# Conner Crowe: full content > More customers from Google and AI like ChatGPT, for ecommerce and service businesses. I run your website, ads and tracking as one system, month to month. This is the full-content companion to https://connercrowe.com/llms.txt. That file is the curated index of the site. This one inlines the text of every answer page, blog post, and case study so the substance is available in a single fetch. Conner Crowe is one person, not a team. The site writes in first-person singular throughout. Static pages such as /pricing/, /services/, /about/ and the frameworks are indexed in /llms.txt with descriptions and are not inlined below. Generated at build time. Content as of 2026-11-16. --- # Benchmarks and tools - AI answers field report: https://connercrowe.com/benchmarks/ai-answers-field-report/ (pooled, dated edition of graded AI-assistant answers about client and snapshot businesses; no per-business data; CSV at https://connercrowe.com/benchmarks/data/ai-answers-field-2026-q4/summary.csv) - What AI says about you: https://connercrowe.com/what-ai-says-about-you/ (free template: ten questions, a check log and a listing-accuracy ledger; PDF at https://connercrowe.com/freebies/what-ai-says-about-you.pdf) - All first-party benchmarks: https://connercrowe.com/benchmarks/ - All free tools: https://connercrowe.com/tools/ --- # Free tools: AI search checks and the AI crawler directory - AI Visibility Check: https://connercrowe.com/tools/ai-visibility-check/ (Free, no signup. Reads a business site the way AI crawlers do (robots.txt rules, a fetch as each AI bot, server-rendered text, business schema, llms.txt), then gives ten buyer questions for the city and service to run in AI assistants and a scorecard of who got named.) - llms.txt Checker: https://connercrowe.com/tools/llms-txt-checker/ (Free, no signup. Checks whether a site serves /llms.txt and whether it follows the published llmstxt.org format, and says plainly what the file does and does not do.) - AI Crawler Checker: https://connercrowe.com/tools/ai-crawler-checker/ (Free, no signup. Reads a site's robots.txt and shows which AI crawlers it allows and which it blocks, bot by bot, with what each bot is for.) - The AI crawler directory: https://connercrowe.com/tools/ai-crawlers/ (12 AI crawlers and fetchers, each with its owner, purpose, user agent and how to allow or block it, cited from the owner's own documentation (last verified October 2026).) ## AI crawler directory: one page per bot - GPTBot: https://connercrowe.com/tools/ai-crawlers/gptbot/ (OpenAI's training crawler. It collects pages that may be used to train OpenAI's foundation models.) - OAI-SearchBot: https://connercrowe.com/tools/ai-crawlers/oai-searchbot/ (The crawler behind ChatGPT search. OpenAI says pages that block it won't be shown in ChatGPT's search answers.) - ChatGPT-User: https://connercrowe.com/tools/ai-crawlers/chatgpt-user/ (ChatGPT visiting a page because a user asked it to. OpenAI says robots.txt rules may not apply.) - ClaudeBot: https://connercrowe.com/tools/ai-crawlers/claudebot/ (Anthropic's training crawler. Blocking it signals Anthropic to leave your future pages out of model training.) - Claude-User: https://connercrowe.com/tools/ai-crawlers/claude-user/ (Claude fetching a page because a user asked a question. Anthropic says it honors robots.txt.) - Claude-SearchBot: https://connercrowe.com/tools/ai-crawlers/claude-searchbot/ (Anthropic's search crawler. Anthropic says blocking it may reduce how visible you are in Claude's search results.) - PerplexityBot: https://connercrowe.com/tools/ai-crawlers/perplexitybot/ (Perplexity's search crawler. Perplexity says it isn't used to collect training data for foundation models.) - Perplexity-User: https://connercrowe.com/tools/ai-crawlers/perplexity-user/ (Perplexity fetching a page to answer a user's question. Perplexity says it generally ignores robots.txt.) - Google-Extended: https://connercrowe.com/tools/ai-crawlers/google-extended/ (A robots.txt control token, not a crawler. It tells Google whether it may use your pages to train Gemini and to ground answers in some of its other systems.) - CCBot: https://connercrowe.com/tools/ai-crawlers/ccbot/ (Common Crawl's crawler. Common Crawl describes the result as an open repository of web crawl data.) - Applebot-Extended: https://connercrowe.com/tools/ai-crawlers/applebot-extended/ (A control token, not a crawler. It decides whether Apple may use what Applebot crawled to train its generative models.) - Meta-ExternalAgent: https://connercrowe.com/tools/ai-crawlers/meta-externalagent/ (Meta's crawler for training foundation models and indexing content for its products.) --- # Answers: Shopify Storefronts Hub: https://connercrowe.com/shopify-storefront/ ## Can I customize a Shopify theme myself or do I need a developer? URL: https://connercrowe.com/shopify-storefront/customize-shopify-theme-yourself-or-developer/ Published: 2026-07-29 | Updated: 2026-10-21 Answer: You can do anything the theme editor exposes: sections, colors, fonts, layout order and most app blocks. Hire a developer once the change needs Liquid, JavaScript or CSS, and above all for tracking. On two of three ecommerce accounts I measured, May to August 2026, 16.6% to 17.2% of counted conversions were cart or checkout events, not purchases. The right answer saves you either a developer invoice or a weekend of breaking your own store. The line between the two is cleaner than most founders think. ### What you can safely do yourself Everything inside the theme editor: reorder sections, add or remove them, change colors and typography from theme settings, swap images, edit navigation, configure app blocks, write collection descriptions. The editor is built so none of this can break the theme. If you make a mess, you reset the settings. Two rules make DIY work safer. **Duplicate the theme before a big session and experiment on the copy.** Shopify lets you preview an unpublished theme end to end, including the cart, so you can walk the whole store before anything is public. This is the same discipline a professional build uses, and it costs you one click. **Change one thing at a time and check it on your phone.** The editor opens on desktop, and plenty of your buyers aren't on one; your analytics will tell you the share. A change that looks composed on a laptop can bury the add-to-cart button on a phone, and you'll never notice from the editor preview. ### Where the developer line sits The moment the change requires the code editor, the calculus flips. Liquid templates, JavaScript, CSS, metafield-driven layouts, structured data, checkout-adjacent anything. Founders can learn Liquid. The reason to hire is that theme code is load-bearing in ways the editor never shows you. The product card you restyle also renders in search results, recommendation blocks, and collection grids. The script you paste in runs on every page and costs load time everywhere. The other developer-territory job is subtraction. Years of app installs leave orphaned code in the theme, and removing it without breaking something requires knowing what still depends on it. Cleanup is quiet, unglamorous work that shows up directly in page speed. ### The jobs that look DIY but aren't Three changes tempt founders because they look small. **Moving the reviews block on the product template.** That's template logic, not a setting. The block you move may be rendered by an app that expects it in a specific place, and the failure mode is a product page that renders fine for you and blank for a buyer on a different variant. **Adding a countdown or announcement widget from a tutorial.** Pasted JavaScript never leaves. It loads on every page, including the ones the promotion ended on six months ago, and nobody remembers where it came from. **"Just hiding" a section with CSS instead of removing it.** The section still loads. You can't see it costing you speed. Each looks like a settings change and each is a code change. There's a fourth, and it's the one that costs real money. Pasting a tracking snippet into the theme because a support article said to. Conversion pixels pasted into theme code can fire on page load rather than on a confirmed purchase. The account side has its own version of this: on two of three ecommerce accounts I measured between May 13 and August 10, 2026, 16.6% to 17.2% of counted conversions were add-to-cart and begin-checkout events, because of which actions the account counted as conversions ([the benchmark](/benchmarks/ecommerce-tracking-accuracy/)). I've rebuilt that mistake often enough to have written the reference architecture down: it lives in [the Tracking Stack](/frameworks/tracking-stack/). ### The rule I use with clients On accounts I run, the founder owns the editor and I own the code. They rearrange the homepage for a promotion without waiting on anyone. Custom cards, template work, speed, and tracking stay on my side, staged on a theme copy and reviewed before going live. That split keeps the store moving daily without the codebase decaying underneath it. You can see the split in the theme library. Across the four Shopify stores I manage, as of October 21, 2026, the theme libraries held 31 staged theme copies against one live theme each. Code work happens on copies, and the live theme only changes when something gets published. That's the split working. Daily control for the person who knows the catalog, code control for the person who knows what the code touches. ### What each side costs | Option | Cost | | --- | --- | | The editor | Free, plus your time | | One developer job | Quoted per job | | A coordinated facelift | From $5,000 | | A custom build | Usually $5,000 to $20,000 | Useful when you're deciding whether to spend a Saturday or a budget line. **The editor: free, plus your time.** Most founders can get further than they expect in an afternoon. **One developer job: quoted per job.** A new section, a template fix, an app wired in cleanly. Get the quote against a written spec. **A coordinated facelift: from $5,000.** Cards, merchandising, reviews, collection copy, and speed designed together and staged as one pass. **A custom build: usually $5,000 to $20,000.** When the change list touches most templates and the catalog structure. Most of my builds launch in two to three weeks from kickoff; a small catalog on a standard structure fits the published two-to-four-week window. Published numbers are on [my pricing page](/pricing/). The trap sits between the second and third bands. Four separate one-off edits from four different people add up fast, leave four people guessing at your priorities, and leave a theme nobody understands afterward. The same money as one coordinated pass buys a designed outcome and one person accountable for it. ### The honest default If your list is colors, sections, and content: do it yourself this afternoon. Duplicate the theme first and check your phone after every change. If your list includes the words cards, template, speed, tracking, or checkout: price a developer. And price the [coordinated version](/storefronts/) before commissioning one-off edits, because five separate patches usually cost more than one planned pass and leave you with a worse theme. The full scope-by-scope pricing is at [what theme customization costs](/shopify-storefront/shopify-theme-customization-cost/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Does my Shopify store need a facelift or a full rebuild? URL: https://connercrowe.com/shopify-storefront/shopify-facelift-vs-full-rebuild/ Published: 2026-07-28 | Updated: 2026-10-21 Answer: Default to a facelift: if checkout converts, pages rank and tracking works, upgrade the theme you have. Rebuild only when the theme blocks the catalog or the buying flow. My storefront work is usually $5,000 to $20,000, with a facelift from $5,000, and most of my builds launch in two to three weeks from kickoff. The redesign industry has one answer to every dated storefront: start over. That puts your checkout, your tracking, and your SEO equity at risk at the same time. Most stores never needed it. Here's the test I run before quoting either project. ### The five signals **Does the checkout convert?** Pull your last ninety days. If conversion rate sits in a healthy band for your category and price point, the machine works and a rebuild risks it for cosmetic gain. **Do the pages rank?** If collections and products hold organic positions, they carry equity a rebuild can burn through URL changes, lost internal links, and structured data that leaves with the old theme. **Does the tracking work?** A working pixel-and-server-side stack is wired into your current templates. Rebuilding means re-wiring and re-validating all of it, and this step is the one designers skip. Priced on its own, that rebuild is typically a $2,500 project, so a rebuild quote that leaves tracking out is $2,500 light before anyone argues about design. **Can the theme show what you sell?** This is the first real rebuild trigger. If the catalog has outgrown the templates, variant-heavy products squeezed into cards that can't hold them, navigation with no room for how buyers shop, then the frame genuinely blocks the catalog. **Can the theme be edited safely?** The second trigger. Legacy architecture, years of stacked app code, changes that break something every time someone touches it. When the marginal edit costs more than it should, the codebase is telling you its age. If signals four and five hold, take the facelift and fix any of the first three on the theme you have. If either four or five fails, price the rebuild. ### The two projects, side by side | | Facelift | Full rebuild | |---|---|---| | Typical range | From $5,000 | Usually $5,000 to $20,000 | | Timeline | Shorter than a rebuild, quoted with the scope | Most launch in two to three weeks; a small catalog fits the published two to four | | URLs | Unchanged | Mapped and redirected | | Structured data | Stays put | Rebuilt and re-tested | | Tracking | Already live, verified | Wired into new templates before launch | | Store downtime | None | None, if staged properly | | Large catalog or migration | Not applicable | Up to $20,000, bigger quoted after the audit | The rebuild column comes from [my own published pricing](/pricing/), where one-time projects run $2,500 to $20,000. ### What a facelift includes [A facelift keeps the theme](/blog/shopify-storefront-facelift-without-a-rebuild/) and upgrades what buyers see: new product cards, a homepage rearranged around how people shop, reviews moved onto product and collection pages, collection copy and internal linking, page-speed work. Everything stages on a theme copy, gets reviewed, and goes live without the store going down. I've run this test on a store with a live Performance Max program: the theme was capable and badly configured, so it got a facelift staged on a theme copy while the campaigns kept running. That staging is routine, not special handling. Across the four Shopify stores I manage, as of October 21, 2026, the theme libraries held 31 staged theme copies against one live theme each. ### What a rebuild includes, when it's earned When the theme genuinely blocks the catalog, the rebuild covers a custom theme in code you own, catalog and collection architecture, page templates built for paid traffic, imagery where the catalog has gaps, and conversion tracking wired before launch. Signal four fails hard when the catalog needs navigation the theme has no room for, or product imagery that doesn't exist yet. No amount of merchandising on the old frame produces either. [Emma Grace Home](/results/150-sku-furniture-catalog-no-photographer/) is the clean example: a furniture catalog of 150-plus products that needed shop-by-room navigation and had no product photography, so it needed a full build rather than a facelift. It launched in under a month with every product image rendered in-house, on a proven theme base with the navigation and sections rewritten in code. Large catalogs and migrations take longer than the usual two to three weeks, so I quote them after the audit. ### Why the industry defaults to rebuilds anyway A rebuild is a bigger invoice and a cleaner project plan. A facelift requires reading someone else's theme and shipping on a live store without breaking it, which is harder to scope and easier to get wrong. Vendors quote the project that suits the vendor. There's a second reason, and it's worth naming. A rebuild lets a builder ignore everything you already have. No inventory of what ranks, no audit of which URLs your campaigns point at, no reading of the app code. Starting over is the cheapest option for the vendor and the most expensive one for you, because you pay to rebuild the parts that were already working. The five signals exist so you can price what the store needs instead. Both projects, with published pricing, live on [my Shopify storefronts page](/storefronts/). On a scoping call I'll tell you which one you need, and the facelift answer comes back more often than my invoice would prefer. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## How do I redesign my Shopify store without losing sales? URL: https://connercrowe.com/shopify-storefront/redesign-shopify-store-without-losing-sales/ Published: 2026-07-29 | Updated: 2026-10-21 Answer: Build on a duplicate theme, verify tracking, redirects and phone speed there, then publish in one action and keep the old theme as a one-click rollback. Across the four Shopify stores I manage, as of October 21, 2026, the theme libraries held 31 staged copies against one live theme each. The store never goes down. "Redesign" and "keep selling" sound like a conflict. On Shopify they aren't, because the platform separates the theme from the store. The catalog, the checkout, the orders, and the customer accounts all live outside the theme. A redesign done properly swaps the frame while the machine keeps running. ### The staging rule Every change happens on a duplicate of the live theme, never on the published one. Shopify allows multiple themes in the library. The duplicate gets the full build while customers shop the original, and the preview link lets you walk the whole store, cart to checkout, before anything is public. This rule alone removes most of the risk founders fear. There's no construction-site phase, no maintenance mode, no weekend the store goes dark. It's also how the stores I run actually look. Across the four Shopify stores I manage, as of October 21, 2026, the theme libraries held 31 staged theme copies against one live theme each. Work happens on copies, and the live theme only changes when something gets published. The corollary: any vendor who asks to edit your live theme directly, or who wants the store password-locked "during the redesign," is telling you how they work. Keep looking. ### What runs in parallel While the new theme builds, three workstreams run alongside it rather than after it. **Tracking.** GA4, GTM, and the platform pixels get wired into the new templates on the staging copy and test-fired there. The single most common redesign injury is a beautiful new site with a dead purchase event, discovered two weeks later as "the ads stopped working." None of that damage is visible in a design review. **Redirects and SEO parity.** Every URL that changes gets a mapped redirect, and the pages that rank get their content and structured data carried over. Two Shopify rules shape this. A URL redirect only fires when the old path returns a 404, so if the old URL still loads a page, the redirect is ignored. And Shopify offers to create the redirect when you change a handle; leave that option on, and check it. Test redirects with an ad's query string attached, too: on one ad account I audited, redirects returning a clean 301 had been stripping the gclid from 860 of 3,519 paid clicks over thirty days ([the write-up](/blog/a-200-is-not-proof-redirects-that-eat-the-gclid/)). The full checklist is in [will a redesign hurt my SEO](/shopify-storefront/will-a-shopify-redesign-hurt-seo/). **Speed.** Measured on the staging theme, on a phone, before cutover. My method is to compare two unpublished themes against each other: a fresh duplicate of the live theme and the new build. Shopify's preview bar loads on any theme you preview, and it skews Lighthouse, so testing a preview against the live store tilts the result. A redesign that ships slower than the store it replaces starts underwater no matter how it looks. ### The cutover checklist Cutover is one action: publish the new theme. What surrounds it matters more than the click. Before you publish, confirm all of it on the staging theme. Every changed URL has a redirect. The purchase event fires and reaches the ad platforms. Structured data coverage on the key templates matches the old store. Page speed measured on a phone, on the pages your ads land on. Collection copy and product copy carried over, or retired on purpose. Publish at a low-traffic hour. Then click through the money paths immediately: product page to cart to checkout to a test purchase, watching the tracking fire live rather than trusting that it will. Keep the old theme in the library untouched. That's your rollback. If anything is wrong, republishing it takes one click and the store is exactly what it was. Don't delete it for a month. The cart lives on the store side, tied to a browser cookie, not in the theme, so a plain publish shouldn't empty anyone's cart. I still test it rather than assume it: add a product on the old theme, publish, reload, and check the cart. A new theme with custom cart code is the case where that check earns its minute. ### The worked example: a facelift under a live Performance Max program It's an ecommerce store I manage. Before I touched the storefront, its Google Ads account had been rebuilt and grown into several Performance Max campaigns. The campaigns sent traffic to the store's existing pages, and server-side tracking ran through its existing checkout. Replatforming would have thrown away conversion history the campaigns were bidding on, and that cost never shows up on a redesign quote. So nothing was replatformed. The work shipped as custom CSS and JavaScript in the theme layer, built and reviewed on a theme copy, and the live theme stayed as it was until the publish. What shipped: product cards that show color options right in the grid, a homepage organized around the brands and categories people actually browse, and the store's 195-plus customer reviews moved onto the pages where buyers decide. The store never went down, and the Performance Max campaigns kept running through the whole project. ### The cutover gate, from a site migration I ran This one wasn't Shopify, but the gate carries over. On [a site migration I ran](/results/hand-wrist-institute-site-and-ads-cutover/), I inventoried all 533 legacy URLs before building anything, 348 of them blog posts. The site was rebuilt and its ad account moved onto it in the same week. After cutover, 40 of 40 sampled redirects were correct, and the way back stayed one record change away the whole window. On Shopify the old theme plays that role. Know your rollback is one publish before you need it, and don't call the cutover done until the redirect sample and a test purchase both pass. ### What to do if the vendor won't stage Two options, and both are fine. Stage it yourself. Duplicate the theme in your admin, give the vendor the duplicate, and hold the publish button yourself. A builder who resists that arrangement is telling you they intend to work on your live store. Or take the smaller project. If your checkout converts and your pages rank, a facelift on the existing theme carries almost none of this risk, because URLs, schema, and content all stay put. The five-signal test for which project you need is at [facelift or full rebuild](/shopify-storefront/shopify-facelift-vs-full-rebuild/). ### What founders underestimate The staging discipline is the safety net. A rollback undoes a bad theme; it doesn't undo a week of ads optimizing against a broken conversion event, or a month of rankings decaying behind missing redirects. Those are the injuries that survive the rollback, and they're the reason tracking and redirects belong in the build rather than the cleanup. The same staged process is how every one of [my Shopify storefront builds](/storefronts/) ships, and the timeline it fits into is at [how long a rebuild takes](/shopify-storefront/how-long-does-a-shopify-store-rebuild-take/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## How long does a Shopify store rebuild take? URL: https://connercrowe.com/shopify-storefront/how-long-does-a-shopify-store-rebuild-take/ Published: 2026-07-28 | Updated: 2026-10-21 Answer: Most of my Shopify builds launch 2 to 3 weeks after kickoff; my published window for a small catalog is 2 to 4 weeks. Large catalogs and migrations run longer and are quoted after the audit. Every window includes tracking, redirects and a phone speed check. Timeline quotes for the same store range from two weeks to six months, which tells you the quote reflects the vendor's process more than the work. Here are the real bands and what moves a project inside them. ### The honest bands **A landing page: about a week.** One template, one purpose, tracking included. If a vendor quotes a month for a landing page, you're paying for their queue, not the work. **A facelift: quoted with the scope, usually shorter than a build.** Custom cards, homepage merchandising, reviews surfaced, speed work. It ships fast because nothing replatforms: every change stages on a copy of the live theme, and the store keeps selling through the whole project. **A small-catalog custom rebuild: two to four weeks.** Theme, catalog architecture, page templates, imagery, tracking. That's the window I publish for a small catalog on [my Shopify storefronts page](/storefronts/). Most of my builds launch in two to three weeks, and four is the honest ceiling for a small catalog with imagery gaps. The dated example is [Emma Grace Home](/results/150-sku-furniture-catalog-no-photographer/): a custom build of 150-plus products, imagery included, that opened in April 2026 less than a month after kickoff. **A large catalog or a platform migration: longer, and priced differently.** Thousands of SKUs with real specification data, a redirect map protecting existing rankings, or part-number and fitment search. That work runs up to $20,000, with bigger migrations quoted after the audit, and it doesn't compress into a month. Any vendor who says it does hasn't done one. **The agency version: longer, for structural reasons.** An agency timeline carries coordination: a project manager, a designer, a developer, and QA passing the work between them, with meetings at each handoff. | Project | Typical window | What sets it | |---|---|---| | Landing page | About a week | One template, tracking included | | Facelift | Quoted with the scope, usually shorter than a build | Staged on a copy of the live theme | | Custom build, most stores | 2 to 3 weeks | One person holds theme, catalog, imagery and tracking | | Custom build, small catalog with imagery gaps | Up to 4 weeks | Imagery production | | Large catalog or platform migration, Plus included | Quoted after the audit | SKU count, redirect map, apps and integrations | ### Where the calendar goes On a multi-person build, most of the calendar is waiting. Imagery is a scheduling problem before it's a production problem: product ships out to be shot, comes back, and the edited files have to arrive before anyone can build the collection pages that display them. Design waits on imagery. Development waits on design. Coordination waits on everyone and reports on the waiting. Very little of that timeline is work. Most of it's queue. When one person holds the theme, the catalog structure, the imagery, and the tracking, those workstreams run in parallel instead of in sequence. That's where the two to three weeks comes from. The work stops waiting on handoffs. ### A cutover week, start to finish On a site migration I ran, I rebuilt the site and moved its ad account onto it in the same week. It wasn't a Shopify store, but the shape of the calendar is the same one I use on Shopify. Before anything was built, I inventoried 533 legacy URLs, 348 of them blog posts. After cutover I sampled 40 redirect rules against the live domain, and 40 of 40 were correct. Rollback stayed one record change away the whole time. The [site and ads cutover case study](/results/hand-wrist-institute-site-and-ads-cutover/) has the full record. The inventory is why the cutover fit in a week. The URLs were mapped before the build started, so launch week was checking, not discovering. ### What the timeline should include A build isn't done when it looks done. The closing stretch of a real timeline holds the work that invisible-until-launch projects skip: conversion tracking wired into the new templates and validated end to end, redirects mapped, page speed measured on a phone over cellular, and a staged launch where the old store keeps selling until the moment of cutover. A quote that beats these windows by skipping those is quoting a different, worse project. The tracking rebuild alone is a $2,500 engagement when bought on its own, which is a useful way to price the omission. ### What stretches a timeline Four things, in order of how often I see them. **Catalog size.** 700 products is a different project than 70. Collection architecture, filtering, and variant handling all scale with the count. **Imagery gaps.** If half the SKUs have one bare supplier photo, producing catalog imagery becomes its own workstream. It's why I run an in-house render pipeline, so imagery doesn't sit in a queue waiting on studio bookings. **Approval lag.** The single biggest variable nobody quotes. A founder who reviews in a day keeps a build on pace. A committee that reviews in two weeks can double the calendar by itself, and no vendor can absorb that from their side. **App surgery.** Subscriptions, bundles, B2B pricing, ERP feeds. Each integration adds build and testing time, and testing is the half that gets underestimated. On Shopify Plus, the work that stretches a calendar is checkout apps, B2B pricing and ERP feeds, so a Plus rebuild gets quoted after the audit, the same as a migration. ### Why my timelines read short The build runs through a system: a written playbook for every repeatable step, templates and checks accumulated across every build before yours, and one senior operator instead of four people passing work between them. Fewer handoffs is most of the answer. The same person doing the theme also wires the tracking, so the two workstreams never wait on each other. The rest is scope discipline: a build with a defined end date has a defined scope, and I'd rather say no to a feature in week one than discover it in week five. ### The question to ask any vendor Ask what happens in week one. A builder with a working system starts building in week one. A process-heavy shop spends weeks one through three on discovery decks. Both can produce a good store. Only one of them does it this quarter. Then ask what happens in the last week, because that's where tracking, redirects, and the speed check either live or don't. Timelines and pricing for my builds are published on [my custom Shopify storefront page](/storefronts/), and the risk checklist for the cutover itself is at [redesign without losing sales](/shopify-storefront/redesign-shopify-store-without-losing-sales/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## How much does a custom Shopify theme cost? URL: https://connercrowe.com/shopify-storefront/how-much-does-a-custom-shopify-theme-cost/ Published: 2026-07-28 | Updated: 2026-10-21 Answer: My custom Shopify builds run $5,000 to $20,000, with conversion tracking included. A small catalog on a standard structure starts at $5,000 and takes 2 to 4 weeks. Large catalogs, migrations and part-number or fitment search run up to $20,000, and bigger migrations are quoted after the audit. Paid Theme Store themes cost $100 to $500. The number depends on which of four things you're buying, and most quotes never say which one. ### The price bands | Option | Price (custom rows are my pricing) | What you get | |---|---|---| | Premium Theme Store theme | $100 to $500 | A template thousands of other stores run, plus the hours to configure it | | Custom build, small catalog | From $5,000 | One person designs on a proven theme base and rewrites what the catalog needs in code | | Large catalog or migration | Up to $20,000 | The catalog is the project, and bigger migrations are quoted after the audit | **$100 to $500: a premium theme from the Shopify Theme Store.** That's the paid range listed on the [Theme Store](https://themes.shopify.com/themes?price%5B%5D=paid) as of October 21, 2026. You're buying a template thousands of other stores run, plus the hours to configure it. For a young store with a small catalog and no paid traffic, this is often the right answer, and anyone who quotes you five figures before asking about your catalog is selling scope you may not need. **From $5,000: a custom build from a senior independent operator.** One person designs the storefront for your brand on a proven theme base and rewrites what your catalog needs in code: catalog architecture, product page templates, page speed, and the storefront details a template can't carry. My builds start at $5,000 for a small catalog on a standard structure and run up to $20,000, depending on catalog size and how much imagery the catalog needs. The published number, with the timeline attached, is on [my pricing page](/pricing/). **Up to $20,000: a large catalog or a migration.** The same one person, but the catalog is the project rather than a detail of it. Thousands of SKUs carrying real specification data, a move off an aging platform, a redirect map that has to protect existing rankings, or a store whose customers search by part number or vehicle fitment. That work doesn't compress into a month, and pricing it like a boutique build would be dishonest. Migrations bigger than that are quoted after the audit. **An agency build: a team on the invoice.** An agency price carries a team: project manager, designer, developer, QA, and the meetings that keep them synchronized. On enormous catalogs and multi-region Plus stores, that team earns its cost. On a founder-led store doing $1M to $30M, most of it's overhead. ### Catalog size moves the number Numbers get slippery in the abstract, so here's the input that moves them most. Across the four Shopify stores I manage, as of October 21, 2026, catalogs range from 3 to 5,116 products. Those aren't the same project at any price. A three-product store needs a sharp product page and a clean path to checkout. A five-thousand-product store needs collection architecture, filtering, variant handling, and search that hold up under that weight, and that's most of the build. Imagery moves it next. [Emma Grace Home](https://emmagracehome.com/), a premium furniture store I launched solo, had 150-plus products and almost no usable photography. It launched within a month because I rendered the imagery myself instead of waiting on a studio calendar, and monthly sales went from $5,387 in April 2026 to $56,182 in September. [The full build](/results/150-sku-furniture-catalog-no-photographer/) is written up as a case study. A 40-SKU store with existing photography and a standard collection structure sits at the floor. The catalog moves the number. ### What moves the price inside a band Four inputs move a custom quote more than anything else. **Catalog size and structure.** A 30-product store and a 700-product store are different projects. Collection architecture, filtering, and variant handling all scale with the catalog. **Imagery.** If half the catalog ships with bare supplier photos, the storefront will look dated in any theme. Rendering or shooting lifestyle imagery is real scope. I run an in-house render pipeline for exactly this gap, and it's part of the quote when the catalog needs it. Priced on its own, that imagery work is a one-time project quoted by SKU count. A first batch of ten lifestyle renders plus one Meta ad set is typically $2,500. **Integrations.** Reviews, subscriptions, bundles, B2B pricing, ERP feeds. Each app the theme has to accommodate adds build and testing time. **Tracking.** GA4, GTM, a server-side container, and platform pixels wired into the new templates before launch. Most quotes leave this out entirely, which is how new sites go live and the ad account goes dark. Bought separately, that rebuild is typically a $2,500 project. ### What the quote should include A complete custom theme quote covers the theme code you own, catalog and collection structure, page templates designed for paid traffic to land on, conversion tracking wired before launch, page-speed work, and documentation the next developer can read. If a quote is missing tracking and speed, it isn't comparable to one that includes them, no matter what the two numbers are. Add $2,500 for the tracking rebuild you'll need afterward and compare the totals. A $6,000 quote with tracking beats an $8,000 quote without it, and it beats a $4,000 quote that leaves your ad account blind for the first month after launch. ### When not to buy custom Skip custom work if your store is pre-revenue, your catalog is under a couple dozen products, or your current theme converts fine and the real complaint is that you're bored of it. A $300 theme plus a merchandising pass beats a $10,000 build you haven't grown into. There's a middle option most founders never get quoted: a facelift on the theme you already run. Better product cards, a homepage that follows how buyers shop, reviews where buyers decide, speed work, all staged on a theme copy. That starts at $5,000 and leaves your URLs, your schema, and your checkout untouched. The test for which one you need is in [facelift or full rebuild](/shopify-storefront/shopify-facelift-vs-full-rebuild/). I sell custom themes and I turn down builds on this basis on scoping calls. A founder who buys a template this year and comes back when the catalog earns a custom build is a better client than one who overbuilds now and resents the invoice. ### The receipt behind my numbers The band I quote comes from shipped work, not a rate card. [The Emma Grace Home catalog build](/results/150-sku-furniture-catalog-no-photographer/) launched within a month, imagery included, with no one else on it. A facelift on a live store shipped new product cards, a homepage reorganized by brand and category, and 195-plus reviews on the storefront, with no downtime. The full offer, with published pricing and the timeline on the card, is on [my custom Shopify storefront page](/storefronts/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## How much does it cost to customize an existing Shopify theme? URL: https://connercrowe.com/shopify-storefront/shopify-theme-customization-cost/ Published: 2026-07-29 | Updated: 2026-10-21 Answer: Changes inside the theme editor cost only your time. Single sections from a freelancer are priced by the hour. Storefront work on my side, from a facelift to a full build, starts at $5,000 and usually runs to $20,000 with tracking included. Customization quotes confuse founders because "customize my theme" covers everything from changing a font to rebuilding the product page. The price tracks the scope, so here are the scopes. ### Free: the theme editor Colors, fonts, section order, logo, announcement bars, most layout toggles. Modern themes expose a lot in the editor, and a founder with an afternoon can get further than most expect. If your complaint is that the homepage sections are in the wrong order, or that the colors are off brand, you don't have a customization project. You have a settings session. Two habits make that session safe. Duplicate the theme before you start and work on the copy, which Shopify lets you preview end to end before publishing. And check every change on your phone, because the editor opens on desktop and plenty of your buyers aren't; your analytics will tell you the share. ### Hourly: section and template work A developer building one new section, modifying a product template, or wiring an app into the theme cleanly. This work is usually billed by the hour, so the quote is the developer's rate times the hours, and a single-purpose job is a small number of hours. Ask for the hour estimate in writing before the work starts. This band is the right buy when the theme is fundamentally fine and one specific thing is missing: a comparison table, a shade-finder block, a custom collection header, a size chart that pulls from metafields instead of a pasted image. The failure mode in this band is accumulation. Five separate freelancers making five separate edits over two years leaves a theme nobody understands, carrying dead code from apps you uninstalled in year one. Each edit was reasonable. The sum is a codebase where the marginal change costs more than it should, because whoever touches it next has to read four other people's work first. The arithmetic is worth doing out loud. Say four one-off edits at $1,250 each is $5,000, plus four separate discovery conversations, four separate people guessing at your priorities, and no shared plan. The same $5,000 spent as one coordinated pass buys a designed outcome and one person accountable for it. If you're about to commission the fourth patch, stop and price the coordinated version instead. ### From $5,000: the coordinated facelift Product cards redesigned for the catalog, a homepage reorganized around how buyers browse, reviews brought onto the storefront, collection copy, internal linking, page-speed cleanup. All designed together, all staged on a theme copy before anything touches the live store. This is the band where the storefront visibly changes caliber without a replatform. Staging is a habit, not a line item. Across the four Shopify stores I manage, as of October 21, 2026, the theme libraries held 31 staged theme copies against one live theme each. Every change gets built and checked on a copy before it publishes. ### A worked example: a premium catalog on a stock theme The store sells premium brands nationally online, and a scaled Performance Max program was already sending it traffic. The ads had improved faster than the site: high-ticket products on a theme nobody had configured past the defaults. The usual answer is a redesign, which means a new theme, a long project, and a cutover where checkout, tracking, and rankings all change at once. What shipped instead, on the existing theme: custom product cards that show variant options right on the collection grid, a homepage organized around the brands and categories people actually shop, and the 195-plus customer reviews brought onto the storefront where buyers make the decision. Every change was built and reviewed on a theme copy, and the live theme stayed as it was until publish. Nothing was replatformed, the theme wasn't rebuilt, and the store never went down. The campaigns kept running through the swap with nothing new to relearn, which was the point, because a replatform would have handed them new pages and new URLs to learn from scratch. That's the shape of a facelift: the same store, presenting what it sells properly. ### Past the facelift: you're pricing a build, not a customization When the change list touches most templates, the catalog structure, and the buying flow, piecemeal customization stops making sense and a custom theme starts. My full storefront builds start at $5,000. Most launch in two to three weeks, and my published window for a small catalog is two to four. Large catalogs and platform migrations run up to $20,000, and bigger migrations are quoted after the audit. That decision has its own math, covered in [what a custom Shopify theme costs](/shopify-storefront/how-much-does-a-custom-shopify-theme-cost/). ### What makes any of these quotes worth it Three questions filter vendors regardless of band. **Does the work get staged on a theme copy, or edited live on your store?** There's no reason to edit a live theme in 2026. A vendor who does is telling you how they handle risk everywhere else. **Does the quote include measuring page speed after the change?** Every added section is load time somewhere. A $600 section that adds two seconds to your best landing page isn't a $600 section, it's a $600 section plus a tax on every click you buy. **Does the person doing the work know which pages your paid traffic lands on?** Customization is merchandising. Someone who has never watched a campaign land on a collection page will optimize for the design review instead of the buyer. The coordinated version of this work, priced, with the case studies attached, lives on [my Shopify storefronts page](/storefronts/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Should I hire a Shopify agency or a freelancer for my store rebuild? URL: https://connercrowe.com/shopify-storefront/shopify-agency-vs-freelancer-for-store-rebuild/ Published: 2026-07-28 | Updated: 2026-10-21 Answer: For a founder-led store, hire a senior independent operator; hire an agency for enterprise or multi-region builds. The deciding questions are who writes the code and who owns the result. My storefront builds are usually $5,000 to $20,000, and most launch in two to three weeks, with tracking wired before launch. The agency-or-freelancer question is framed wrong, and the framing is what gets founders burned. Both labels cover an enormous quality range. The variables that decide your outcome sit underneath the label. ### The two questions that decide it **Who touches the code?** At an agency, the person who sold you the build is rarely the person building. The work lands with a mid-level developer, reviewed at milestones. With a freelancer, the person you interview is the person building, which is either great or terrible depending entirely on who they are. Seniority of the actual hands predicts your outcome better than company size does. **Who owns the result after launch?** A rebuild gets judged on what the store converts, not on how it looks at handoff. Most builders, agency or freelance, hand off and leave. If the builder also runs your traffic, they inherit every decision they made in the build, which changes how they make them. This is the reason I build storefronts as part of a marketing program instead of as standalone web design. ### The three shapes, with the numbers attached | | Agency | Marketplace freelancer | Senior independent operator | |---|---|---|---| | Typical rebuild quote | Priced for a full team | Per job or per hour, against your spec | From $5,000, most $5,000 to $20,000 | | Timeline for the same scope | Set by their staffing | Depends on their queue | Most launch in two to three weeks | | Who writes the code | Mid-level developer | The person you interviewed | The person you interviewed | | Tracking included | Usually a separate line | Usually not | Wired before launch | | Bench if someone leaves | Yes | No | No | | Large catalog or migration | Core strength | Rarely | Up to $20,000, bigger quoted after the audit | My column matches [the pricing I publish](/pricing/). ### Where an agency is the right call An honest read, because the agency shape wins real scenarios: you run a Plus store with multi-region requirements, ERP integrations, and a security review process. You need bench depth so the project survives one person leaving. You need five workstreams staffed at once and your team is used to working through a project manager. At that scale, the agency quote carries real coordination work, and a solo builder is the risk, not the bargain. ### Where a freelancer is the right call Small scope, tight budget, clear spec. A theme customization, a section build, a template fix. Marketplace freelancers do this well, and hiring an agency for it wastes money on process. The failure mode is handing a freelancer an open-ended rebuild with no spec, then discovering the gaps at launch. Freelancers price against the spec you wrote. If the spec omits redirects, structured data, and conversion tracking, so will the build, and none of those omissions are visible in a design review. ### The third option the question hides A senior independent operator: one person with agency-level experience who scopes, designs, builds, and stays on the account. You get the seniority an agency puts on the sales call, on the keyboard, without the junior layer the retainer pays for. My storefront builds start at $5,000, and most land between $5,000 and $20,000. Most launch in two to three weeks from kickoff, and every build ships with conversion tracking wired in before launch, because I'm the same person running the paid traffic afterward. Large catalogs and platform migrations run up to $20,000, with bigger migrations quoted after the audit, where the catalog itself is the project rather than a detail of it. ### Why one operator moves faster Handoffs eat more of a build calendar than effort does. When a designer, a developer, a photographer and a project manager pass work between them, much of the schedule goes to waiting and re-explaining. One person doing the theme, the imagery and the tracking never waits on themselves, which is most of why my builds usually launch in two to three weeks from kickoff. Large catalogs and migrations take longer, and I quote them after the audit. ### The trade, stated plainly There's no bench. If I'm out for a week, the project waits a week. For a founder-led store doing $1M to $30M, that trade usually wins. For a store that needs several teams staffed at once, it starts to lose, and I'll say so on the call rather than take the work. ### What to ask any builder before signing Ask who personally writes the code and how senior they are. Ask to see two live stores they built and what changed in conversion after launch. Ask whether the quote includes conversion tracking in the new templates, and if the answer is a blank look, walk. Ask what happens in month two when something breaks. Ask who holds the redirect map. I'll answer the live-store question first. [Emma Grace Home](https://emmagracehome.com/) is a store I launched solo that opened in April 2026, and its monthly sales went from $5,387 that month to $56,182 in September, from the store's Shopify sales report. [How I built it](/results/150-sku-furniture-catalog-no-photographer/) is written up in full. The answers separate real builders from proposal factories faster than any portfolio does. The way I answer those questions is documented on [my Shopify storefronts page](/storefronts/), and the honest solo-versus-agency comparison, including the rows the agency wins, is on [my solo consultant vs agency page](/vs/agency/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Should I redesign my Shopify store or buy a new theme? URL: https://connercrowe.com/shopify-storefront/shopify-redesign-vs-new-theme/ Published: 2026-07-28 | Updated: 2026-10-21 Answer: Buy a theme when yours is legacy architecture and a modern template covers your catalog: Shopify's Horizon is free, and paid Theme Store themes cost $100 to $500. Commission custom work when the catalog has outgrown templates; I build those on a proven theme base, from $5,000. On most stores, merchandising the theme you already run comes first. "The store looks dated" has three fixes: a new theme, free or $100 to $500, a merchandising pass on the theme you have, or custom work. The last two are storefront work from $5,000. Which one you need depends on why it looks dated, and most founders skip that diagnosis. ### When a new theme is the answer A theme swap earns its cost in two situations. First, your current theme is legacy architecture. If it predates Shopify's Online Store 2.0, you're locked out of sections on every page, flexible metafields, and most modern apps. A current theme removes real constraints beyond the visual ones. This is the clearest buy on the page: $100 to $500 removes a ceiling you've been working under for years. Second, your catalog fits what good templates do well. A focused catalog, standard product types, no unusual buying flow. A well-configured theme covers that store, and the low price is the whole point. Shopify's own current free theme is Horizon, launched in May 2025 as a family of ten free themes, and the paid themes listed on the [Theme Store](https://themes.shopify.com/themes?price%5B%5D=paid) on October 21, 2026 ran $100 to $500. Configure it properly, spend the savings on imagery, and move on. Budget the configuration honestly. A theme is a purchase; making it look like your brand is a project. Two to three days of your own time in the editor, or paid freelance help against a clear spec. A $300 theme that nobody configures reads exactly as generic as the one it replaced. ### When a new theme won't fix it A theme changes the frame around your content. It doesn't change the content. If the product photos are bare supplier shots, the collections are thin, and the product descriptions are two manufacturer sentences, the new theme will present the same weaknesses in nicer typography. Founders swap themes, see no lift, and conclude design doesn't matter. The design was never the problem. The catalog was. This is the most common $500 mistake in ecommerce, and it's expensive twice: once for the theme, and once for the year that passes before anyone diagnoses the real constraint. A redesign, meaning custom work, earns its cost when the catalog or the buying flow has outgrown templates: room-based navigation for a large furniture catalog (the Living Room, Bedroom, Dining Room and Bathroom menus I built for [Emma Grace Home](https://emmagracehome.com/)), variant-heavy products that need custom cards, a brand identity strong enough that template constraints visibly cheapen it, or landing page templates built for the paid traffic you send. I start from a proven theme base and rewrite the templates, sections and navigation your catalog needs in code, so the store fits the catalog instead of fighting a template's defaults. Those builds start at $5,000, and most launch in two to three weeks from kickoff; a small catalog on a standard structure fits the published two-to-four-week window. Large catalogs and platform migrations run up to $20,000, with bigger migrations quoted after the audit. ### The option most stores should pick first Before either, price the merchandising pass: upgrade the storefront you have. Sharper product cards, a homepage rebuilt around the way people shop, reviews moved onto product and collection pages, collection copy, internal linking, page-speed work. It attacks the content, which is usually the real reason the store reads as dated. That's storefront work from $5,000, staged on a copy of the theme you already run. ### The pattern I see most The common case is a premium catalog on a capable theme nobody configured past the defaults, where the theme is current and perfectly capable. Every instinct says buy a better theme. The diagnosis usually says merchandising. A new theme would cost less in dollars and deliver less, because it moves the same weak content into a nicer frame. Catalog size matters more than looks here. Across the four Shopify stores I manage, as of October 21, 2026, catalogs range from 3 to 5,116 products. A template that suits one end of that range rarely suits the other, which is why the catalog, not the look, decides this. ### The decision in one pass Run the three questions in order. **Is the theme legacy architecture that blocks modern features?** If yes, a new theme is table stakes. Buy it, configure it properly, then reassess. **Does a template cover your catalog and buying flow?** If yes, buy the template and put the difference into imagery and merchandising. Imagery is almost always the higher-return half of that split. **Has the catalog outgrown templates, or does paid traffic need pages templates can't provide?** Then custom work is justified, and the numbers for that are in [what a custom theme costs](/shopify-storefront/how-much-does-a-custom-shopify-theme-cost/). If you answered no, yes, no, you have a merchandising project and a $300 theme, and the five-signal version of this test is at [facelift or full rebuild](/shopify-storefront/shopify-facelift-vs-full-rebuild/). ### What I'd tell you on a call I sell custom builds, so weigh the incentive, but the scoping call runs the same three questions and the answer is frequently the cheap one. A founder who buys a $300 theme this year and comes back for custom work when the catalog earns it makes a better client than one who overbuilds now and resents the invoice. The offer, when the answer is custom, is on [my custom Shopify storefront page](/storefronts/), with the published numbers on [my pricing page](/pricing/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Should my Shopify ads land on product pages or a landing page? URL: https://connercrowe.com/shopify-storefront/shopify-ads-product-page-or-landing-page/ Published: 2026-07-29 | Updated: 2026-10-21 Answer: Land each ad on the page that continues its promise: a product ad on its product page, a category ad on a merchandised collection, an argument on a landing page built for it. In one service account I run, ad groups converted at 11.76% on the best destination and 0.88% on old blog posts. The question behind the question is whether you need to build anything. Often you don't. The ad's promise picks the destination, and only one of the four cases requires new pages. ### Case one: the ad sells one product Land on the product page. A buyer who clicked a Shopping ad or a single-product creative wants the price, the photos, and the buy button, not a detour. The work is making the product page worthy of the click: real photography or renders, variant clarity, reviews visible, speed. If your product pages can't carry paid traffic, that's a storefront problem. The tell is a campaign with healthy click-through and a conversion rate nobody can explain. Before rebuilding the campaign, open the product page on a phone and read it as a stranger who has never heard of you. ### Case two: the ad sells a category Land on a merchandised collection. Someone who clicked "solid oak dining tables" wants to browse the range, so the collection page needs honest filtering, consistent tiles, and enough copy to hold the ranking the ads are paying to reinforce. The most common failure here is a collection page that's a bare grid: no copy, no ordering logic, no reason to trust the brand over the next tab. A collection is merchandising, and merchandising is the highest-return storefront work on most stores because it improves pages the ads already pay for instead of building new ones. ### Case three: the ad makes an argument This is where a dedicated landing page earns its build cost. Bundles and offers the catalog doesn't express. Comparisons against a category the buyer currently uses. Education-first products where the buyer needs to understand the mechanism before the price makes sense. Lead-gen for services attached to the store. A product page can't carry those arguments because its job is to sell one SKU to someone already convinced of the category. A useful test: if the ad's headline is a claim rather than a product name, the destination needs to continue the claim. Sending argument traffic to a product page is why "the ads don't work" so often means "the destination dropped the thread." If I'm already running your ads, landing pages and site upkeep are a $1,500-a-month add-on to the monthly program. A full storefront build is a separate project, usually $5,000 to $20,000. Published numbers are on [my pricing page](/pricing/). ### Case four: the homepage Never on purpose. The homepage is a router for people who already know the brand. Paid traffic pays per click, and a router click is a wasted one. The exception that proves it: pure brand-name search ads, where the visitor typed your name and wants the front door. Service sites are different again, because there the home page is often the offer page. On a store it's a router. ### The worked example: one account, a tenfold spread by destination This one's a service account rather than a store, from [a site migration I ran](/results/hand-wrist-institute-site-and-ads-cutover/). Over the thirty days before cutover, the ad groups pointing at the site's home page converted at 11.76% and 11.06%. The ad groups pointing at old blog posts converted at 0.88% and 1.11%, and one pointing at a condition page converted at 8.65%. Same account, same practice, roughly a tenfold spread by destination. That made [repointing the eight blog-post ad groups at purpose-built landing pages](/blog/the-51-percent-i-didnt-move/) an easy call, and it's the number to pull on any account before building anything new. The home page doing well there fits case four: on a service site, the home page is often the offer page. On a store, run the same check on your own campaigns. On most Shopify stores, cases one and two matter more than case three. ### What this means for the build list Most Shopify stores buying traffic need zero new pages and better existing ones: product pages and collections made worth the spend. That's facelift work, storefront work from $5,000, not new pages. Stores running argument-led campaigns need one to three landing pages, built fast, tracked from day one, and disposable once the campaign ends. Either way, two things belong in the spec. Conversion tracking wired into the destination before spend resumes, because a page the ad platform can't measure is a page that teaches the algorithm nothing. And page speed measured on the destination specifically, on a phone, because speed on the pages you buy traffic to is priced into every click. The fix order is at [why Shopify stores get slow](/shopify-storefront/why-is-my-shopify-store-slow/). When I run the ads and build the destination, the two get designed together, which is the point of [the storefront work](/storefronts/). The free [Google Ads Setup Checklist](/audit/) covers the destination check alongside the account settings, with no email gate on the download. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Why is my Shopify store so slow? URL: https://connercrowe.com/shopify-storefront/why-is-my-shopify-store-slow/ Published: 2026-07-29 | Updated: 2026-10-21 Answer: Shopify stores get slow from app scripts on every page, oversized images, too many font files and old theme code; Shopify's hosting is rarely the cause. Google calls a page good at LCP of 2.5 seconds or less, INP of 200 milliseconds or less and CLS of 0.1 or less, measured at the 75th percentile of page loads. Slow is expensive twice: buyers leave before the page paints, and the ad platforms charge you more to send traffic to a page they score as bad. The causes rank in a predictable order. ### The thresholds to measure against | Core Web Vital | Good, at the 75th percentile of page loads | |---|---| | Largest Contentful Paint (LCP) | 2.5 seconds or less | | Interaction to Next Paint (INP) | 200 milliseconds or less | | Cumulative Layout Shift (CLS) | 0.1 or less | Google's Core Web Vitals define "good" as a Largest Contentful Paint of 2.5 seconds or less, an Interaction to Next Paint of 200 milliseconds or less, and a Cumulative Layout Shift of 0.1 or less. Google measures each at the 75th percentile of page loads, split by mobile and desktop ([web.dev](https://web.dev/articles/vitals), checked October 21, 2026). Those are the numbers to hold a store against, on the pages your ads actually land on. ### Cause one: app scripts Every app you install earns the right to inject script on every page, and most take it whether the page uses the app or not. The review widget loads on the blog. The bundle app loads on the contact page. The abandoned-cart tool loads on a policy page nobody abandons a cart from. Uninstalling helps less than founders expect, because many apps leave their code in the theme after removal. That's the sediment problem in cause four, arriving early. **Diagnosis:** open your storefront with your browser's network tab and list the third-party domains. Any you can't match to an app you use is a suspect. Then cross-check against your app list and your billing. Plenty of stores are paying for at least one app nobody has opened in a year, and it's still loading on every page view. **The fix:** an app audit. Remove what you don't use, then have the leftover code stripped from the theme. This is the highest-return speed work on most stores. ### Cause two: images Furniture-sized hero images shipped at 4,000 pixels wide for a phone screen. Product photos uploaded straight from the supplier at print resolution. Shopify's CDN resizes images when the theme asks correctly. Older themes and hand-edited ones often don't ask correctly, so the browser downloads the full-resolution file and shrinks it on the client, paying the full transfer cost for none of the benefit. **Diagnosis:** if your largest homepage image weighs more than a few hundred kilobytes on your phone, the theme isn't sizing responsively. **The fix:** lives in the theme's image rendering, not in manually shrinking uploads forever. Fix it once in the template and every future upload inherits the fix. This is worth real attention for high-consideration catalogs, where buyers want to see the product at full scale and much of the traffic arrives on phones. The imagery has to be big enough to sell and small enough to arrive, and only the template can hold both. ### Cause three: fonts Each font family and weight is a file the browser must fetch before text settles. Themes configured with a display font, a body font, and five weights of each ship ten font files when the design uses a few. Two families, two or three weights total, covers almost any brand and cuts font weight by half or more. This is a theme settings change on most modern themes, which puts it inside the founder's own reach. ### Cause four: the theme itself Legacy themes and heavily patched ones carry years of sediment: sliders nobody uses, scripts from apps deleted in 2023, CSS for sections that no longer exist. Past a point, cleanup inside the old theme costs more than moving to a current one. That threshold decision is covered in [facelift or full rebuild](/shopify-storefront/shopify-facelift-vs-full-rebuild/). ### What's almost never the cause Shopify's hosting. The platform serves fast globally, and switching plans won't fix a slow store. Two more false leads worth naming. A speed-optimizer app, which adds a script to fix a script problem. And a theme swap bought for speed alone, which moves your slow content into a fast frame and leaves you at roughly the same number. ### Where speed sits in the build, and why that saves money Speed work costs less inside a project than after one. On my storefront builds, page speed is a build spec item measured before launch rather than a cleanup engagement later. A build that starts at $5,000 includes it, and most of my builds launch in two to three weeks from kickoff. Bought as remediation afterward, the same work is a separate engagement, plus whatever the slow months cost in ad spend on pages the platforms were scoring down. The pattern repeats with tracking, which is the other thing redesigns leave for later. Wired during the build it's part of the number. Bought afterward it's typically a $2,500 project on its own. The published prices are on [my pricing page](/pricing/). ### The fix order and the measurement rule Apps first, images second, fonts third, theme sediment last. That order is by return per hour, and it holds on almost every store I open. Measure on a phone over cellular before the first change and after every change, using the same page each time. One variable, one measurement, or you'll never know which change earned the improvement. When the fix lives in a new or edited theme, my method is to compare two unpublished themes against each other: a fresh duplicate of the live theme and the edited one. Shopify's preview bar loads on any theme you preview, and it skews Lighthouse, so a preview tested against the live store isn't a fair comparison. Measure your paid landing pages specifically, not your homepage. Speed there is priced into every click you buy, and the homepage is rarely where paid traffic lands. Which page paid traffic should land on is its own decision, covered at [product page or landing page](/shopify-storefront/shopify-ads-product-page-or-landing-page/). ### What to do this week If you have never audited the apps, start there and nowhere else. Open the network tab, count the domains, match them to your app list and your invoice, and remove what nobody uses. Then have the residue stripped out of the theme, because uninstalling an app doesn't uninstall its code. If the store is already lean and still slow, the problem is the theme's age, and the honest fix is structural. When I build or facelift a storefront, speed ships inside the project rather than as a cleanup later, which is cheaper in both directions. The way that works is on [my Shopify storefronts page](/storefronts/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Will redesigning my Shopify store hurt SEO? URL: https://connercrowe.com/shopify-storefront/will-a-shopify-redesign-hurt-seo/ Published: 2026-07-28 | Updated: 2026-10-21 Answer: It can, and most losses come from changed URLs without redirects, not the design itself. On a site migration I ran, I inventoried 533 legacy URLs before building and confirmed 40 of 40 sampled redirects after cutover. On Shopify, that means mapping every changed handle, because a redirect only fires once the old URL returns 404. Inventory first, redirect everything. The redesign itself doesn't hurt SEO. The five specific mistakes redesigns make hurt SEO, and every one of them is preventable if someone inventories what the store ranks for before anything gets touched. ### Risk one: URLs that change without redirects Shopify keeps product and collection URLs tied to handles, so a theme change alone doesn't move them. The damage happens when the rebuild "cleans up" handles, restructures collections, or drops pages entirely. Every changed URL needs a 301 redirect, and every dropped page needs a decision, not a 404. The redirect map ships with the launch, not as a post-launch cleanup, and it should be a document you can read rather than a promise. Two Shopify specifics shape that map. First, a URL redirect only takes effect on a broken URL. Shopify's [help center](https://help.shopify.com/en/manual/online-store/menus-and-links/url-redirect) says that if the old URL still loads a valid page, the redirect won't work, so a product still live at its old handle can't be redirected away until it returns a 404. Second, when you edit a product or collection handle in the admin, Shopify offers to create a URL redirect from the old address; leave that option on, and check it. The inventory comes before the build. On the last site migration I ran, I crawled 533 legacy URLs before anything was built, 348 of them blog posts, and after cutover I sampled 40 redirect rules against the live domain. All 40 were correct. That one wasn't Shopify, but the method carries over unchanged: list every URL first, then prove the redirects after launch. The failure is quiet. A 404 doesn't alert anyone. It shows up six weeks later as a collection that used to bring traffic and now brings none, by which point the vendor has been paid and moved on. ### Risk two: internal links that quietly disappear Rankings lean on internal linking: the collection copy that links related categories, the blog posts that point at products, the footer paths crawlers follow. New templates frequently ship without the link blocks the old theme carried, and the loss is invisible in a design review because the page looks cleaner without them. The fix is an internal-link inventory of the top-ranking pages before the rebuild, checked against the new templates before launch. I take this seriously on my own site for the same reason I take it seriously on client stores. Internal links are a ranking signal you fully control, and they're the first thing a redesign deletes for aesthetic reasons. ### Risk three: structured data that leaves with the theme Product schema, review stars, and breadcrumbs mostly live in theme code. Swap the theme and the rich results that took months to earn can vanish in a single deploy. (FAQ markup no longer earns a rich result at all: Google stopped showing FAQ rich results in May 2026.) Run the old store's key templates through a structured-data test before the build starts, save the output, and hold the new templates to the same coverage. This takes an afternoon and it's the cheapest insurance on the list. ### Risk four: slower templates Redesigns add weight: hero videos, extra font families, app embeds, animation libraries. Core Web Vitals are a ranking input and a conversion input, and a redesign that ships slower than the store it replaced starts underwater. Speed belongs in the build spec with a number attached, measured on a phone over cellular, not on the designer's laptop over office wifi. Google's "good" thresholds, judged at the 75th percentile of page loads, are a Largest Contentful Paint of 2.5 seconds or less, an Interaction to Next Paint of 200 milliseconds or less, and a Cumulative Layout Shift of 0.1 or less ([web.dev](https://web.dev/articles/vitals)). Measure the same page before and after. One testing habit of mine: I compare the new theme and the old one as two unpublished theme copies, never a preview against the live store, because the Shopify preview bar skews Lighthouse on the theme being previewed. The causes and the fix order are in [why Shopify stores get slow](/shopify-storefront/why-is-my-shopify-store-slow/). ### Risk five: thinner content New designs favor clean, sparse pages, and collection descriptions or product copy get cut for looking cluttered. The copy was carrying rankings. Content parity is a launch checklist item: everything that existed either survives, moves somewhere better, or gets retired on purpose with a redirect. "We tightened the copy" is a decision that needs an owner who understands what the copy was earning. ### The version of this that also burns your ad account The same inventory logic applies to paid traffic, and the damage arrives faster there. Campaigns point at specific URLs. Change them without updating the account and ads land on redirects or dead pages while Quality Score pays the price. New landing page templates also need the conversion tracking re-wired and validated before spend resumes. Redirects can break the ad account even when they work for SEO. On one account I audited, redirects returning a clean 301 were stripping the gclid from 860 of 3,519 paid clicks in a thirty-day window. Every status check passed, and the conversions came back attached to nothing. A redirect map for a store running paid traffic has to preserve the query string, not just the status code. The full failure mode is in [A 200 Is Not Proof](/blog/a-200-is-not-proof-redirects-that-eat-the-gclid/). This is the argument for the person who runs the traffic running the build. I know which URLs the spend lands on and which pages rank before I touch anything, because I'm the one who will be answering for both next month. ### The low-risk path If your pages rank and the theme isn't blocking the catalog, [a facelift](/blog/shopify-storefront-facelift-without-a-rebuild/) carries almost none of these risks. On one live store I chose a facelift over a rebuild for exactly this reason, and because the theme stayed, the URLs, schema, and collection copy stayed with it. That decision test is in [facelift or rebuild](/shopify-storefront/shopify-facelift-vs-full-rebuild/). When a full rebuild is earned, the five risks above are closable with an inventory-first process: rankings inventoried, links inventoried, schema captured, speed budgeted, content mapped, all before the theme work starts. That's how [my Shopify storefront builds](/storefronts/) run, and the staged cutover that protects the launch itself is at [redesign without losing sales](/shopify-storefront/redesign-shopify-store-without-losing-sales/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ --- # Answers: Conversion Tracking Hub: https://connercrowe.com/conversion-tracking/ ## How do I know whether Meta or Performance Max drove a sale? URL: https://connercrowe.com/conversion-tracking/meta-or-performance-max-which-drove-the-sale/ Published: 2026-10-15 | Updated: 2026-10-21 Answer: You can't tell from either platform's report, because both claim the same sale. Judge both against your Shopify orders or booking system. On one store I manage, Google claimed 0.83 purchases per online order and Meta 0.07. Then test: hold one channel back in a region or period and watch the total. Calls and bookings need their own tracking. Every store owner running both platforms hits this the first time they add the two dashboards together. Both will usually claim the same order, and both are telling the truth by their own rules. ### Why both platforms claim the same order A shopper clicks your Instagram ad on Monday, searches your brand on Thursday, clicks a Performance Max Shopping ad and buys. Meta counts the sale because the Monday click falls inside its default 7-day click window. Google counts it because of Thursday's click. The store recorded one order. Each platform measures with its own window and its own rules, and each is built to show its own value. Add the two dashboards together and you'll always count some sales twice. It's two scorekeepers watching the same game. Performance Max adds a second wrinkle: it bids on your own brand searches unless you stop it, so it gets credit for people who were already coming to you. The fix is brand exclusions on Performance Max, plus your brand terms as phrase negatives on a shared list applied to every non-brand campaign. That's covered in [why Performance Max gets credit for Shopping conversions](/blog/why-performance-max-gets-credit-for-shopping-conversions/). ### Pick one neutral source of truth For a store, that's the order record in your store platform. For a service business, it's the booking system or the CRM where a lead becomes a signed client. The platforms are useful for steering inside each channel. The neutral record is what you judge them against. Then compare totals over the same period. If the platforms together claim more sales than the store made, some of those claims overlap or are being counted twice. The question becomes what each channel adds on top of the other. ### What the gap looked like on two stores I run I pulled recent data for two online stores whose ad accounts I manage and put the platform claims next to each store's own order record. To keep the stores unidentifiable I'm publishing ratios, not order counts. | | Store 1 | Store 2 | |---|---|---| | Window | August 17 to October 14, 2026 | July 1 to September 30, 2026 | | Google Ads purchases claimed per order | 0.83 (online orders) | 0.51 (all orders) | | Meta purchases claimed per order | 0.07 | Meta not running | | Combined platform claims per order | 0.89 | 0.51 | **Neither store's platforms claim more than the store sold.** That's the first thing to check, and both pass. Claims below the order count are normal: some buyers come back without clicking an ad, some come through email or search, some buy in ways the tags never see. **Store 1 still had a duplicate hiding inside a healthy number.** For several weeks in September a second purchase action, from the Google tag, was set to count toward bidding alongside the main one from the store's sales-channel app. About one in eight of the purchases Google claimed in this window came from that second action, so some sales were counted twice and bidding learned from the inflated figure. It has since been set to stop counting. The total still looked believable, which is why this kind of duplicate survives for weeks. **I nearly published the wrong ratio myself.** My first pull compared a full quarter of ad claims against the store's order export, and showed Google claiming one and a half purchases per order. The order API only returns the last 60 days unless the app has full order history access, so half the quarter's orders were missing. Same window on both sides, every time, or the comparison means nothing. Neither number tells you what the ads caused. They tell you whether the scoreboard is sane enough to start testing from. ### Test what each channel really adds The honest answer comes from switching something off and watching the total. - **Hold out a region.** Pause Meta in one state or city for a few weeks and compare total sales there against similar areas where it kept running. - **Hold out a period.** Pause one channel for a set window, long enough for the data to settle, and watch whether total orders move. - **Watch new vs returning buyers.** A channel that mostly reaches people who already know you is worth less than its report says. None of this needs expensive software to start. It needs clean tracking and the patience to change one thing at a time. ### Phone calls and bookings: the sales nobody counts For a service business, the biggest attribution gap usually isn't Meta versus Google. It's that most of the real conversions happen on the phone or in a booking system the ad platforms never see. The fix has two parts. First, give each ad source its own tracked number so a call is tied to the ad that caused it, and record the calls so you can score which ones were real leads. Second, send qualified leads and booked jobs back to Google as offline conversions, so the bidding learns from the ones that turned into revenue instead of every ring. The setup is in [how to track phone calls as Google Ads conversions](/lead-quality/track-phone-calls-conversions-google-ads/) and [importing closed deals into Google Ads](/lead-quality/import-closed-won-deals-google-ads-offline-conversions/). The full measurement setup I build is documented as [The Tracking Stack](/frameworks/tracking-stack/), and [the free ads audit](/audit-request/) shows where your current numbers are double counting. ## How do I set up the Meta Conversions API the right way? URL: https://connercrowe.com/conversion-tracking/meta-conversions-api-setup/ Published: 2026-06-28 | Updated: 2026-10-30 Answer: Run the Conversions API beside the pixel. Send the same events from both, give each pair the same event_id and event name so Meta keeps one, and pass hashed customer data for matching. Dedup is where builds break. On one store I audited, two CAPI builds with mismatched event IDs double-sent every purchase for six months. ### What the Conversions API is The Conversions API (CAPI) sends conversion events to Meta from your server instead of from the browser. The pixel still fires client-side. CAPI runs alongside it, server-side. The point is coverage. Browser events get blocked by ad blockers, iOS settings, and consent tools. Server events don't, because they leave from your infrastructure, not the visitor's device. Treating CAPI as a switch you flip to "fix tracking" is how stores end up worse off. On one store I audited, Meta's 2024 CAPI auto-setup was still live in GTM beside a newer server-side build. Both sent every purchase, their event IDs didn't match, and Ads Manager carried inflated purchase counts for six months. A bad CAPI build makes reporting worse than no CAPI at all, because now you have two streams that double-count or contradict each other. The architecture is what makes it work. ### The two-stream model Run the pixel and CAPI as a matched pair. Every key event fires twice: once from the browser through the pixel, once from your server through CAPI. PageView, ViewContent, AddToCart, InitiateCheckout, Purchase. Both sides send the same event for the same user action. That sounds like double-counting, and it would be, except for one field. You send a shared `event_id` on both the pixel call and the server call, with the same event name on both. When Meta receives that pair for the same pixel within 48 hours, it keeps one and discards the duplicate. Without event IDs, Meta can fall back to the event name plus `fbp` or `external_id`, but that fallback generally only works when the browser event arrives first, so don't build on it. This is deduplication, and it's the single most important part of the setup. Get it wrong and your Purchase count inflates, your ROAS looks better than reality, and the account optimizes toward a number that doesn't exist. So the build order is: pixel first, CAPI second, dedup keys wired before you trust a single number. ### What CAPI needs to match A server event with no identity is close to useless. Meta has to tie the event back to a person to attribute it and to optimize. You do that by passing customer information parameters, hashed with SHA-256 before they leave your server. The fields that move match quality the most: - Email (hashed) - Phone (hashed) - `fbp` and `fbc` cookies, captured from the browser and forwarded to the server - IP address and user agent - External ID, if you have a stable customer id The `fbc` cookie carries the click id from the ad click. Forwarding it's what lets a server-side Purchase get attributed back to the specific ad that drove it. Skip it and you lose attribution on a large share of conversions even though the event itself arrived. Match quality shows up in Events Manager as the Event Match Quality score per event. Below "Good" on Purchase, fix identity before you touch anything else. ### How to send the events There are three common paths, in rough order of how much control they give you. **Direct server-to-server.** Your backend posts events to the Graph API endpoint. Most control, most match potential, most engineering. This is the right call for a custom stack or a high-spend account where match quality pays for itself. **Server-side Google Tag Manager (Stape or your own container).** The browser sends events to a server container you own, and that container forwards them to Meta with full data. This is what I reach for on Shopify and most lead-gen sites. You get server-side identity and dedup without writing API calls by hand, and one container can feed Meta, GA4, and TikTok from the same event stream. **Native platform integration.** Shopify's built-in Meta channel and partner connectors send CAPI for you. Fast to turn on. The catch is you give up control of dedup keys and the exact data sent, so verify in Events Manager that events deduplicate and match well before you rely on it. Auto-setup tools carry the same risk. On the store with the six-month double send, the old auto-setup had also written the CAPI access token into a GTM tag in plaintext, so every saved container version held a live token anyone with read access could copy. Keep the token in a variable, and check what any automated setup left behind before you trust it. ### Verify before you trust it Setup isn't done when the connection turns green. It's done when you have confirmed the numbers reconcile. 1. Open Events Manager and check that each event shows both "Browser" and "Server" as sources. 2. Confirm the deduplication rate is high. Low dedup means your `event_id` isn't matching across the two streams. 3. Check Event Match Quality on Purchase and Lead. Push the weak ones up with more identity fields. 4. Compare Meta's reported Purchase count against your real backend over the same window. Shopify orders, your CRM, your database. If Meta is meaningfully higher, you have a dedup or double-fire problem. If it's much lower, you have a coverage or match problem. That last step is the one people skip, and it's the only one that tells you the truth. The platform UI will happily show you a healthy-looking setup that's double-counting. Reconciling against the source of record is how you catch it. ### Where it sits in the stack CAPI is one layer of a tracking stack, not the whole thing. It depends on a clean event layer feeding it, consistent event names across pixel and server, and a consent setup that doesn't strip the data you need. If your GA4 and Meta numbers already disagree, adding CAPI on top of a broken foundation just gives you a third number to reconcile. The store with the double send had the same problem underneath: it was paying for first-party loading, yet its theme still pulled GTM from googletagmanager.com, so ad blockers caught it on every visit and recovery from them read 0 percent. Safari recovery on the same server container was working, at 45 percent of events. The [full audit](/blog/bing-pixel-your-value-here/) is on the blog. Fix the event layer and identity capture first, then add the server stream, then verify it reconciles. In that order it earns its place. Out of order it adds noise. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## How does Shopify Checkout Extensibility break my tracking, and how do I fix it? URL: https://connercrowe.com/conversion-tracking/shopify-checkout-extensibility-tracking/ Published: 2026-06-28 | Updated: 2026-10-22 Answer: Checkout Extensibility removed checkout.liquid and Additional scripts, so any purchase pixel pasted there stopped firing. Rebuild checkout events as Web Pixels under Customer events and send the purchase server-side. On one store, the broken purchase event left Google Ads well short of GA4's count until I fixed it, validated it end to end, and the two landed within 4%. ### What Shopify changed For years, Shopify stores tracked the checkout with code pasted into it. Plus stores could edit `checkout.liquid`, and stores on every plan had the "Additional scripts" box for the order status and thank-you page. That's where most stores pasted their Meta Pixel, GA4 tags, TikTok pixel, Pinterest tag, and any custom purchase tracking. It worked because those scripts ran directly on the checkout and order status pages. Shopify retired that model. `checkout.liquid` and the additional scripts field are gone, replaced by Checkout Extensibility. Checkout is now built from Shopify's own upgraded checkout plus app extensions, and you no longer get to inject raw script tags into it. The order status page and post-purchase page moved with it. So if your purchase event lived in additional scripts or `checkout.liquid`, it stopped firing the day your store migrated. The storefront tracking up to the cart still runs. Checkout and purchase are the part that goes dark. ### The deadlines have passed [Shopify ran this in stages](https://shopify.dev/docs/storefronts/themes/architecture/layouts/checkout-liquid), and every stage has now passed. On Plus, `checkout.liquid` no longer works on the information, shipping, and payment steps. On August 28, 2025, Shopify sunset `checkout.liquid` and Additional scripts on the Plus thank-you and order status pages. On non-Plus stores, script tags on those pages ended on August 26, 2026. If nobody rebuilt the tracking, assume any purchase tracking that lived in those scripts is dead. Check Settings, then Checkout, to see what your store is running now. Here's what the upgrade broke and what kept working. Plus stores lost `checkout.liquid` too, and the fix is the same: Web Pixels plus the server-side setup below. | Broke when the checkout upgraded | Still works | |---|---| | Pixels and tags pasted into Additional scripts | Theme-page events: page views, product views, add to cart | | Code in `checkout.liquid` on Plus | App pixels registered under Customer events | | Script-tag apps that injected code into checkout | Server-side purchase events sent from the order | | Custom purchase scripts and their dedup logic | The pixels from Shopify's Google & YouTube and Meta channel apps | The trap is that the storefront still looks healthy. PageView, ViewContent, and AddToCart keep reporting because those fire on theme pages, not checkout. Only the bottom of the funnel breaks. That's exactly the part you bid on, so a store can run for weeks thinking conversions are flat when really the purchase event quietly went to zero. ### Where checkout tracking lives now The replacement is the Web Pixels API, which Shopify surfaces in the admin as Settings, Customer events. This is a sandboxed JavaScript environment that subscribes to standard events: `checkout_started`, `payment_info_submitted`, `checkout_completed`, and the earlier funnel events like `product_viewed` and `product_added_to_cart`. The sandbox is the important word. A Web Pixel runs in a Web Worker, isolated from the page DOM. It can't read the page, can't touch cookies the way an inline script could, and can't rely on globals you used to set in the theme. You get a structured event payload and a small API for storage. Code written for additional scripts won't drop into a Web Pixel unchanged. It has to be rewritten against the event schema. There are two ways to use it: - **App pixels.** Meta, GA4 via Google and YouTube channel, TikTok, and Pinterest all ship Shopify apps that register their own Web Pixel automatically. For most stores this is the cleanest path. Connect the channel, confirm the pixel is registered under Customer events, and the platform handles the event mapping. - **Custom pixels.** For anything the apps don't cover, you write a custom pixel in the Customer events screen, subscribe to the events you need, and forward them yourself. One boundary on this: Google documents that running Google tags inside the custom pixel sandbox isn't a supported implementation, and enhanced conversions are one of the things that may not work correctly from there. Use custom pixels to forward events to your own endpoint or to platforms that support it, and keep Google Ads conversions on the Google & YouTube app or on the page itself. A theme-page conversion doesn't need the sandbox at all, which is [the defect I shipped into my own build](/blog/shopify-custom-event-no-subscriber-google-ads-conversion/). ### Rebuild the purchase event server-side Client-side checkout tracking inside the sandbox is more limited than the old inline script, and browser-side events still get blocked by ad blockers, iOS, and ITP. So the reliable fix is to stop depending on the browser for the sale. One store I work with shows what's at stake. After its checkout migration, the purchase event broke, and Google Ads recorded far fewer conversions than GA4 did. Once I fixed the event and validated it end to end, the two counts landed within 4% of each other. On a server-side build I run, measured over ten days, the first-party server path delivered about a third of all events and half of purchases that browser tracking prevention would otherwise have dropped. Send the purchase server-side. For Meta that's the Conversions API. For GA4 that's the Measurement Protocol, or a server-side GTM container. The cleanest version routes the order through a server endpoint that fires the purchase with the order ID, value, currency, and hashed customer data, then dedupes against any browser event using a shared event ID. Shopify's order webhook, or the `checkout_completed` event passed to your server, gives you a trustworthy purchase signal that doesn't depend on the customer's browser finishing a script. That's the signal you want feeding your ad platforms, because it reconciles to real orders in Shopify. A practical setup I keep coming back to: - Web Pixels (app or custom) for the browser-side funnel events, so platforms still get the in-session signals they use for optimization. - Conversions API and Measurement Protocol for the purchase, fired from the order, deduped by event ID. - A single source of truth for revenue, which is Shopify orders, not the pixel. ### How to confirm it's working Don't trust the admin toggle. Verify each layer. 1. Check Settings, Customer events, and confirm each platform's pixel is registered and the custom pixels you expect are present. 2. Place a real test order. Watch Meta Events Manager Test Events and the GA4 DebugView for `checkout_completed` and purchase firing once, not zero times and not twice. 3. Reconcile a day of Shopify orders against purchases reported in Meta and GA4. If the platform numbers sit far below Shopify, something in the chain is dropping. 4. Confirm deduplication. If browser and server both fire without a matching event ID, you'll double-count and your ROAS will read better than it is. Step 3 is the one worth doing properly, because the gap runs in both directions. On the one account in my [Ecommerce Tracking Accuracy Benchmark](/benchmarks/ecommerce-tracking-accuracy/) where I could read both the order table and the ad account, the ad platform's counted purchases over 90 days equaled 58.5% of the store's entire order book, and its all-conversions purchase total equaled 65.9%, while the store's own journey data classified none of those orders as paid. That's one account, N = 1, and it's enough to show the comparison is worth running on your own store. If you migrated to Checkout Extensibility and your reported conversions fell off a cliff while Shopify revenue held steady, look at the checkout events first, because they're the ones the migration moved. The sale still happened. The tracking just lost the page it used to live on. Rebuild it on Web Pixels plus server-side events, then verify against real orders before you trust a single number in the ad dashboards. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## What are enhanced conversions in Google Ads and how do I set them up correctly? URL: https://connercrowe.com/conversion-tracking/enhanced-conversions-google-ads/ Published: 2026-06-28 | Updated: 2026-10-21 Answer: Enhanced conversions send hashed first-party data, such as a checkout or form email, so Google can match each conversion to the ad click a cookie lost. One failure I find is the setting marked enabled while the hashed-email field arrives empty. As of October 2026, enhanced conversions for leads was on in 18 of my 28 lead-gen accounts. ### What enhanced conversions do Enhanced conversions take the first-party data a customer already gave you at checkout or on a lead form, hash it with SHA-256, and send it alongside the conversion event. Google matches that hashed identifier against signed-in Google accounts to attribute the conversion back to an ad click that the cookie alone lost. The problem they solve is recovery. Safari and Firefox cap or strip third-party cookies, iOS limits click identifiers, and consent banners block a slice of traffic outright. So a real conversion happens, but Google never connects it to the click that drove it. That click looks like wasted spend. Smart Bidding then optimizes against an incomplete picture. Enhanced conversions hand Google a second way to make the match, so more of your real conversions show up in the account and bidding gets a cleaner signal. The data never leaves in plain text. When you pass email, phone, name, and address to the Google tag, directly or through GTM, the tag normalizes and hashes them with SHA-256 before anything is sent. You don't hash them yourself on the web path. Google compares hashes, not raw values. ### One switch, two different jobs Google used to ship two products under one name, each with its own setting. In 2026 Google began combining enhanced conversions for web and enhanced conversions for leads into a single on/off setting ([Google's notice](https://support.google.com/google-ads/answer/16884284)). Even with one switch, the two jobs it covers are still different. **Enhanced Conversions for Web** improves measurement for conversions that already fire on your site, like a purchase or a form submit. It supplements the existing conversion with hashed user data so more of those events match. This is the e-commerce path. **Enhanced Conversions for Leads** is for businesses that close offline. The lead submits a form, the tag captures the email with the lead, and later you upload the offline conversion (the closed deal) keyed to that same email. Google ties the ad click to the eventual sale. If you run a service business and close on the phone or in person, this is the one you want. Know which job your conversion is doing before you touch anything. If the business closes offline and the closed deal never comes back as an upload, the lead side does nothing for bidding. Across the 33 Google Ads accounts I manage, in the 30 days to October 21, 2026, enhanced conversions for leads was on in 18 of the 28 lead-gen accounts, and only 6 of those 28 bid on an offline conversion import. ### The three setup paths Since April 2026 Google accepts user-provided data from the tag, Data Manager and the API at the same time, so these aren't exclusive. They differ in how much you control. **Google Tag (gtag.js) with automatic collection.** The lightest path. You enable enhanced conversions in the conversion action settings, and the tag scrapes user data from the page using CSS selectors or automatic detection. It works, but it's fragile. A theme update that renames a field or moves the confirmation data breaks the scrape silently. I treat automatic collection as a starting point, not a finished build. **Google Tag Manager with manual variables.** The path I default to. You capture email, phone, and address into dataLayer variables at the point the user submits, then map those variables into the conversion tag's user-provided data fields. It survives front-end changes because you control where the data comes from, and you can see the values flowing in GTM Preview before you publish. **An API or offline import.** For leads that close in a CRM. You push the identifier and the conversion value back through an API or a scheduled upload. Since 15 June 2026, Google has moved offline conversion imports and enhanced conversions for leads uploads to the Data Manager API. The Google Ads API keeps them only for developer tokens Google allowlisted because they were already sending these uploads in the first half of 2026, so a new build starts on Data Manager. On an API upload you normalize and hash the identifiers yourself, to Google's spec. This is where value-based bidding gets real fuel, because you can send the actual deal size, not a flat lead value. ### What decides your match rate Setup is half the job. The match rate is what tells you whether it worked, and a few things move it. Send more fields. Email is the strongest single identifier, but email plus phone plus name plus address gives Google more chances to land a match. Let the tag do the hashing. The Google tag and GTM's user-provided data variable normalize and hash the values for you, so send them plain and don't pre-hash them. Hand-normalizing (lowercase the email, strip whitespace, phone in E.164) and hashing only belongs on an API upload, where a messy string won't match the clean hash on Google's side. Respect consent. Under Consent Mode v2, the signal that governs enhanced conversions is ad_user_data. If the user denies it, the user data shouldn't be sent. Bolting enhanced conversions onto a site with no consent signal is a compliance problem, and in the EEA it will also throttle what Google can model. ### How to verify it's live Don't trust a green checkmark. Open the conversion action in Google Ads and look at [the diagnostics for that action](/blog/no-recent-enhanced-conversions-alert/). Within a couple of days of real traffic you should see the diagnostics report on the data coming in. How well it matches depends on how many fields you send and how clean they are. A match rate near zero means the data isn't arriving, the fields are empty or malformed, or consent is blocking it. Cross-check in GTM Preview or the browser network tab that the user-provided data fields actually carry values when the conversion fires. An empty field hashes to nothing and matches nothing. On a $40K-a-month account I inherited, enhanced conversions read as enabled while the hashed-email field was being dropped before it reached Google, so Google received the wrapper without the payload. The [audit I run on inherited accounts](/blog/google-ads-tracking-audit-guide/) checks for it. On Shopify, don't build this inside a custom pixel. Google's [custom pixel limitations page](https://support.google.com/google-ads/answer/16000892) lists enhanced conversions among the features that may not work from that sandbox, and recommends its Google & YouTube app for sitewide measurement. Pick one install path. The most common Shopify break I find is a duplicate install, the Google & YouTube app plus a GTM Google tag firing the same purchase twice. ### Where this sits in the stack Enhanced conversions are one layer, not the whole tracking system. They make your existing conversion measurement more complete. They don't fix a conversion that never fires, a checkout event that breaks on mobile, or a lead form that GTM can't see. They also won't tell you that the action you're enhancing is the wrong action. On two of the three accounts in my [Ecommerce Tracking Accuracy Benchmark](/benchmarks/ecommerce-tracking-accuracy/), measured May to August 2026, 16.6% to 17.2% of what the Conversions column counted was add-to-cart, not a sale. My Q3 account benchmark found 0 of 4 purchase accounts doing it, so it's a defect to check for, not a rate. Raising the match rate on that only sharpens a signal pointed at the wrong event. Get the base conversion firing reliably first, confirm it in a controlled test, then layer enhanced conversions on top to recover the matches the cookie lost. Built in that order, you get a measurable lift in recorded conversions and a Smart Bidding signal that reflects what actually happened. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## What conversion value should I send to Google and Meta, and how do I keep them reconciling? URL: https://connercrowe.com/conversion-tracking/conversion-value-to-send-google-meta/ Published: 2026-06-28 | Updated: 2026-10-21 Answer: Send Google and Meta the same number: the order subtotal after discounts, before tax and shipping, in one currency, with the order ID, server-side where possible. Then check what the value column holds. On two of three stores I measured from May to August 2026, 19.8% to 25.8% of reported conversion value was cart value. ### Pick one definition of "value" and send it everywhere When Google and Meta disagree on revenue, the first thing I check is whether each one received a different number for the same order. One got the grand total with tax and shipping, the other got the subtotal. One got the value in USD, the other in the customer's local currency. Once that happens, no amount of dashboard tweaking will make them line up. So the first decision is what "value" means, and it has to be made once. My spec is the order subtotal after discounts, before tax and shipping, in your store's base currency, sent with the order ID. That's the number closest to what the customer paid you for the goods, and it maps to margin. Tax and shipping inflate ROAS without adding a dollar you can spend. If you send the grand total to one platform and the subtotal to the other, your blended ROAS is a fiction. Reconcile gross to gross. The platforms receive the value at the moment of purchase, so compare them against the same orders' subtotals in Shopify before any refunds come out, and handle refunds and cancellations as a separate step. Comparing platform value to a Shopify figure that has already netted out returns makes a healthy setup look inflated. Whatever you choose, write it down. The definition is a one-line spec: "value = order subtotal after discounts, excluding tax and shipping, in USD, keyed to the order ID." Every conversion tag, server event, and offline import then has to honor that one line. Then check the platform side, because a setting can change the number after it arrives. On one Shopify account I took over, [a $32.99 order showed in Google Ads as $148.95](/blog/google-ads-revenue-doesnt-match-ga4-shopify/). The dataLayer was clean. The prior agency had turned on Customer Lifecycle Value Optimization with a hardcoded $115.96 added to every new-customer conversion, and Smart Bidding had chased the padded number for nine months. The platform can also be aimed at the wrong event entirely: when I audited [Emma Grace Home](https://emmagracehome.com/), its Google Ads was [optimizing toward add-to-carts](/results/unassigned-revenue-dual-ga4-fix/) instead of buyers. ### Send the same number to Google and Meta Google Ads expects a `value` and a `currency` on the conversion. Meta's Purchase event expects `value` and `currency` in the same way. Send identical figures. - **Google Ads:** the `value` parameter on the purchase conversion action, with `currency_code` set explicitly. Don't leave currency to default. Enhanced conversions don't carry a value of their own. They attach hashed customer data to the conversion so Google can match it, and the value still rides on the conversion tag or, for offline conversions, on the import row. - **Meta:** the `value` and `currency` custom data on the Purchase event, sent through the Conversions API. The pixel-only value and the CAPI value must match, or Meta's deduplication will keep one and you won't always know which. If your store sells in multiple currencies, convert to a single base currency before the event fires, and tag the original currency in a separate parameter for your own records. Either way, send the currency code on every event. A value that arrives without one gets read in the account's default currency, so an order in yen can land as the same number of dollars. ### Fire it server-side, with the order id as the key Browser tags drop value constantly. Ad blockers, ITP, a customer who closes the tab on the thank-you page before the pixel loads. Every dropped purchase is revenue the platform never sees, which is one reason ad-platform revenue can read lower than Shopify. Send the purchase from the server: Shopify's order-created event, a server-side GTM container, or a direct Conversions API call. Server events fire whether or not the browser cooperated. On both platforms, attach a stable identifier so the browser event and the server event collapse into one: - **Meta:** `event_id` on both the pixel and the CAPI call. Same id, same order, deduplicated. - **Google:** the `transaction_id` (order id) on the conversion. This also lets you import corrections later without double-counting. Use the actual order id, not a timestamp or a random string. When you reconcile a month later, the order id is what lets you trace a single sale across Shopify, Google, and Meta. ### Decide how you handle refunds and discounts Two edge cases need a written rule. **Discounts.** If a customer uses a 20% code, do you report the value they paid or the full subtotal? Report what they paid. The platforms optimize toward the value you feed them, and you want them chasing real revenue, not list price. Pick one rule and apply it to both Google and Meta. **Refunds.** A sale that gets refunded is still sitting in the platform as revenue unless you send a correction. Google accepts refund adjustments through the conversion-adjustment upload (a negative or retraction keyed to `transaction_id`). Meta's handling is weaker, so many stores accept a known gap there and account for it in reporting rather than trying to claw each refund back. Whatever you do, document the rule, because a 10% refund rate quietly inflates reported ROAS by roughly 10% on both platforms. On the one store in my [Ecommerce Tracking Accuracy Benchmark](/benchmarks/ecommerce-tracking-accuracy/) where I could read the order table, about a sixth of the orders in the window were refunded and about a tenth were cancelled, and the ad platform got no restatement for any of them. That's one store, and the platform's purchase count still included orders the store no longer held. Cancellations follow the same rule as refunds. Send the subtotal after discounts at purchase, then correct for cancellations and refunds in their own step, so you always know which number moved and why. ### Reconcile on a cadence, not a hunch Once the same value is firing server-side to both platforms with an order id, the platforms will still not match Shopify exactly, and that's fine. Attribution windows differ, view-through differs, the platforms claim credit Shopify assigns elsewhere. The goal isn't identical numbers. The goal is a stable, explainable gap. Before you trust the ratio, confirm the value column is measuring revenue at all. When I measured three ecommerce accounts for the benchmark, from May to August 2026, 19.8% to 25.8% of reported conversion value on two of the three was cart value rather than purchase revenue. A ratio built on that number is stable and wrong. Shopify is the ledger. The platform numbers are for steering inside each platform, and the ratio is how you catch them drifting. Pull three numbers for the same date range: Shopify order subtotals after discounts and before refunds, Google Ads conversion value, Meta purchase value. Track the ratio of each platform to Shopify week over week. When the value definition is clean, those ratios hold steady. When someone edits a tag, swaps the total for the subtotal, or breaks the currency field, the ratio jumps, and that jump is your alarm. If you haven't pinned the definition yet, start there before touching bids or budgets. On one $40K-a-month inherited account, [fixing the value signal along with the rest of the stack](/blog/google-ads-tracking-audit-guide/) made reported ROAS fall once the inflated value came out, while contribution-margin ROAS rose. The dashboard number looked worse and the business number got better. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## What is Google Consent Mode and how does it work? URL: https://connercrowe.com/conversion-tracking/google-consent-mode-explained/ Published: 2026-06-28 | Updated: 2026-10-27 Answer: Google Consent Mode is an API that tells your Google tags what a visitor agreed to, so they fire normally, send cookieless pings, or hold back. It doesn't collect consent. On one store I audited, visitors accepted the banner and the Google Ads beacons still went out as gcs=G1--, because nothing passed the yes along. ### What Google Consent Mode is Google Consent Mode is a JavaScript API that sits between your cookie banner and your Google tags. Your consent management platform (CMP) tells Consent Mode what the user agreed to. Consent Mode then tells every Google tag on the page how to behave. That's the whole job. It's easy to confuse the two halves. The CMP collects the choice. Consent Mode reads that choice and adjusts tag behavior. If you only install one half, you get a banner that looks finished but tags that still ignore consent, or tags that go silent because no signal ever reaches them. Consent Mode controls four signals: - `ad_storage`: whether advertising cookies can be set - `analytics_storage`: whether analytics cookies can be set - `ad_user_data`: whether user data can be sent to Google for ads - `ad_personalization`: whether data can be used for personalized ads and remarketing The last two arrived with Consent Mode v2 in November 2023, when Google began strengthening enforcement of its EU user consent policy, and the enforcement tied to them took effect in March 2024. If you use Google tags on EEA, UK and Swiss traffic for personalized ads or remarketing, those signals have to reach Google. Without them, remarketing and audience lists stop populating from that traffic. ### Default state, then update Consent Mode runs in two steps. First you set a default state before any tag fires. For EEA traffic the safe default is `denied` across all four signals. Then, once the user interacts with the banner, your CMP calls an `update` to flip the relevant signals to `granted` or leave them `denied`. ```js gtag('consent', 'default', { ad_storage: 'denied', analytics_storage: 'denied', ad_user_data: 'denied', ad_personalization: 'denied', wait_for_update: 500 }); ``` The `wait_for_update` value tells Google tags how long to hold before assuming denial. That window matters. If a tag fires before the default loads, it runs with no consent context, and you get the worst of both worlds: a hit that may breach consent and a signal Google can't model from. ### Basic mode vs advanced mode | | Basic mode | Advanced mode | |---|---|---| | Google tags load | Only after the user grants consent | On every page | | When consent is denied | Nothing fires, no pings | Cookieless pings with no identifiers or cookies | | Conversion modeling | No modeling input | Pings feed the modeling | This is the distinction most setups get wrong, and it changes what data you actually keep. In **basic mode**, Google tags don't load at all until the user grants consent. If consent is denied, nothing fires. No pings, no modeling input. You lose all visibility into denied-consent users. In **advanced mode**, tags load on every page but adjust behavior based on consent. When consent is denied, tags send cookieless pings: no identifiers, no cookies, just an anonymous signal that an event happened. Google uses those pings as input for conversion modeling, which estimates the conversions you can't directly observe. Advanced mode is what most advertisers want, because it feeds the modeling that fills the gap left by denied consent. Basic mode is simpler and more conservative, but you forfeit the modeled conversions. Pick deliberately. An account can land in basic mode by accident when the CMP blocks the tags from loading, and the owner never realizes modeling was on the table. ### Advanced mode still sends a request before consent "Cookieless" doesn't mean nothing leaves the browser. In advanced mode Google's tags load on arrival and send a ping while consent is denied. Google's [consent mode overview](https://developers.google.com/tag-platform/security/concepts/consent-mode) lists what it carries: a timestamp, the user agent, the referrer, whether the visitor arrived from an ad click, the consent state and a random number for the page load. No cookies are read or written, and Google truncates the IP for ads when ad storage is denied. The same page says the full page URL is still collected when ad storage is denied, and GA4's `collect` request shows it in `dl=`. For most US sites that's a measurement trade many advertisers accept for modeled conversions. The calculation changes on pages whose address names something sensitive: a condition, a treatment, a legal matter. A request from `/practice-areas/dui-defense/` tells Google what the page is about before the visitor has chosen anything. On those templates I run basic mode, or no Google tags at all, and measure through calls and booked appointments instead. I'm making a measurement call here, and I'm not a lawyer. Whether a pre-consent request matters where you operate is a question for your counsel. (Section added October 27, 2026; legal context is dated in [the pre-consent change log](/pre-consent-tracking/#what-changed).) The [pre-consent browser test](/pre-consent-tracking/check-what-your-website-sends-before-consent/) shows which mode your site is in. ### Conversion modeling When consent is denied and you're in advanced mode, Google fills the reporting gap with modeled conversions. It uses the cookieless pings plus aggregate, observed behavior to estimate conversions from users it couldn't track individually. Google requires a traffic threshold before modeling kicks in. Small accounts may not hit the volume needed, so the modeled numbers can stay flat even with advanced mode configured correctly. Modeled conversions show up inside your Google Ads conversion columns. They're estimates, not raw counts, so reconciling them against a server-side source or your backend will never be exact. That's expected behavior, not a bug. ### Where Consent Mode quietly breaks Consent Mode problems rarely show up as error messages. They're silent gaps you only find when conversions drop or audiences stop growing, or when you read the beacons. The failures to check for: **No default state set.** Tags fire before the consent default loads, so the `update` never has anything to override. Check that the default block runs first in the data layer, above your GTM container or gtag snippet. **CMP and Consent Mode not wired together.** The banner records a choice but never pushes an `update` call. Consent stays at the default forever. In GTM, confirm the CMP template is firing the consent update, not just writing a cookie. **Granted, but the beacons say otherwise.** On [one store I audited](/results/consent-mode-denied-for-consenting-users/), returning visitors had accepted the banner, the consent cookie and the CMP's own API both said yes, and every consent-scoped Google beacon still went out carrying `gcs=G1--`. Two causes stacked. For a returning visitor, the CMP's free tier never sent the granted update, because that feature sat behind the paid tier. A bridge tag that did push the update ran on DOM-ready, after the page-view and conversion beacons had already left. The fix read the consent cookie server-side and printed the granted state at the very top of the page, ahead of every tag, and the beacons flipped to `G111` the same day. **v1 signals only.** A setup built before 2024 sends `ad_storage` and `analytics_storage` but not `ad_user_data` or `ad_personalization`. European remarketing audiences slowly drain. Audit the tag for all four signals. **Basic mode by accident.** The CMP is set to block Google tags entirely, so advanced mode and its modeling never run. If your conversion volume from European traffic looks like a cliff rather than a slope, check whether tags are being blocked rather than adjusted. **Consent checks layered on top in GTM.** Some accounts use a CMP, native Consent Mode, and a GTM trigger condition that also gates the tag. Three layers, each capable of blocking, and a tag that fires far less than anyone expects. Simplify to one source of truth, and clear out what's left over. When I audited [my own container](/blog/click-text-trigger-trap/), a paused Cookiebot tag and two orphan consent triggers were still sitting there months after I'd removed Cookiebot. ### How to verify it works Open Google Tag Assistant or the browser console and watch the consent state. You want to see the `default` event fire first with denied values, then an `update` event after you interact with the banner, with the signals you granted flipping to `granted`. In GTM preview mode, the Consent tab shows each tag's required consent and whether it was met at fire time. Then read what the page actually sends. Accept your own banner, reload, open the network tab, filter for `collect` or `google`, and read the `gcs` parameter. After `G1`, the first digit is ad storage and the second is analytics storage: 1 is granted, 0 is denied, and a dash means no state was set. `G111` is both granted. `G100` is both denied. `G1--` after you said yes means nothing on the page passed your choice to Google. Run the test as a returning visitor, not in a fresh incognito window, because the first-visit path is the one every setup gets tested on, and the failure above only appeared on the second visit. If the default fires, the update fires, all four v2 signals are present, and your tags read them, Consent Mode is doing its job. If conversions still look wrong after that, the problem is downstream in your tag configuration or attribution, not in Consent Mode itself. Consent Mode governs the gate. It doesn't fix what happens once the tag is through it. Consent also gets blamed for gaps it didn't cause. When I measured three ecommerce accounts from May to August 2026 for the [Ecommerce Tracking Accuracy Benchmark](/benchmarks/ecommerce-tracking-accuracy/), the gaps I could measure came from configuration: on two of three accounts the Conversions column carried cart events, and on all three, 99.4% to 99.8% of All conversions wasn't a purchase. No account in that sample exposed a modeled-versus-observed split through the API, so consent effects couldn't be measured at all. Check the definitions before you blame the banner. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## What is server-side tracking? URL: https://connercrowe.com/conversion-tracking/what-is-server-side-tracking/ Published: 2026-06-28 | Updated: 2026-10-22 Answer: Server-side tracking sends conversion events to Meta and Google from a server you control, alongside the browser pixel, so less is lost to browser tracking limits. On one store I audited, it recovered 45 percent of events and 51 percent of purchases from Safari tracking prevention, but 0 percent from ad blockers, because GTM still loaded from Google's domain. ### What it does Server-side tracking moves the conversion event off the visitor's browser and onto a server you control. The browser still fires a pixel. But a server endpoint, usually a server-side Google Tag Manager container, sends a second, matching event straight to Meta or Google from your own infrastructure. That second path is the point. Browser-only tracking loses data to ad blockers, iOS App Tracking Transparency, Safari's cookie limits, and slow page loads that fire before the pixel does. The server path doesn't run in the browser, so most of those losses never touch it. Here's what it is, what it recovered on stores I've measured, and when you need it. ### Browser-side vs server-side, side by side In a browser-only setup, the chain is short. A visitor buys, the Meta pixel fires in their browser, the event travels from their device to Meta. If anything between the page and Meta breaks, the conversion is gone and you never know it happened. In a server-side setup, two things fire for the same purchase: - The **browser event** fires as before, from the visitor's device. - The **server event** fires from your server, carrying the order ID, the value (the order subtotal after discounts, before tax and shipping, in one currency), email and phone (hashed), and the click IDs from the ad that drove the visit. Both events share a single event ID. The ad platform sees both, matches them on that ID, and counts the purchase once. This is the deduplication contract. Get the ID wrong and you either double-count or drop the server event entirely. On one store I audited, two Meta server builds sent purchases with mismatched event IDs, and Meta counted those orders twice for six months. ### Why the browser stopped being enough Three forces broke browser-only tracking, and none of them are reversing. **Ad blockers.** Blockers strip the Meta and Google pixels, and the GTM loader itself, before they fire. Those buyers convert. Your reporting never sees them. A server container only helps here if the loader comes from your own domain. On the store I mentioned above, the theme still loaded GTM from googletagmanager.com, the standard blocklists caught it, and ad-blocker recovery read 0 percent while the store paid for the upgrade that should have fixed it. **iOS and Safari.** Apple's App Tracking Transparency and Intelligent Tracking Prevention cut the cookie lifetime browsers will hold and limit what the pixel can read. On a high-ticket purchase the buyer researches for weeks, and a 7-day cookie window erases the link between the first ad click and the eventual sale. On the same store, the server path was recovering 45 percent of events and 51 percent of purchases from Safari's tracking prevention. A second store I work with showed the same pattern: the server path recovered a real share of the purchases tracking prevention had been dropping. **Page speed and timing.** Pixels fire late in the page load. A buyer who taps through to confirmation and closes the tab fast can leave before the browser event sends. The server event has no such race condition. The result is a gap. Meta reports fewer conversions than your Shopify admin shows. Your cost per purchase looks worse than it is. The algorithm optimizes against incomplete data, so it bids on the wrong people, and spend leaks. ### How a server-side setup is built The standard architecture in 2026 looks like this. 1. **Web GTM** sits on your site and captures the event in the browser, the same as today. 2. **A server-side GTM container** runs on a subdomain of your own domain, for example `gtm.yourstore.com`. Hosting it on your domain is what gets past blockers, because the request now looks like first-party traffic to your own site. That holds only when the GTM loader script comes from your domain too. 3. The server container receives the event and forwards it to **Meta's Conversions API** and **Google's enhanced conversions** endpoint, server to server. 4. Each forwarded event carries a shared **event ID** for deduplication and hashed customer data for matching. For most Shopify and lead-gen brands I run this through a hosted server container like Stape, which removes the need to manage your own cloud server. The platform setup matters less than getting the event ID, the customer-data hashing, and the dedup contract right. ### Server-side tracking isn't magic A few things it doesn't do, so you spend on it for the right reasons. It doesn't invent data the browser never had. If your pixel was firing on the wrong page or your purchase event was missing the order value, the server path will faithfully send the same broken data. Fix the event quality first. On that second store, the Shopify pixel feeding the server was subscribed only to checkout completed. Over ten days the server saw thousands of page views and a steady run of purchases, but almost no add-to-carts or product views. Once the missing events were wired in, the ad account's add-to-cart action, which had recorded almost nothing the month before, recorded hundreds over the next month. The quietest version I've found was on [Emma Grace Home](/results/unassigned-revenue-dual-ga4-fix/), a furniture store I designed. It had the full server stack: a server-side GTM container behind a first-party subdomain, with GA4, Meta CAPI and Microsoft Ads wired through it. Zero ecommerce events reached it, because the Shopify app that should have pushed them to the data layer had one setting unchecked. Once I checked it, events reached the server container within minutes. It doesn't replace the browser pixel. You run both. The browser event still does work the server can't, like building view-through audiences. They're partners. It doesn't fix attribution disagreements between platforms on its own. Meta, Google, and your Shopify admin count differently by design. Server-side narrows the gap by recovering lost events. It doesn't make three different measurement systems agree to the decimal. ### When you need it You need server-side tracking when: - You spend real money on Meta or Google and the platform's reported conversions sit well below your Shopify or CRM truth. - Your buyers research for days or weeks, so short browser cookie windows are dropping attribution. - A meaningful share of your audience is on Safari or iOS, or runs ad blockers. Check the browser split in GA4 before you decide. You can wait when you spend little, your conversion event is still broken at the browser level, or you have no one to maintain the container. A half-built server setup that double-counts is worse than a clean browser pixel. If your reported numbers and your real sales have stopped reconciling, that's the signal. Fix the event in the browser, then add the server path, then check that both sides deduplicate to one clean purchase. That's the order I rebuild in. The full architecture is in my [Tracking Stack reference](/frameworks/tracking-stack/), and the [free 25-page Setup Checklist](/audit/) covers the conversion settings you can check on your own account first. Server-side tagging also changes what a browser test can see: the browser only shows the hop to your own server. [The pre-consent browser test](/pre-consent-tracking/check-what-your-website-sends-before-consent/) covers what to ask for instead. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## What's the difference between Google Tag Manager and GA4? URL: https://connercrowe.com/conversion-tracking/gtm-vs-ga4-difference/ Published: 2026-06-28 | Updated: 2026-10-30 Answer: Google Tag Manager decides which tags fire on your site and when; GA4 is one analytics destination those tags report to. The break I find most is a duplicate install sending the same event twice. On one Shopify store I audited, the Google & YouTube channel injected four extra Google scripts, including a second copy of the GTM container. ### They aren't competing tools | | GTM | GA4 | |---|---|---| | What it is | A tag-management system | An analytics product | | Job | Decides what fires, when, and with what data attached | Collects events, sessions and conversions, then reports on them | | Role in the stack | The plumbing | A destination, the fixture the plumbing feeds | | Can run without the other | Yes, sending data only to other platforms | Yes, with the gtag.js snippet hardcoded | People line these two up as if you pick one. You don't. Google Tag Manager and GA4 sit at different layers of the same stack and do different jobs. Google Tag Manager (GTM) is a tag-management system. It's a single container snippet you place on your site, and from then on you add, edit, and fire tracking tags from the GTM interface instead of editing site code. Those tags can be a GA4 tag, a Meta Pixel, a Google Ads conversion tag, a TikTok pixel, a server-side endpoint, or a custom HTML script. GA4 (Google Analytics 4) is an analytics product. It's a destination. It collects events, sessions, and conversions, then gives you reports on traffic, behavior, and attribution. GA4 has to receive data from somewhere, and on most sites that somewhere is a GA4 tag fired by GTM. So the relationship is plumbing and fixture. GTM is the plumbing that decides what fires, when, and with what data attached. GA4 is one fixture the plumbing feeds. You can run GA4 without GTM by hardcoding the gtag.js snippet, and you can run GTM without GA4 by sending data only to other platforms. On a real ecommerce or lead-gen site you almost always run both. ### What each one does #### GTM: the firing layer GTM works on three primitives. - **Tags** are the code that runs: a GA4 event, a conversion pixel, a remarketing script. - **Triggers** decide when a tag fires: a page view, a button click, a form submission, a custom dataLayer event like `purchase`. - **Variables** pull in the values a tag needs: order value, transaction ID, page path, a click element's text. The dataLayer is the spine. It's a JavaScript object your site (or your platform) pushes structured data into, and GTM reads from it. A clean dataLayer with `purchase`, `value`, `currency`, and `transaction_id` is the difference between tracking that reconciles and tracking that doesn't. #### GA4: the reporting layer GA4 takes the events GTM sends and organizes them. Every interaction is an event. Some events get flagged as conversions (now called key events). GA4 builds sessions, assigns channels, runs attribution models, and lets you build explorations and audiences. It doesn't decide what fires on your site. It reports on what it receives. ### Why the distinction matters in practice The break I find most often on Shopify stores is a duplicate install: the Shopify Google & YouTube channel plus GTM, or two GA4 installs, sending the same event twice. Someone says GA4 is wrong, when GA4 is faithfully reporting what it received twice. On [one store I audited](/blog/bing-pixel-your-value-here/), the Google & YouTube channel had auto-injected four gtag scripts at the bottom of every page, and one of them was the store's own GTM container, which the theme already loaded. The container ran twice on every page, and two GA4 properties were collecting production data, so the property the team reported on undercounted. The same pattern shows up with two GA4 installs. On [Emma Grace Home](/results/unassigned-revenue-dual-ga4-fix/), a furniture store I designed, two GA4 installs fed one property, the web container loaded twice, and the purchase tag sent `Purchase` with a capital P, which GA4 ecommerce doesn't read as `purchase`. Remove the duplicate load before you fix anything that changes what counts, or the fix doubles too. The other GTM-side breaks are a tag firing before the dataLayer is populated, or firing with a missing `transaction_id`. In every case GA4 is reporting faithfully. The data going in is broken, and the break is in the firing layer. Three patterns to check for: - **Double-counting.** GA4 is installed once through the gtag.js snippet hardcoded on the theme or injected by the Shopify Google & YouTube channel, and a second time through a GA4 tag in GTM. Pageviews and revenue both inflate. The fix is to pick one delivery method, not to touch GA4 settings. - **Pageview-only tracking.** GTM fires the Google tag (called the GA4 Configuration tag until Google renamed it in September 2023) on every page but never sends ecommerce or lead events. GA4 looks alive, sessions count, and yet GA4 and Meta or Google Ads never reconcile because the conversion events were never wired. - **Race conditions.** The GA4 tag fires on the standard page-view trigger before the platform pushes purchase data into the dataLayer, so the value lands empty. The tag fired. The number is zero. If you can't tell which layer a number lives in, you can't debug it. That's the entire reason to keep these straight. ### How they fit with the rest of the stack GTM is where you also feed everything that isn't GA4. The Meta Pixel, Google Ads conversions, and server-side tagging through a tool like Stape all run as tags in the same container, reading the same dataLayer. That shared source is what gets your platforms to agree. When GA4, Meta, and Google Ads all read the same `purchase` event with the same value, they reconcile. When each one is wired separately by a different person at a different time, they drift. I put a number on that drift in the [Ecommerce Tracking Accuracy Benchmark](/benchmarks/ecommerce-tracking-accuracy/): one store carried three purchase tags installed by three different routes, and over the window where all three were live they disagreed by 30.1% on the same event. GA4 sits downstream as a reporting and audience source. You can pipe GA4 audiences and conversions back into Google Ads, which is useful, but that's GA4 acting as a data source, not as a tag manager. ### What to do Think of it this way. GTM is the question "what fires, when, with what data." GA4 is the question "what happened on my site, organized into reports." One is delivery. One is destination. If you're setting up tracking from scratch, install GTM first, build a clean dataLayer, then add a GA4 tag inside GTM along with your ad-platform tags. That order keeps every platform reading one source. If your numbers already disagree across platforms, don't start in GA4. Start in GTM and the dataLayer, because that's where the firing logic lives, and firing logic is where duplicate installs and empty values start. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## Why don't my GA4 and Meta conversion numbers match? URL: https://connercrowe.com/conversion-tracking/ga4-meta-conversion-numbers-dont-match/ Published: 2026-06-28 | Updated: 2026-10-30 Answer: GA4 and Meta won't match, because each credits the same orders under its own rules. I expect Meta 10 to 30 percent above GA4's paid-social count, and GA4 paid conversions within 15 percent of Google Ads plus Meta. Outside those bands, check tracking first. On one store I audited, two Conversions API builds double-sent Meta purchases for six months. ### They count different things on purpose GA4 and Meta both report conversions, but they answer different questions. Meta answers "how many of these orders can I credit back to an ad someone saw or clicked." GA4 answers "which clicks it saw should share credit for this key event," and by default it splits that credit with data-driven attribution. Its Traffic acquisition report works differently again: it files each session under its last non-direct click. Those are different definitions, so the totals won't line up. Expecting them to is the mistake, not the gap itself. I treat Shopify as the ledger. The order count inside the Shopify admin is the number no platform can overrule. Meta and GA4 are both attempting to attribute those orders, and each uses its own rules to do it. So the first move is never to ask which platform is right. It's to ask how far apart they're and whether that distance is explainable. The band I hold Meta to is 10 to 30 percent above GA4's paid-social conversions, wider on stores with a high iOS share. Inside that band, the numbers are doing their job. Outside it, something is broken. ### The three reasons the numbers diverge #### Attribution windows Meta's default is a 7-day click, 1-day view window. A person clicks your ad Monday, comes back through a Google search Thursday, and buys. Meta still claims that order inside its 7-day window. GA4 saw both visits. Its session report files the order under organic search, the last non-direct click, and its data-driven model gives Meta a share at most. Same order, two different owners. Multiply that across a few hundred orders and the totals drift apart with nothing actually wrong. View-through conversions widen the gap further. Meta counts a conversion when someone saw the ad and didn't click, then bought within a day. GA4 has no concept of that order being Meta's. So Meta will almost always report more conversions than GA4 credits to paid social. #### Modeling and the iOS gap Since iOS 14.5, Meta fills part of what it can't observe with statistical modeling. When the pixel and Conversions API can't tie an order to a person, Meta estimates it. Ads Manager doesn't show you which conversions were modeled, so you can't audit that share line by line. Read Meta conversions as directional, not as ledger truth. GA4 models too, mostly for visitors who decline consent when Consent Mode is on. Neither platform's number is the ledger. The store's order record is. #### Deduplication and double firing This is the one that signals a real problem. If your Meta pixel and your Conversions API both fire a purchase and they aren't deduplicated by a shared `event_id`, Meta counts the same order twice. The same thing happens in GA4 when a purchase event fires on both the thank-you page and a server-side tag without a `transaction_id` to dedupe against. On one store I audited, Meta's 2024 Conversions API auto-setup was still firing beside a newer server-side build. Both sent every purchase, the event IDs didn't match, and Meta counted each order twice for six months. Meta reliably drops the duplicate when the browser and server events share the same `event_id` and event name within 48 hours. When the platform total is higher than your Shopify order count for the period, double firing is the first thing to rule out. ### How to read the gap in 15 minutes | Relationship | What I expect | | --- | --- | | Meta vs GA4 paid-social conversions | Meta 10 to 30 percent above | | GA4 paid-channel conversions vs Google Ads plus Meta | Within 15 percent | | GA4 paid-channel conversions vs Shopify paid orders | Within 20 percent | Pull five numbers for the same thirty-day window and put them in one row of a spreadsheet: 1. Google Ads conversions 2. Meta conversions 3. GA4 paid-channel conversions, split into paid search and paid social 4. Shopify paid orders 5. Shopify revenue after discounts, before refunds Now check three relationships. Meta should run higher than GA4's paid-social number, inside that 10 to 30 percent band. GA4's paid-channel conversions should land within 15 percent of the Google Ads plus Meta sum, and within 20 percent of Shopify's paid orders. Compare revenue gross to gross. The value each platform receives should be the order subtotal after discounts, before tax and shipping, in one currency, with the order ID, and refunds and cancellations get handled separately. If all three hold, your tracking is healthy and the divergence is just the platforms doing their jobs differently. Stop there. If either relationship breaks, you have a tracking problem, not a performance problem, and chasing it through bid changes will waste budget. The two failure shapes are the platform reporting more than Shopify (double counting or undeduplicated events) and the platform reporting far less than Shopify (a tag that stopped firing, a consent banner blocking the pixel, or a checkout the tag never sees). ### When the gap is a real problem A few patterns tell me to stop reading and start auditing: - The platform total exceeds the Shopify order count for the same period. That's double counting, not strong performance. - The gap is widening week over week with no change in iOS share or channel mix. Something in the tracking stack drifted. - GA4 paid conversions collapsed after a theme update, an app install, or a consent-banner change. The tag is firing in fewer sessions than it used to. - Meta conversions are flat while Shopify orders climb. The Conversions API may have lost its match quality, and Meta is modeling blind. For a sense of what a sane scoreboard looks like, I put the platform claims next to the order record on one store I manage, August 17 to October 14, 2026. Google Ads claimed 0.83 purchases per online order and Meta claimed 0.07, so together they claimed fewer sales than the store made. That passes the first check. One in eight of Google's claims still came from a duplicate purchase action, which is why a believable total isn't proof. The [full comparison](/conversion-tracking/meta-or-performance-max-which-drove-the-sale/) is its own answer. It's one store, so read it as a worked example rather than a rate. The fix lives in the deduplication contract between the pixel, the Conversions API, and the GA4 tag, plus the reconciliation back to Shopify. My [Tracking Stack reference](/frameworks/tracking-stack/) covers both. My [free ads audit](/audit-request/) is where I check this against your real account, and the [free 25-page Setup Checklist](/audit/) covers the conversion settings you can check yourself first. ### The honest read GA4 and Meta not matching is the normal state of a healthy account. Two platforms, two attribution definitions, two amounts of modeling. The job isn't to force them into agreement. It's to know the size of the gap you should expect, watch it over time, and treat Shopify as the number that settles every argument. When the gap stays inside the band, leave it alone. When it breaks the band, fix the tracking before you touch a single bid. [The wasted ad spend hub](/wasted-ad-spend/) walks through the rest of the leaks. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## Why don't my Google Ads conversions match my Shopify orders? URL: https://connercrowe.com/conversion-tracking/google-ads-conversions-dont-match-shopify-orders/ Published: 2026-09-20 | Updated: 2026-10-30 Answer: Google Ads and Shopify count on different clocks: Google credits the click date, Shopify the order date. On one account in 2026, switching the date basis alone moved the count 5.8 percent. Over thirty days, a count gap inside 10 percent is reconciled, 10 to 25 percent needs a named cause, and above 25 percent is a tracking problem. The two numbers will never be equal, and an owner who expects them to be ends up chasing a gap that isn't a defect. The useful question is how wide the gap can be before it stops being attribution and starts being a broken event or a setting. This page holds the band I use for every pair, and the six causes I rule out when a gap sits outside it. ### Different clocks, same orders Google Ads credits a purchase to the day of the click that earned it, inside a click-through window of 1 to 90 days depending on the action, plus a view-through window. Shopify writes the order on the day the card was charged. Same order, two dates, and on a thirty-day report the edges never line up. That effect alone is measurable. In my [Ecommerce Tracking Accuracy Benchmark](/benchmarks/ecommerce-tracking-accuracy/), switching one account from click-date to conversion-date reporting moved its total by 5.8 percent on count and 7.4 percent on value, with nothing else changed. Then there's what happens after the sale. On the one account where I could read the store's order table, about a sixth of orders were refunded inside the window and about a tenth were cancelled, and the ad platform was never sent a restatement for any of them. Google keeps every one as a conversion. Shopify's net figures drop them. A gap is the normal state. This is the band I hold every account to, one row per pair. | Comparison | What's being compared | Window | Band | |---|---|---|---| | Google Ads purchase conversions vs Shopify orders attributed to paid search (GA4 last-non-direct, or Shopify Sales by traffic source) | Count. Google on click date, Shopify on order date, gross orders with refunds and cancellations left in | 30 days | Inside 10 percent, reconciled. 10 to 25 percent, name the cause. Above 25 percent, tracking problem | | Google Ads conversion value plus Meta purchase value vs Shopify revenue after discounts, before refunds, attributed to paid | Revenue | 30 days | Inside 15 percent, healthy. Above 15 percent, a setting or an event is padding the value | | GA4 paid-channel conversions vs Shopify paid orders | Count. GA4 sessionizes on last non-direct click | 30 days | Within 20 percent | | GA4 paid-channel conversions vs Google Ads plus Meta reported conversions | Count | 30 days | Within 15 percent | | Meta purchases vs Shopify orders | Count. Meta models the iOS share | 30 days | 10 to 40 percent, Meta higher, wider on high-iOS stores | | Meta purchases vs GA4 paid-social conversions | Count | 30 days | 10 to 30 percent, Meta higher | | CRM closed-won deals vs offline conversions imported to Google Ads and Meta | Count. Same records on both sides, by deal date | 7 days | Under 5 percent | The first row is the one this question is about. Between 10 and 25 percent, I want a named cause from the list below before anyone touches a bid. Compare gross to gross, Shopify's gross orders and gross revenue in the same window, so refunds don't open a gap on their own. Gross here means order subtotals after discounts and before refunds, not Shopify's Gross sales line, which is before discounts. ### The setting that pads revenue A Shopify client's $32.99 order landed in Google Ads as $148.95. The dataLayer was clean and GA4 said $32.99. The prior agency had switched on Customer Lifecycle Value Optimization with a hardcoded $115.96 incremental value for new customers, so every new-customer conversion arrived as revenue plus $115.96, and Smart Bidding had been chasing that number for nine months. The setting is two clicks deep under Conversions, then Summary, and it's the first thing I check when the count reconciles and the revenue doesn't. [The full receipt](/blog/google-ads-revenue-doesnt-match-ga4-shopify/) is on the blog. ### Add-to-cart value counted as revenue The column reads "Conv. value". Nothing in the interface says what fed it. On two of the three accounts in the benchmark, measured May to August 2026, between 16.6 and 17.2 percent of counted conversions and between 19.8 and 25.8 percent of reported conversion value came from add-to-cart and begin-checkout events, because those actions had been set to count inside the bidding metric. One Shopping campaign on one of those accounts had no purchase goal at all. In 90 days it reported 5.96 conversions, every one a cart or checkout event, and the algorithm bid to maximize cart value. In my Q3 account benchmark, a separate pull, 0 of 4 purchase-tracked accounts counted cart, page-view or engagement actions as primary, so this is a defect to rule out, not a rate. This is why the revenue band sits at 15 percent rather than 25. The smallest value defect I have measured was 19.8 percent, and a wider band would call it normal. Open the conversion-action list and read the category of every action that counts. The name tells you nothing. ### Two purchase events, one order Three purchase-category conversion actions were live at once on one benchmark account: the store sales-channel tag, a tag-manager deployment, and an analytics-imported purchase event. Over their common live window they recorded 382.02, 321.59 and 266.99 purchases for the same store, a 30.1 percent spread between the highest and the lowest. The account counted only one of them. The other two existed to disagree with it. The same defect shows up on Meta when the browser pixel and the Conversions API both fire a purchase without a shared event_id, which counts one order twice and inflates reported ROAS with it. Inside Google Ads the tell is a conversion rate far above the account's own history on a high-volume campaign, which usually means branded overlap or a de-duplication failure. The dedup contract, one event ID shared by browser and server, is in the [Tracking Stack](/frameworks/tracking-stack/). ### The purchase event that went dark On one store I work with, Google Ads saw about 68 percent of the purchases GA4 recorded. After a checkout migration, the purchase event no longer fired for every completed order, so part of the store's sales never reached the Google Ads tag. Once I repaired that event and tested it from click to confirmation, the two numbers came within 4 percent of each other inside two weeks. A gap that opens on a specific date usually has a specific cause, and a checkout change is the most common one I find. [What Checkout Extensibility breaks, and the order to fix it in](/conversion-tracking/shopify-checkout-extensibility-tracking/), is its own answer. ### Revenue that landed in Unassigned On [Emma Grace Home](/results/unassigned-revenue-dual-ga4-fix/), a furniture store I designed and run marketing for, GA4 was crediting paid revenue to Unassigned instead of a channel. The cause was a duplicate install, the most common Shopify break I find: two GA4 installs on one property, and the web container loading twice. The purchase event was also named with a capital P, and GA4 only treats lowercase `purchase` as a purchase. When GA4 can't resolve a session to a channel, Shopify's paid orders and Google's conversions each look fine on their own and refuse to meet in the middle. Count your installs before you touch an attribution setting. ### The five-number reconciliation I run weekly Fifteen minutes in a spreadsheet, five numbers, same thirty-day window: Google Ads conversions, Meta conversions, GA4 paid-channel conversions, Shopify paid orders, Shopify revenue after discounts, before refunds. I read them against the table above in a fixed order. 1. Google Ads conversions against Shopify paid orders. Inside 10 percent, move on. Outside 25, stop here. 2. GA4 paid conversions against Shopify paid orders, within 20 percent, and against the Google Ads plus Meta sum, within 15 percent. A miss on either points at a channel-mapping or session problem, the kind two GA4 installs produce. 3. Platform conversion value against Shopify paid revenue after discounts, before refunds, within 15 percent. If the count band holds and the revenue band doesn't, the cause is a value setting: lifecycle padding or cart value counted as revenue. If neither holds, an event is missing or doubled. Either way, no bid, budget, or structure change goes in until the count gap is back inside 10 percent, because every one of those decisions is computed from the column that's wrong. A gap that holds steady for a month is attribution. One that jumps in a single week is a broken tag with a date on it. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## Why is my Shopify conversion tracking not working? URL: https://connercrowe.com/conversion-tracking/shopify-conversion-tracking-not-working/ Published: 2026-06-28 | Updated: 2026-10-22 Answer: Shopify tracking usually breaks at one of four points: a Customer Events pixel that only hears checkout_completed, event names your GTM triggers miss, purchases sent twice without a shared event ID, or blocked third-party pixels. Place one test order: zero hits means a missing event, two means duplicates. On one store, purchases looked normal while add-to-carts barely registered. When a Shopify store owner tells me conversion tracking is broken, they mean one of two things. Either the numbers in Meta and GA4 don't match Shopify, or the numbers are missing entirely. Both trace back to a short list of failure points. I check them in order. ### Broken, or just different? Before you call it broken, put a number on the gap between what the ad platforms report and what Shopify recorded. | Gap, platform purchases against Shopify paid orders, 30 days | What it usually means | What I check | |---|---|---| | Inside 10 percent | Reconciled. The rest is click date against order date | Nothing, move on | | 10 to 25 percent | Something specific, still unnamed | Attribution window, counting setting, refunds | | Above 25 percent, platform lower | An event is missing or blocked | The Customer Events pixel, first-party loading, consent | | Above 25 percent, platform higher | Purchases are counted twice | Duplicate installs, missing event_id dedup | ### Start with funnel event volumes, not the purchase The fastest tell isn't the purchase event. It's everything before it. Pull event volumes for the last 10 days, from GA4 or your server container, and put `view_item`, `add_to_cart`, and `begin_checkout` beside PageView and purchase. You're looking for an event that sits at a tiny fraction of page views while purchases look normal. On one store I audited, ten days of server logs showed thousands of page views and a steady run of purchases, but only a handful of product views and add-to-carts. Its custom pixel listened for the completed checkout and nothing else. Once the pixel subscribed to the full funnel, the ad account's add-to-cart action went from almost nothing the month before to hundreds in the first full month. An event that low is broken, not underperforming. No amount of bid tuning fixes it, because the platforms have nothing to optimize toward. That's the problem to fix first. ### The four causes I check #### 1. The Customer Events pixel is missing or incomplete Shopify moved checkout to a sandboxed environment. The old approach of dropping pixel snippets into theme.liquid doesn't capture checkout anymore. Tracking now lives under Settings, then Customer events. Pixels installed by a channel app show up there on their own, and your own code goes in as a custom pixel on the same screen. The common breakage: the pixel exists but subscribes only to `checkout_completed`. That gives you a purchase number and nothing else. View item, add to cart, and begin checkout never fire, so your funnel goes dark before checkout. Subscribe to the full event set, not just the purchase. #### 2. Event names don't match your GTM triggers This one is quiet and expensive. The Customer Events pixel emits an event, but the event name doesn't match what your GTM Web Container trigger is listening for. The trigger never fires. The event never reaches GA4 or your server container. The harsher version is a published event with no subscriber at all. A theme file can call `Shopify.analytics.publish()` and Shopify accepts it whether or not any pixel is listening, so the producer looks finished while nothing consumes it. I walk through [finding that one in my own build](/blog/shopify-custom-event-no-subscriber-google-ads-conversion/), including why the fix was to take the conversion out of the sandbox rather than write a subscriber for it. Open your GTM Web Container, list your triggers, and compare each trigger name against the exact strings your pixel pushes to the dataLayer. A mismatch as small as `add_to_cart` versus `addToCart` kills the whole chain silently. Nothing errors. The event just disappears. Case counts too. On [Emma Grace Home](/results/unassigned-revenue-dual-ga4-fix/), a furniture store I designed, the purchase tag sent `Purchase` with a capital P. GA4 ecommerce binds to lowercase `purchase`, so that one character kept purchases out of GA4's ecommerce reports even when the event fired. #### 3. Purchases are sent twice The break I find most often on Shopify is a duplicate install: the Shopify Google & YouTube channel plus GTM, or two GA4 installs, firing the same event twice. On [one store I audited](/blog/bing-pixel-your-value-here/), the Google & YouTube channel had injected a second copy of the GTM container, so every dataLayer event fired into two instances. Meta has its own version. If you run a Meta browser pixel and a server-side CAPI feed at the same time without a shared dedup key, Meta counts each purchase twice. On that same store, a legacy CAPI auto-setup and a later server-side build both sent every purchase, with event IDs that didn't match, for six months. The store owner sees great ROAS in Meta, terrible reconciliation against Shopify, and can't trust either number. The fix is a stable key shared between the browser event and the server event. Meta deduplicates when the event name matches and the pixel's `eventID` equals the server's `event_id`. Without a matching ID, Meta falls back to `fbp` or `external_id`, which generally only works when the browser event arrives before the server one. Set `event_id` explicitly (the order ID works for Purchase) and pass the identical value from both the browser pixel and the server tag. For duplicate installs, the fix is to remove one delivery route, not to dedupe two. #### 4. Pixels load third-party and get blocked If your Meta Pixel still loads from `connect.facebook.net` and your GA4 still loads from Google's domains, ad blockers and browser privacy defaults drop part of your traffic before any event fires. On the store with the duplicate container, the server path was recovering 45% of events and 51% of purchases from Safari's tracking prevention, and adblock recovery read 0% because the theme still loaded GTM from `googletagmanager.com` instead of the paid first-party loader. Check what's loading on your storefront. GTM and GA4 should come from a first-party subdomain like `load.gtm.yourbrand.com`. If they're still loading third-party, you're leaving recoverable conversions on the table. First-party loading for GTM and GA4 is the higher-priority fix here. Routing the Meta Pixel server-side is a separate, paid step and I only recommend it when Meta is a large line item. ### Consent can also zero out your data Before you tear apart the pixel, check consent. Pull your GA4 requests by consent choice. If a large share reads "Not set," Consent Mode may be failing at the source. The direct test is to accept your own banner, reload as a returning visitor, and read the `gcs` parameter on an outbound Google request. On [one store I audited](/results/consent-mode-denied-for-consenting-users/), visitors had said yes and the beacons still carried `gcs=G1--`. One caution. A high "Not set" share can be old data from before a consent platform was installed, not proof that consent is broken now. Verify what's actually deployed before you call it missing. Look in the storefront for an active consent platform and confirm it fires before GTM, not after. ### The diagnosis order I follow 1. Funnel event volumes against PageView. This tells you if the problem is broken events or just attribution noise. 2. The Customer Events pixel. Confirm it exists and subscribes to the full funnel, not only `checkout_completed`. 3. Event name matching between the pixel and your GTM triggers. 4. Deduplication keys between browser and server. 5. First-party versus third-party script loading. 6. Consent Mode coverage. I start at steps 1 and 2 because a broken event looks like a quiet one. Across 33 Google Ads accounts I manage, in the 30 days to October 21, 2026, 16 had at least one biddable website conversion action that recorded zero conversions. Some of those are low-volume, and some are broken without anyone knowing. A broken action rarely announces itself: on [one contractor account](/results/dead-form-conversion-fix/), renaming a form button from Send to Submit broke the click-text trigger behind the only website conversion, and it sat Inactive with nobody told. In the funnel case above, the owner would have looked at the purchase pixel, and the purchase was the one event still firing. The damage is upstream, in the events the ad platforms need to find buyers in the first place. One more check belongs on that list, and it sits above all six: confirm the actions your account counts as conversions are purchases. In my [Ecommerce Tracking Accuracy Benchmark](/benchmarks/ecommerce-tracking-accuracy/), covering May to August 2026, two of three accounts counted add-to-cart events inside the Conversions column, at 16.6% to 17.2% of the total, and one campaign had no purchase goal on it at all. Tracking that fires correctly into the wrong definition still reads as broken from the outside. If you have run through this list and the numbers still don't reconcile, the next layer is server-side observability. Turn on outgoing request logging in your server container so you can see exactly what's being forwarded to Meta, Google Ads, and GA4. You can't fix what you can't watch. The full architecture is in my [Tracking Stack reference](/frameworks/tracking-stack/). Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ --- # Answers: Lead Quality Hub: https://connercrowe.com/lead-quality/ ## Do I need a CRM to run Google Ads for a service business? URL: https://connercrowe.com/lead-quality/do-i-need-a-crm-to-run-google-ads-service-business/ Published: 2026-06-29 | Updated: 2026-10-21 Answer: Yes, though the first place you mark leads qualified can be a call-tracking platform like WhatConverts. In 33 Google Ads accounts I manage, 30 days to October 21, 2026, only 6 of 28 lead-gen accounts bid on an offline import. Start with a booked appointment or WhatConverts-qualified call as the one primary conversion, then import closed deals from a CRM. ### The answer in brief You can launch Google Ads without a CRM. You can't run it well without one. The reason is mechanical. A service business converts to a call or a form fill, not a checkout. That lead might close in three days or three months, and the deal might be worth $500 or $50,000. Google Ads has no idea which is which. The only system that knows is your CRM. The first fix doesn't need the CRM at all: make a booked appointment or a qualified call the primary conversion, so Google stops counting junk fills. Once closed deals live in a CRM, a later and optional layer sends each closed-won deal back to Google Ads, tied to the original click, so bidding can lean toward the leads that turn into work. Skip both, and the algorithm finds the cheapest leads it can, which are almost always the worst ones. The closed-won import is the rare part. Across 33 Google Ads accounts I manage, 30 days to October 21, 2026, 27 of 28 lead-gen accounts bid on calls and 18 of 28 had enhanced conversions for leads on. Only 6 of 28 bid on an offline import, the import that tells Google which leads turned into work. ### What the CRM has to expose A CRM earns its place in this stack only if it does three things. Plenty of CRMs are installed and still fail all three. - **Lead source on every record.** The gclid (Google's click ID) and any UTM parameters have to land on the contact when the form submits or the call connects. If your CRM has a clean pipeline but no idea where each lead came from, the offline conversion has nothing to match against. This is the most common break I find. - **Stage transitions with timestamps.** Lead to MQL to SQL to closed-won, each with a date. Timestamps are how you measure time-to-close and how you confirm a deal is real before you import it. - **Closed-won exposed via webhook or API.** When a deal flips to closed-won, the CRM has to fire that event outward. A win locked inside a dashboard nobody can read programmatically is a win Google Ads will never see. Salesforce, Pipedrive and GoHighLevel all do this. The brand matters less than the wiring. For most of my lead-gen clients, the qualification lives in WhatConverts, which tracks the calls and forms and is where a lead gets marked qualified. A CRM with a beautiful pipeline and no lead-source field is worse than a spreadsheet, because it looks finished. ### What to do before you have a full CRM At low volume, you don't need to buy Salesforce to start. You need one place where every lead's source is captured and you can mark which ones closed. A call-tracking and lead platform like WhatConverts can be that place before you own a CRM. It records each call or form with its source, and someone on your team marks the lead qualified when it books. A spreadsheet holds up in the early stage if the gclid makes it onto each row. The form passes the gclid, your booking tool stores it, and you update a "closed" column by hand. Once a week you upload those wins to Google Ads as an offline import, and enhanced conversions match the hashed email or phone back to the original click. Google began combining the web and leads settings for enhanced conversions into one in 2026 ([Google's notice](https://support.google.com/google-ads/answer/16884284)). That's the same closed-won signal a full CRM sends, run by hand. It holds as long as one person can keep every row current. The moment that stops being true, the manual version breaks. People forget to log closes. The gclid gets dropped on one form. You lose track of which uploads already happened. That's the threshold where a real CRM with a closed-won webhook pays for itself. The pipeline behind it is small once the data is clean. The closed-won webhook I build runs on a Cloudflare Worker and is about 150 lines of TypeScript including auth, error handling, and the reconciliation query. New builds post to Google's Data Manager API. Since 15 June 2026 the Google Ads API only takes offline conversion uploads from developer tokens that were already sending them. ### The cost of skipping it Run Google Ads with no closed-won feedback and here's what happens. You set the primary conversion to "form submitted." Smart Bidding optimizes toward more of those, as instructed. It learns which keywords, times, and audiences produce the most form-fills per dollar. Those are usually the tire-kickers, the wrong-service searches, the people who fill out anything. Your lead count climbs. Your cost-per-lead drops. Your sales team gets quieter. The platform did what you told it. You told it the wrong thing, because the CRM that knew better was never connected. It's common. In my [Q3 2026 benchmark](/benchmarks/google-ads-account-benchmarks/), 6 of 29 lead-gen accounts counted a non-lead action or a click proxy as a primary conversion. Fixing the signal doesn't have to wait for a full closed-won import. On one fence contractor's account, the only website conversion was a click proxy. It fired when someone pressed a button labeled "Send," whether or not the form went through, and it went dead when the button was renamed "Submit." I moved the conversion onto the form plugin's own success event, which fires only after a submission sends, and it recorded its first real lead shortly after. The whole fix lived in Tag Manager ([the dead form case study](/results/dead-form-conversion-fix/)). The CRM is layer one of the [Lead Quality Stack](/frameworks/lead-quality-stack/) for a reason. Everything downstream, the call tracking, the server-side container, the offline imports, depends on a source of truth that knows which leads turned into money. The [free 25-page Google Ads Setup Checklist](/audit/) covers the conversion settings that decide what Google counts as a lead, and [the free ads audit](/audit-request/) reads your real account, back in writing within 24 hours of read-only access. Worked example: Greenacre Law. Every qualified call imported at $0, so a $2,500 review looked like a $15,000 lawsuit. I scored 236 recorded paid calls in the first live pass, and a backfill priced 132 qualified leads by matter type with the firm’s own numbers. https://connercrowe.com/results/law-firm-call-scoring-by-matter-type/ ## How do I import closed-won deals into Google Ads as offline conversions? URL: https://connercrowe.com/lead-quality/import-closed-won-deals-google-ads-offline-conversions/ Published: 2026-06-29 | Updated: 2026-10-21 Answer: Store the click ID (GCLID, GBRAID or WBRAID) on the CRM record, and when the deal closes, upload it with close time and deal value to its own offline conversion action within 90 days of the click. New builds use Data Manager. Of 28 lead-gen Google Ads accounts I manage, 6 bid on one (30 days to October 21, 2026). ### The mechanism in three moves Offline conversion imports close the loop between a Google Ads click and a deal that closes weeks later. There are three moving parts. First, capture the click ID. Google appends a `gclid` parameter to most ad clicks, and on some iOS traffic it sends a `gbraid` or `wbraid` instead. A small script reads whichever one arrives from the URL and writes it into its own hidden field on your form. Google's [Data Manager API accepts](https://developers.google.com/data-manager/api/devguides/events/google-ads/offline/send-events) a `gclid`, a `gbraid` or a `wbraid` as the ad identifier, so store all three fields rather than one. When the form submits, the ID rides into the CRM on the new contact or deal record. Second, store it. The click ID lives on the CRM record as a custom property next to the lead. It sits there through the whole pipeline: lead, booked appointment, proposal, closed-won. It has to survive every stage transition without getting wiped, and it has a shelf life. Google's [GCLID import guide](https://support.google.com/google-ads/answer/7012522) requires a click-to-conversion cycle under 90 days or under 14 days, depending on the data source. Some connected sources, such as BigQuery, import only the last 14 days on each run. If your deals routinely close later than that, upload an earlier stage you trust, like the booked consult or estimate, as its own conversion action, and send closed-won only on the deals that land inside the window. Third, push the win back. When the deal moves to closed-won, you send an offline conversion to Google Ads carrying that original click ID plus the conversion time and the deal value, against a conversion action set up for closed-won deals only. Google matches the ID to the click that started it and credits the campaign, ad group, and keyword that produced the actual customer. That's the loop. The CRM is the ledger, and the win flows back to the ad platform where the bidding happens. ### Why this changes how Smart Bidding behaves Without the loop, Smart Bidding optimizes against form-fills. The algorithm does what you tell it to. It finds the cheapest conversions. The cheapest form-fills are usually the worst leads: price shoppers and the contact who filled out four firms' forms at once. The algorithm has no idea those leads never close, so it buys more of them. Feed closed-won deals back in with their dollar value, move bidding to a value-based strategy, and the target changes. Now the algorithm sees which clicks turned into signed clients and which dollar amounts they carried. It bids toward revenue. The expensive keyword that produces a $40,000 matter starts winning the auction over the cheap keyword that produces 30 junk fills. Few accounts are set up this way. Across 33 Google Ads accounts I manage, 30 days to October 21, 2026, 6 of the 28 lead-gen accounts bid on an offline import, while 27 of 28 bid on calls. The import also isn't the first fix. A law firm I worked with shows the order. I fixed the conversion signal and rebuilt the account: WhatConverts-qualified calls became the conversion, forms got required-field validation so a half-filled form stopped counting, and the account was rebuilt by case type with filtering ad copy and 300+ negatives. In 90 days spend fell 50%, qualified signups doubled, and cost per qualified lead fell by about half. There was no signed-case import behind that. Google Ads never saw a signed case, every qualified call still imported at $0, and I didn't isolate which change did what. On call-first accounts a click-ID import has a second limit: only a small share of calls carry a click ID. Match the rest by the caller's number and call time with enhanced conversions instead. ### Three ways to build it | Method | How it sends | The trade | |---|---|---| | Zapier or similar connector | Fires on closed-won, filters out records without a click ID, runs the Send Offline Conversion action | You pay per task, and it can break quietly on a plan change or an auth expiry | | Call-tracking platform | WhatConverts can send leads once they are marked quotable or given a quote or sales value | Best when calls are the lead | | Custom webhook plus Cloudflare Worker | The CRM fires a webhook and the Worker formats and posts the conversion. New builds go to the Data Manager API | You own the code, and a failed upload is yours to catch | | Spreadsheet upload | A manual offline conversion file | Tedious and it lags, but it proves the loop works | Pick the method that matches your stack and your tolerance for maintenance. **Zapier or a similar connector.** A Zap fires when a deal hits closed-won, filters out records without a click ID, and runs Zapier's [Google Ads Send Offline Conversion action](https://help.zapier.com/hc/en-us/articles/8496054576909-How-to-set-up-Google-Ads-Send-Offline-Conversion-action) with the GCLID and deal value. That action needs a GCLID and an "import from clicks" conversion action. No code to maintain. The trade is that you pay per task and add a dependency that can break quietly on a plan change or an auth expiry. **Your call-tracking platform.** When calls are the lead, the platform that tracks them can often do the sending. Most of my lead-gen clients run WhatConverts for call and form tracking and qualification, and its Google Ads integration [can send leads once they're marked quotable](https://www.whatconverts.com/help/docs/integrations/google-ads-integration/understand-the-google-ads-integration/) or get a quote or sales value, with that value attached. The CRM's job is to tell it which leads qualified and which closed. **Custom webhook plus a Cloudflare Worker.** A CRM workflow fires a webhook when the deal closes. A Cloudflare Worker receives it, formats the offline conversion, and posts it to Google. New builds go to the Data Manager API. From June 15, 2026, the Google Ads API [stopped accepting](https://support.google.com/google-ads/answer/16884284) offline conversion and enhanced-conversions-for-leads uploads from developer tokens that hadn't already been sending them. Tokens that were sending them were allowlisted and keep API access for now, though Google recommends moving them to the Data Manager API too. You own the code, it costs near nothing to run, and you control the exact payload and timing. The trade is that a failed upload is yours to catch. **A spreadsheet upload, to start.** Before any automation, you can export closed-won deals with their click IDs and upload them to Google Ads as a manual offline conversion file. It's tedious and it lags, but it proves the loop works and validates that your click IDs actually matched clicks before you spend time on a webhook. ### Where this sits in the larger stack Importing closed-won deals is layer eight of the [Lead Quality Stack](/frameworks/lead-quality-stack/), the loop that's unique to lead generation. It only works if the layers under it are wired. The click ID has to land on the form and survive into the CRM. The CRM has to expose closed-won through a webhook or API. Your primary conversion in Google Ads should already be a qualified lead, which on my accounts means a booked appointment, a booked consult or estimate, not a raw form-fill. The offline import then sharpens a signal that's already pointed the right way. When I audit this, I check two things in order. First, is the primary conversion a qualified lead, a booked appointment or a WhatConverts-qualified call, rather than a form fill? Second, where your CRM holds closed deals, are they reaching Google Ads with the click ID and a value? If the first is wrong, you're optimizing toward the wrong event before the imports even matter. If you want the click-ID path and the offline import checked against your own account, the [free 25-page Google Ads Setup Checklist](/audit/) covers the conversion settings around it, with no email gate, and [the free ads audit](/audit-request/) reads the whole conversion setup on your account, back within 24 hours of read-only access. Worked example: Greenacre Law. Every qualified call imported at $0, so a $2,500 review looked like a $15,000 lawsuit. I scored 236 recorded paid calls in the first live pass, and a backfill priced 132 qualified leads by matter type with the firm’s own numbers. https://connercrowe.com/results/law-firm-call-scoring-by-matter-type/ ## How do I stop spam form submissions from inflating my Google Ads conversions? URL: https://connercrowe.com/lead-quality/stop-spam-form-submissions-inflating-google-ads-conversions/ Published: 2026-06-29 | Updated: 2026-10-21 Answer: Count the lead only after your server accepts it, not on the submit click, and set the action to one conversion per click. Across 33 Google Ads accounts I manage, 30 days to October 21, 2026, 3 of 28 lead-gen accounts counted a biddable lead action as every conversion. Then bid on booked appointments, not raw fills. ### Why the spam counts in the first place Almost every spam-inflation problem I diagnose comes down to one wiring mistake. The conversion fires on the click of the submit button, not on a real submission. A bot lands on the page, fills the fields, clicks submit, and the GTM trigger registers a form submission the instant that button is pressed. It never checks whether the form passed validation. It never checks whether the server accepted the lead. The click is the event, so the bot gets counted. Google Ads then sees that conversion. Smart Bidding sees it too. Because the algorithm optimizes toward whatever you call a conversion, it starts buying more of the traffic that produces those cheap form-fills. On accounts I take over, more than 30% of the form leads being counted are typically spam or junk. That's my experience on takeovers, not a measured benchmark. The spam isn't a reporting nuisance. It's steering your budget toward worse traffic. The tag doesn't even need a submit click to count junk. One home-services account in my [home-services benchmark](/home-services-paid-search-benchmarks/) reported a 114% conversion rate because it counted a 50% page scroll and starting to type in a form as conversions. It logged more conversions than clicks and no completed leads. There are two fixes, and you need both. One lives in the browser tag layer. One lives in how you define a conversion. I'll take them in order. ### Fix the trigger: gate the event on server confirmation The browser tag container (GTM web) is a traffic cop, not the form processor. Its job is to fire the conversion event when a real lead has been accepted, and only then. That means the trigger can't watch the submit button. It has to watch for proof that the submission succeeded. Concretely: - **Stop using the submit-click or generic Form Submission trigger.** Both fire before validation. Replace them with an event that only exists after the server has accepted the lead. - **Fire the conversion on the server-confirmed thank-you state.** The cleanest signal is a redirect to a thank-you URL the server serves only after it has validated and stored the lead, or a custom dataLayer event your backend pushes on a successful response. Either way, the trigger reacts to server confirmation, not a button press. - **Add a honeypot field and check it server-side.** A hidden field that humans never see but bots fill in. If it has a value, the server rejects the submission and never pushes the success event. No event, no conversion. It catches the simple bots that fill every field, at no cost. - **Add reCAPTCHA v3 and read the score on the server.** v3 shows no checkbox. It returns a score from 0.0 to 1.0 for each request. Pass the token to your backend, verify it there, and push the success event only above your threshold. Low-score submissions never become a conversion. reCAPTCHA keys are now managed in a Google Cloud project, and Google's migration moves existing v2 and v3 keys over without code changes. The principle under all four is the same. The conversion event belongs downstream of the validation, not in front of it. If a submission can reach Google Ads without your server confirming it's a real lead, your trigger is in the wrong place. I've seen what the click version costs. On one fence contractor's account, the only website conversion fired on a click whose button text read "Send." A click trigger like that counts button presses, so a submission that failed validation or got caught by the spam filter still counted as a lead. When someone renamed the button to "Submit," it stopped counting anything at all. I rebuilt it on the form plugin's own success event, which fires only after the mail is sent. The [dead form case](/results/dead-form-conversion-fix/) walks through the fix. One more setting sits outside the tag. Open the conversion action and check how it counts. A lead action should count one conversion per click. Set to every conversion, one person who submits twice counts as two leads, and one bot that submits ten times counts as ten. Across 33 Google Ads accounts I manage, 30 days to October 21, 2026, 3 of 28 lead-gen accounts had a biddable lead action set to every conversion. ### Fix the definition: one primary conversion, and it's a qualified lead Cleaning up the trigger stops the obvious bot fills. It doesn't stop the low-quality humans, the out-of-area inquiries, the wrong-service requests. For that you change what you let Google Ads optimize toward. In a service business, your one primary conversion should be a **qualified lead, not a raw form-fill.** For me that means a booked appointment: a booked consult or a booked estimate. A form-fill is an action. A booked appointment is something your CRM or lead platform records after a person on your team has talked to the lead. When the qualified lead is the primary conversion, spam never reaches the signal Smart Bidding learns from, because it never gets qualified. Gating spam is half the fix. Whether to bid on form fills or on qualified leads at all is its own decision, covered in [form fills or qualified leads](/lead-quality/optimize-google-ads-form-fills-or-qualified-leads/). On one law firm account I took over, every form submit counted as a conversion, spam and bot traffic included. Fixing the conversion signal and rebuilding the account cut spend in half and doubled qualified signups in 90 days. That works like this: 1. The form-fill lands in your CRM with its gclid attached, captured at the click. 2. Qualification happens in the CRM or lead platform. A lead that's spam, out of area, or out of scope never books, so it never advances. 3. Only the booked record is imported back into Google Ads as the conversion, with the gclid and the hashed email or phone so enhanced conversions can make the match. Google began combining the web and leads settings for enhanced conversions into one in 2026 ([Google's notice](https://support.google.com/google-ads/answer/16884284)). Now your bidding optimizes against leads a human confirmed were real. Keep the raw form-fill as a secondary, observe-only conversion if you want a read on volume, but keep it out of the bid strategy. Run Smart Bidding as tCPA on the booked-appointment conversion, or value-based bidding once qualified values flow. Never run Maximize Conversions against raw fills, or you reward the algorithm for finding more of the junk you're trying to escape. ### The order to do it in Patch the trigger first. It's fast and it stops the bleeding the same day. Then build the qualified-lead definition, because that's what protects you long term. The trigger fix removes bots. The definition fix removes everything a human would call a bad lead. Together they mean the number in your Google Ads account starts matching the number of real opportunities your team works. This sits inside the broader architecture in my [Lead Quality Stack](/frameworks/lead-quality-stack/), where the browser tag layer (L02) and the qualified-lead conversion (L06) are two of eight layers that have to agree. To find where your own form tracking is leaking before you touch anything, my [free 25-page Google Ads Setup Checklist](/audit/) walks the conversion setup you can check on your own account, and [the free ads audit](/audit-request/) reads the whole conversion setup end to end. Worked example: Greenacre Law. Every qualified call imported at $0, so a $2,500 review looked like a $15,000 lawsuit. I scored 236 recorded paid calls in the first live pass, and a backfill priced 132 qualified leads by matter type with the firm’s own numbers. https://connercrowe.com/results/law-firm-call-scoring-by-matter-type/ ## How do I track phone calls as conversions in Google Ads? URL: https://connercrowe.com/lead-quality/track-phone-calls-conversions-google-ads/ Published: 2026-06-29 | Updated: 2026-10-30 Answer: Turn on call reporting so Google's forwarding numbers count calls from your ads and website, then run a call-tracking platform like WhatConverts to record, score and send only qualified calls back as the conversion you bid on. Across 20 lead-gen accounts I manage, the typical one had 46.1% of ad calls run under 60 seconds. ### What Google's built-in call tracking does, and where it stops Google Ads ships with call conversions through Google forwarding numbers, and call reporting has to be on in the account settings for them to work. A call from a call asset or a call-only ad counts as a conversion once it runs past a minimum call length you set on the conversion action. The default is 60 seconds. It reaches further than many people assume. With [website call conversions](https://support.google.com/google-ads/answer/6100664), a Google forwarding number dynamically replaces the number on your site for visitors who clicked an ad. So the customer who clicks your ad, reads three pages, and then dials the number in your header is counted, as long as the Google tag and the call snippet are installed and verified. Two gaps remain. Google only sees callers who clicked an ad, so the customer who finds you in a Google Business Profile, an organic result, or an AI answer that pulled your number never reaches it. And duration is the only test, so a 90-second wrong number counts the same as a 90-second consult booking. The duration bar is where most accounts sit. Across 20 lead-gen Google Ads accounts I manage with 10 or more ad calls, 30 days to October 21, 2026 (1,265 calls from call assets and manual dials): - The typical account had 46.1% of its calls run under 60 seconds, missed calls included, with a range of 20% to 88%. - The typical account missed 4.0% of its calls. Pooling every answered call, the middle one ran 110 seconds. - The typical account had 19.5% of its answered calls end under 30 seconds. Across all 28 lead-gen accounts I manage in the same window, 40 of 47 biddable call conversion actions used Google's 60-second default, and 5 accounts bid on a call action that counts calls of 20 or 30 seconds. A shorter bar counts more hang-ups and wrong numbers as leads. A longer one still can't tell a booking from a long sales pitch. Duration sorts calls by length, not by whether they were leads. ### Dynamic number insertion across every source The fix is a dedicated call-tracking platform sitting in front of all your traffic. Most of my lead-gen clients run WhatConverts for call and form tracking and qualification. CallRail works the same way. The mechanism is dynamic number insertion. A snippet on your site swaps the displayed phone number based on how the visitor arrived, so a Google Ads click sees one number, an organic visit sees another, and a paid social click sees a third. That swap is what carries the click identifier through. When a visitor lands with a gclid, the swap ties that gclid to the number they dial. Now the phone call has a source. You can attribute it to the campaign, ad group, and keyword that produced it, the same way you would a form fill. Pools of numbers rotate per session so two visitors at once never collide. Without dynamic number insertion, every off-ad call lands in one undifferentiated bucket. With it, calls become as traceable as clicks. ### Record, transcribe, and score the call A tracked call is the start. The lead-quality question is whether the call was any good. WhatConverts and CallRail both record and transcribe calls, which lets you score them. Set a definition of a qualified call before you start. The conversion I set Smart Bidding to bid on is the booked appointment: a booked consult or estimate. For a law firm that's a potential client in your practice area and jurisdiction who books a consultation. For an HVAC company it's a service-area homeowner who books an estimate, not a vendor pitch or a job applicant. Raw calls stay secondary. Transcripts let you tag calls against that bar, by hand at first and with rules once you see the pattern. Scoring holds up when you test it. On a law firm I work with, the first live scoring pass read 236 calls and sorted them into 67 qualified, 110 not viable, and 59 follow-ups. Before trusting it, I checked it against calls the firm's own team had labeled by hand, and on clear rejections it matched 6 for 6. A hard rule allows one qualified conversion per caller per month, so a caller who rings five times doesn't count five times. This is the same logic the rest of the Lead Quality Stack runs on. The primary conversion should be a qualified lead, not a raw form-fill, and the same holds for calls. A pile of duration-only call conversions teaches Smart Bidding to chase volume. ### Send the qualified call back to Google Ads Once a call is scored qualified, that event has to reach Google Ads. In the setups I run, the call-tracking platform usually sends it, because that's where the call, the score and the value already live. Anything fired from the site goes through the server-side container, the same path the form conversions take. Don't count on a click ID for most calls. On that law firm, only about 15% of paid calls carried one. A call import keyed on click IDs alone would have missed most of the qualified calls. The phone-and-time import with enhanced matching, which matches the caller's number and the call time to the ad interaction, was the rail that carried them. Most call platforms also write the call into your CRM as a lead record, which is where the qualification stage lives. That keeps phone and form leads in one ledger instead of two systems that never reconcile. When this is wired, the qualified call becomes a primary conversion. Smart Bidding sees phone and form leads together, and tCPA has the data it needs. If you later send a value with each qualified call, Maximize Conversion Value can use it. On the law firm, the firm priced its 13 matter types in three tiers, $2,500, $9,000 and $15,000, as its own price list for scored calls. ### The side effect of the swap Once dynamic number insertion is live, the number rendered on your page is often not the one you own. That matters the moment someone reads a number off the page to reuse it, whether that's a developer rebuilding the site or you setting up a call asset. A rendered number is a lease, not a deed. Numbers released when a call-tracking account lapses get recycled, and Google says the same of its own: forwarding numbers "are property of Google, they can change or be reassigned." Checking it's less obvious than it looks. Reading the raw server HTML tells you what the page is swapping from. It doesn't tell you who owns the line. A source-level tracking number, the single static number CallRail assigns to one campaign, sits in that server HTML too and never changes between fetches. A `curl` fetch also runs no JavaScript, so it can't see a browser-side swap at all. Compare what a browser renders against what the server sent: if they disagree, something is swapping. Then confirm the underlying number against the business's own records before it goes into a rebuild or a listing. ### What the fix buys you Calls carry most lead-gen accounts. Across 33 Google Ads accounts I manage, 30 days to October 21, 2026, 27 of the 28 lead-gen accounts bid on calls. On those accounts the call conversion is part of the bidding signal, and on call-only accounts it's all of it, so a duration-only call action becomes the whole definition of a lead. My [Q3 2026 Google Ads benchmark](/benchmarks/google-ads-account-benchmarks/) is a separate dataset, classified on primary actions rather than biddable goals. It has more on how call-heavy accounts set their primary conversions. Get it right and the rest of the stack compounds. When a phone lead becomes closed-won in the CRM, that win can import back into Google Ads against the click ID or the caller's number and call time, so bidding optimizes toward calls that actually become customers. The call layer sits in the middle of the architecture, between your CRM and your ad platforms. See how it connects in the [Lead Quality Stack framework](/frameworks/lead-quality-stack/), or run the [free 25-page Google Ads Setup Checklist](/audit/) to see what your account is missing today. If you want it diagnosed on your own account first, the [free custom audit](/audit-request/) comes back within 24 hours of read-only access. Worked example: Greenacre Law. Every qualified call imported at $0, so a $2,500 review looked like a $15,000 lawsuit. I scored 236 recorded paid calls in the first live pass, and a backfill priced 132 qualified leads by matter type with the firm’s own numbers. https://connercrowe.com/results/law-firm-call-scoring-by-matter-type/ ## Should I optimize Google Ads for form fills or qualified leads? URL: https://connercrowe.com/lead-quality/optimize-google-ads-form-fills-or-qualified-leads/ Published: 2026-06-29 | Updated: 2026-10-21 Answer: Qualified leads. In a 14-day test on one service account, AI Max produced 41% more form fills and 43% fewer qualified leads, and cost per qualified lead rose 76%. Make the booked appointment, a booked consult or estimate, the one primary conversion, keep raw fills and calls secondary, and bid Target CPA on it. ### The short answer Qualified leads. Not form fills. The conversion you mark as primary is the target Smart Bidding aims at, and the algorithm is good at hitting whatever you point it at. Point it at raw form fills and it finds the cheapest possible form fill. The cheapest form fill is almost always the worst lead. The tire-kicker, the wrong service area, the person who will never book, the bot that slipped past your form validation. This is the signal swap, the S pass in my CLOSE Audit. In my Lead Quality Stack it sits in the Google Ads layer (L06), whose rule is one primary conversion, a qualified lead, with everything else demoted to secondary. A form-fill primary conversion is bad data dressed up as a result. ### Why a form-fill primary teaches the algorithm to find junk | Change under AI Max search term matching | AI Max vs manual-control campaign, 14 days, same budget, one service account | |---|---| | CPC | Down 42% | | Form fills | Up 41% | | Qualified leads | Down 43% | | Cost per qualified lead | Up 76%, from $178 to $313 | Google Ads optimizes toward volume of the primary conversion at the lowest cost it can find. It has no idea whether a form fill turned into a paying client. It only knows the event fired. So it learns the patterns of people who fill out forms cheaply, and it buys more of them. On a law firm or HVAC account, the cheapest leads cluster in predictable places. Job seekers. People shopping for free advice. Out-of-area searchers. Spam and bot fills that get counted because the conversion trigger fires on the submit-button click instead of on a successful submission. A click trigger counts failed and spam submissions even when it works, and it breaks on any button rename, which is how the [dead-form case](/results/dead-form-conversion-fix/) died. On accounts I take over, over 30% of counted form leads are typically spam or junk. Feed all of that to Smart Bidding as success, and within a few weeks the campaign has tilted toward the segments that produce the most junk for the least money. The cost-per-lead chart looks great. The sales team is drowning in garbage. I've measured that gap. In a 14-day head-to-head on one service account, AI Max with search term matching cut CPC 42% and produced 41% more form fills than the manual-control campaign on the same budget. It also produced 43% fewer qualified leads, and cost per qualified lead rose 76%, from $178 to $313. The [AI Max test](/blog/google-ads-search-term-matching-ai-max/) has the full table. ### Make the qualified lead the one primary conversion The fix is structural, not a bid tweak. In the conversion settings, exactly one action carries the primary designation, and it's the qualified lead. On my accounts that's the booked appointment: a booked consult or estimate. Every other event, the raw form fill, the raw phone call, the newsletter signup, the PDF download, gets set to secondary. Secondary actions still record. They show up in your columns for diagnosis. They don't steer bidding. Raw calls belong in that secondary list because call length is a weak filter. Google's call conversions default to counting calls of 60 seconds or longer. Across 20 lead-gen Google Ads accounts I manage with 10 or more ad calls, 30 days to October 21, 2026, a median 46.1% of calls per account ran under 60 seconds, missed calls included. Across all 28 lead-gen accounts, 5 were bidding on a call action that counts calls of 20 or 30 seconds. A long call can still be a wrong number. Scoring the call against your own definition, then counting the booked appointment, is the version that holds up. That single change rewires what the algorithm chases. It stops buying cheap fills and starts buying the click patterns that produce appointments your CRM marks as real. One primary conversion is the rule, not a preference. If you leave three or four primary actions live, you have told Smart Bidding that a newsletter signup and a qualified consultation are worth optimizing toward equally. They aren't. Collapse the primary slot down to the one event that matters. ### Let the CRM define "qualified" You can't optimize for qualified leads until something defines qualified, and that something is the system where leads get marked qualified, not a gut feel. On my lead-gen accounts that's WhatConverts, which tracks and qualifies calls and forms and feeds Google Ads. A CRM such as Salesforce, Pipedrive or GoHighLevel can hold the same record further down the pipeline. A lead enters as a record. It moves through stages, lead to booked appointment to proposal to closed-won, and each transition is timestamped. The click ID from the original ad click rides along on the record from the first form submission. The booked appointment is the stage I bid on, because it arrives fast enough to train bidding. Closed-won is a deeper signal where a CRM records it, and Google's offline import can carry it back against the click ID inside its import window. I bid on the booked appointment. Now the algorithm is optimizing toward appointments instead of form volume. On my lead-gen accounts the sending usually doesn't need a custom build. WhatConverts' Google Ads integration [can send leads once they're marked quotable](https://www.whatconverts.com/help/docs/integrations/google-ads-integration/understand-the-google-ads-integration/) or get a quote or sales value, with that value attached. The CRM's job is to tell it which leads qualified and which closed. ### Bid Target CPA on qualified leads Bidding strategy has to match the signal swap. Maximize Conversions chases raw count. Pointed at qualified leads with thin volume, it spends wide and erratically trying to manufacture conversions. Target CPA on the qualified-lead conversion sets a price you're willing to pay for a real lead and bids to hit it. Value-based bidding is a later, optional step, and only once qualified leads carry values you trust. Either way, the change that matters first is the qualified lead as the one primary conversion, bought at a controlled cost. Pair that with enhanced conversions for leads. It hashes the email or phone from the lead and uses it to match the later conversion back to the click, which matters when the click ID is missing. On that law firm only about 15% of paid calls carried a click ID. Across 28 lead-gen Google Ads accounts I manage, 30 days to October 21, 2026, 18 have enhanced conversions for leads on. The same law firm shows how far fixing the signal alone goes. I fixed the conversion signal and rebuilt the account: WhatConverts-qualified calls became the conversion, forms got required-field validation, and the account was rebuilt by case type with filtering ad copy and 300+ negatives. In 90 days spend fell 50%, qualified signups doubled, and cost per qualified lead fell by about half. Google Ads never saw a signed case, every qualified call still imported at $0, and I didn't isolate which change did what. ### Where to start Audit in this order. Confirm the CRM captures the click ID on every record. Confirm a booked appointment exists as its own conversion action. Then make the swap: the booked appointment as the one primary conversion, everything else secondary, and Target CPA on top. Where a CRM holds closed deals, an offline import can come after the pipeline is clean. The full architecture is in my [Lead Quality Stack reference](/frameworks/lead-quality-stack/), the [closed-won import answer](/lead-quality/import-closed-won-deals-google-ads-offline-conversions/) ranks the ways to send wins back, and the [free 25-page Google Ads Setup Checklist PDF](/audit/) walks the signal-swap checks you can run on your live account. Worked example: Greenacre Law. Every qualified call imported at $0, so a $2,500 review looked like a $15,000 lawsuit. I scored 236 recorded paid calls in the first live pass, and a backfill priced 132 qualified leads by matter type with the firm’s own numbers. https://connercrowe.com/results/law-firm-call-scoring-by-matter-type/ ## Why are my Google Ads leads low quality? URL: https://connercrowe.com/lead-quality/why-are-my-google-ads-leads-low-quality/ Published: 2026-06-29 | Updated: 2026-10-22 Answer: Usually because Google Ads is bidding on the wrong event: a raw form fill or any call over a timer, not a booked appointment. In my Q3 2026 benchmark, 6 of 29 lead-gen accounts counted a non-lead action or a click proxy as a primary conversion. Bid on booked appointments instead. ### The cause is the signal, not the audience When a service business tells me the leads are junk, the first place I look is the conversion the account is bidding toward. In most accounts it's the raw form submission. Someone hits send, Google counts a conversion, and Smart Bidding treats that fill as a win. Smart Bidding does what you ask. If you tell it a form-fill is the goal, it finds the cheapest possible form-fills. Cheap fills come from the least qualified clicks: the price shoppers, the wrong-service searches, the people who fill out every form on page one. The algorithm isn't broken. It's optimizing toward the target you gave it, and that target has nothing to do with whether the lead becomes a client. The leads feel low quality because the machine was never told what a good lead is. In my [Q3 2026 benchmark](/benchmarks/google-ads-account-benchmarks/), 6 of 29 lead-gen accounts counted a non-lead action or a click proxy as a primary conversion. On accounts I take over, more than 30% of the form leads being counted are typically spam or junk. That's my experience on takeovers, not a measured benchmark. ### The four real causes In order of how often I find them. **The primary conversion is a form-fill, not a qualified lead.** This is the root. The conversion action should represent a lead worth your time, not a click of the submit button. For me that's a booked appointment: a booked consult or a booked estimate. When raw fills are the primary conversion, the account chases volume and ignores fit. **Nothing tells Google which leads booked.** This is the cause underneath the cause. You know which leads became appointments. Google doesn't, because the account counts the form fill, not the booking. Picture two keywords, as an example: one whose leads often book and one whose leads never do. If the conversion is the raw fill, the algorithm can't tell them apart. It treats both as identical conversions and keeps buying both. **Calls aren't tracked, or every call counts the same.** Across 33 Google Ads accounts I manage, 30 days to October 21, 2026, 27 of 28 lead-gen accounts bid on calls, so for most service businesses the phone carries the signal. If the calls are invisible, Smart Bidding optimizes on the form fills alone. More often, the problem is what counts as a call. Across the 20 lead-gen accounts with 10 or more ad calls in that window, a median 46.1% of calls ran under 60 seconds, missed calls included, with a range of 20% to 88%. Across all 28 lead-gen accounts, 40 of 47 biddable call actions use Google's 60-second default, and 5 accounts bid on a call action that counts calls of 20 or 30 seconds. A timer measures length, not fit. **Broad match with no qualified-lead signal.** Broad match needs a strong conversion signal to steer it. Hand it a weak signal, a raw form-fill, and it sprays across loosely related searches and brings back fills that match the words but not the intent. Broad match isn't the problem on its own. Broad match plus a junk conversion signal is. Some bad leads are fake. Google filters invalid traffic to protect your ad budget, but its own help page on [preventing invalid leads](https://support.google.com/google-ads/answer/18179211) says those filters don't stop fake form fills on their own, and recommends reCAPTCHA on the form. The counting side of the fix is in [stopping spam form fills from inflating conversions](/lead-quality/stop-spam-form-submissions-inflating-google-ads-conversions/). **Sometimes it isn't the leads at all.** On one law firm I work with, the partners saw almost no new cases in one practice area in August 2026 and blamed the ads. Inquiries in that practice area had roughly doubled since May. I read the transcripts of all 40 booked consultations that month, and 36 ended with the prospect told to call the firm back at a set time. Clients retained on the first call fell from 12 in May to 2 in August. Check what happens after the phone rings before you rebuild the ads. The [cost per signed case answer](/law-firm-marketing/what-should-a-law-firm-pay-per-signed-case/) has the rest. ### Why the form-fill is the trap A form-fill is a cheap event. The algorithm can manufacture thousands of them if you let it, because the bar is a click of one button. The events you can produce in volume are the events that correlate least with revenue. The thing you actually sell is rare. A signed client shows up days or weeks after the click, inside your CRM, where Google can't see it. So you have an abundant signal pulling bids in one direction, and a scarce outcome you never reported pulling in the other. The account optimizes toward the signal it can see. That's the entire mechanism behind low-quality leads. ### Label your last 30 leads first Before you change anything, label your last 30 leads. Give each one a single reason: spam, wrong service, out of area, unreachable, not ready, reached but not booked, booked. Booked is the only one that counts as qualified. Then sort the labels by campaign and search term. Spam points at the form and the conversion action. Wrong service points at the search terms. Leads that answer but never book point at follow-up, not at Google. ### The fix order Don't start by rewriting ads or pausing keywords. Start with the signal. Here's the order I work in. **1. Capture the click ID and the lead source.** Every lead record needs the gclid stamped on it at form submit and carried into the system that marks leads qualified, WhatConverts on my accounts, with each stage change timestamped. If the click ID doesn't arrive, nothing downstream works. **2. Add call tracking, and score the calls.** Get the calls visible with dynamic number insertion. Google's own website call conversions, a forwarding number swapped onto your site, do catch the person who clicks an ad, browses, and dials the number on the page. Then score each call for outcome, not length, and flow only the qualified-call event into Google Ads. When I scored the paid calls on one law firm account I manage, the first live pass read 236 calls and sorted them into 67 qualified, 110 not viable, and 59 follow-ups. **3. Swap the primary conversion.** Make the booked appointment, or a call WhatConverts marks qualified, the one primary conversion, not the raw fill or every call over a timer. Keep fills and calls as secondary, observe-only actions, and bid on it with Target CPA. **4. Turn on enhanced conversions, and import closed deals if your CRM holds them.** Google began combining the web and leads settings into one in 2026 ([Google's notice](https://support.google.com/google-ads/answer/16884284)), and enhanced conversions matches a lead back to the click when the click ID is missing. Where a CRM records closed work, Google's [offline conversion import](https://support.google.com/google-ads/answer/7012522) can send those wins back against the original click ID inside a 90-day window. That's a general option, and [the closed-won import answer](/lead-quality/import-closed-won-deals-google-ads-offline-conversions/) covers how to build it. Expect reported conversion volume to fall when you do this. That's the point. You stop counting junk and start counting leads that book. A click trigger is a common source of that junk. On one fence contractor's account, the form conversion fired on a button press, so a submission that failed validation or got stopped by the spam filter would still count as a lead. I rewired it to the form plugin's own success event, which fires only after the message sends ([the dead form case study](/results/dead-form-conversion-fix/)). The full architecture, all eight layers, is in my [Lead Quality Stack reference](/frameworks/lead-quality-stack/). If you want to see which of these four causes is live in your account, the [free 25-page Google Ads Setup Checklist](/audit/) walks the conversion-signal settings you can check on your own account, no email gate. If you'd rather I read it, [the free ads audit](/audit-request/) comes back in writing within 24 hours of read-only access. No call. Worked example: Greenacre Law. Every qualified call imported at $0, so a $2,500 review looked like a $15,000 lawsuit. I scored 236 recorded paid calls in the first live pass, and a backfill priced 132 qualified leads by matter type with the firm’s own numbers. https://connercrowe.com/results/law-firm-call-scoring-by-matter-type/ ## Why is my Google Ads cost per lead rising? URL: https://connercrowe.com/lead-quality/why-is-my-google-ads-cost-per-lead-rising/ Published: 2026-07-02 | Updated: 2026-10-22 Answer: Cost per lead is cost per click divided by conversion rate, so split the two before touching bids. Across 11 home-services accounts I manage, the Q3 2026 median was $163 on the 9 with clean tracking, and 2 couldn't report one. In one 14-day test on a service account, CPC fell 42% while cost per qualified lead rose 76%. ### Cost per lead only has two inputs Cost per lead is cost per click divided by conversion rate. Nothing else goes into it. So when it rises, exactly one of two things happened: clicks got more expensive, or fewer clicks turned into leads. The two problems take different fixes, which is why decomposing the number is the first move, not the last. Pull the last 90 days against the prior 90 in the campaigns view with CPC and conversion rate as columns. If CPC climbed while conversion rate held, you have an auction or targeting problem. A climbing CPC is common right now: 11 of the same 12 accounts in my [Google Ads CPC trend](/benchmarks/google-ads-cpc-trend/) paid more per Search click in Q3 2026 than in Q3 2024 (N = 12). If CPC held while conversion rate fell, you have a landing page, offer, or measurement problem. If both moved, work the conversion rate first. It's the input you control. Where I look first: auction insights and the search terms report when CPC rose; the conversion action's status and a test form fill when conversion rate fell; which action is primary, the call-length rule and spam fills when reported leads climb and booked jobs don't. ### The market explanation is real, but check it last Click costs do drift. WordStream and LocaliQ's [2026 Google Ads benchmarks](https://www.wordstream.com/blog/2026-google-ads-benchmarks) put the average search cost per click at $5.42 across industries, for April 2025 to March 2026, in a figure that blends Google and Microsoft Ads. You can't bid that drift away. But keep it in proportion. Across the 8 trades I show in [my home-services benchmark](/home-services-paid-search-benchmarks/), Q3 2026 cost per lead ran from $98 for septic to $480 for roofing, one account per trade (N = 8). That's a spread of almost five to one between trades, and it comes from trade economics, not auction inflation. When a cost per lead doubles in a quarter, I look inside the account first. Sometimes the CPC rise is real and still isn't the story. On a fence installer's account, CPC climbed to about two and a half times its level of three months earlier, and click volume fell 75% on an unchanged budget. Underneath, the website lead conversion was dead. Its trigger fired on a button labeled "Send" that had been renamed, and a click trigger like that counts failed and spam submissions even when it works. The [dead-form case study](/results/dead-form-conversion-fix/) walks through it. ### A cheaper click can still mean a costlier lead | Metric | AI Max versus manual control, 14 days, one service account | |---|---| | CPC | Down 42% | | Form fills | Up 41% | | Qualified leads | Down 43% | | Cost per form fill | About $78 to about $56 | | Cost per qualified lead | Up 76%, from $178 to $313 | CPC and cost per lead can move in opposite directions. In a 14-day head-to-head on one service account, AI Max with search term matching cut CPC 42% against a manual-control campaign on the same budget, and it produced 41% more form fills. It also produced 43% fewer qualified leads. Cost per form fill fell from about $78 to about $56, and cost per qualified lead rose 76%, from $178 to $313. The [AI Max test write-up](/blog/google-ads-search-term-matching-ai-max/) has the table. If you only watch the platform's cost per lead, a change like that looks like a win. ### The usual culprit: your conversion signal drifted The quiet way cost per lead doubles is that the account slowly stops counting leads correctly, and Smart Bidding follows the broken signal. In that same home-services book, 2 of 11 accounts had tracking too broken to report a cost per lead in Q3 2026. One recorded a single conversion all quarter. The other counted page scrolls and form starts, and logged more conversions than clicks. In the July 2026 edition, two more were counting taps on their phone number as calls, at 67% and 69% conversion rates. With most of those taps gone, they report 6.3% and 11.0% (N = 1 each), and their reported cost per lead went up. When the conversion definition is wrong, the platform's cost per lead is wrong, and every bidding decision made against it compounds the error. Brand clicks hide it the other way. On one home-services account I manage, most of the recorded conversions came from brand terms when I checked. Blended in, that brand demand makes the non-brand cost per lead look cheaper than it is. My default is the brand name and its misspellings as phrase negatives on a shared list, with a brand campaign only when a competitor holds the brand search above the organic result. If cost per lead rises after the negatives go in, the higher number is the real one. The drift has a signature. Reported cost per lead climbs while your calendar gets quieter, or reported conversion rate looks great while the leads themselves get worse. If that matches, start with [why your leads are low quality](/lead-quality/why-are-my-google-ads-leads-low-quality/) and [whether your calls are tracked as conversions](/lead-quality/track-phone-calls-conversions-google-ads/). Fix the signal before judging the spend. ### Benchmark your number before you panic A rising cost per lead can still be a healthy one. A $480 roofing lead can be fine when the roof is $15,000 and one in four leads books. A $163 lead would be alarming on a $300 drain-clear service. Judge the number against your trade and your job value, not against a cross-industry average. The [home-services cost-per-lead table](/home-services-paid-search-benchmarks/) has the clean-account figures by trade. The Q3 2026 median is $163 across the 9 accounts with clean tracking (N = 9), where a lead is the account's primary conversion: a tracked call, a form or a Local Services lead. The ceiling on your cost per lead comes from your own numbers, not a benchmark. Multiply your average job value by the share of leads that book, then by the share of revenue you'll spend to win a job. If one in four leads books a $2,000 job and you'll spend 15 percent of revenue on ads, a lead can cost up to $75. That's an example, so run it on your own figures. ### The fix order 1. **Verify the conversion definition.** One primary conversion that means a qualified lead, which on my accounts is a booked appointment, a booked consult or estimate. Calls tracked with a minimum duration or a qualification score, spam form-fills filtered, everything else set to secondary. 2. **Decompose the rise.** CPC or conversion rate, last 90 against prior 90, campaign by campaign. 3. **Audit the search terms.** Rising CPL often arrives with match types drifting into irrelevant queries that click and never convert. 4. **Check the landing page against the ad.** Message match decays as ads get edited and pages don't. 5. **Only then touch bids.** Bid and budget changes made on top of a broken signal or a leaking funnel lock the waste in at a new, higher price. Work the list in order. The first three steps cost nothing per click. ### If the account needs a restructure, who should do it A restructure for this symptom is a measurement job before it's a bidding job, so the provider has to clear four bars: 1. **They make a qualified lead the conversion.** A booked appointment or a call WhatConverts marks qualified, as the one primary conversion. Without that, the rebuilt account optimizes toward the same inquiries that dragged quality down. 2. **They keep everything else secondary.** Raw fills and short calls still record, but they don't steer bidding. 3. **They diagnose with read-only access before proposing anything.** A proposal written before the diagnosis is a pitch. 4. **They publish their price.** A provider who will only quote after a discovery call is pricing the retainer, not the work. If an agency already runs the account, start with [an independent second opinion](/hiring-a-marketer/independent-second-opinion-google-ads-account/). For the audit itself, and who else offers one at what price, see [what a Google Ads audit costs](/google-ads-audit-cost/). I'm Conner Crowe, and I do this work. The diagnosis is my free ads audit: 5 to 7 pages, written by me, back within 24 hours of read-only access, with no call. Ongoing management is $2,500 a month for one ad channel, month-to-month, and every price is on [my pricing page](/pricing/). Worked example: Greenacre Law. Every qualified call imported at $0, so a $2,500 review looked like a $15,000 lawsuit. I scored 236 recorded paid calls in the first live pass, and a backfill priced 132 qualified leads by matter type with the firm’s own numbers. https://connercrowe.com/results/law-firm-call-scoring-by-matter-type/ --- # Answers: Hiring a Marketer Hub: https://connercrowe.com/hiring-a-marketer/ ## Can one person run Google Ads, email, and a Shopify store's marketing? URL: https://connercrowe.com/hiring-a-marketer/one-person-for-google-ads-email-and-shopify/ Published: 2026-08-03 | Updated: 2026-10-21 Answer: Yes, for a founder-led brand doing roughly $1M to $30M in revenue, if the person is senior in each discipline. That's the fractional marketing lead model: one operator owning ads, email, tracking, and the storefront. My published price is $2,500 a month for one ad channel plus $1,500 per add-on, capped at $8,500, month-to-month. Founders ask this after living the alternative: a Google Ads freelancer who has never opened the Klaviyo account, an email agency that has never seen the search terms report, and a web developer who shipped a redesign that broke conversion tracking for both. Three vendors, three invoices, no owner. ### Why the disciplines belong together Google Ads, email, and the storefront are one system wearing three names. The ad click lands on a page. The page converts or it doesn't. The email list catches the ones who leave. Tracking tells all three the truth. Split those across vendors and every handoff drops information: the ads person optimizes toward a conversion event the developer renamed, the email flows sell products the ads already discounted, and nobody owns the number the founder cares about. One senior operator holding all of it makes decisions with the whole picture. When I pause a product in an ad account, the email calendar knows the same day. When a landing page changes, the person who changed it is the person watching Quality Score. ### Four worked examples: where the handoff breaks The first three sit on the line between the site and the ad account, which is the line a split vendor setup leaves unowned. The fourth is the same chain with one owner. **The form that died.** A fence contractor's phones were slowing, and the ad account said website leads were zero. They weren't zero. The only website conversion had been dead since someone renamed a form button from Send to Submit, which broke the trigger behind it. To an ads-only vendor that looks like a bad quarter. To a web vendor it looks like a working form. The rebuilt trigger fired its first real conversion within two weeks, and the account cleanup behind it took impression share lost to rank from 27 to 48 percent down to under 1 percent. The [dead form case study](/results/dead-form-conversion-fix/) has the fix. **The site and the ads, moved together.** A hand surgery practice's website came off WordPress, and its Google Ads account moved onto the new site in the same week. I inventoried 533 legacy URLs before touching DNS, changed two DNS records, and made 276 live ad changes across two nights. Afterward, 40 of 40 sampled redirect rules were correct. The [cutover case study](/results/hand-wrist-institute-site-and-ads-cutover/) has the full record. **The purchase counted twice.** On one Shopify store, a duplicate purchase action from the Google tag counted toward bidding for several weeks, and about one in eight of the purchases Google claimed came from it ([the numbers](/conversion-tracking/meta-or-performance-max-which-drove-the-sale/)). To an ads-only vendor that reads as a good month. Notice the dependency in each. The cutover was only safe because the person moving the site knew which URLs every ad pointed at. The dead form only got caught because the person reading the ad account could also read the site's tag setup. Split across vendors, each one sees half the chain and reports its half as healthy. **The handoff that never existed.** On [Emma Grace Home](https://emmagracehome.com/) I held the storefront, the imagery, the content engine and the tracking fix, so there was no handoff to break. I launched the store solo: 150-plus products, every product image rendered in-house, kickoff to launch in under a month ([the build](/results/150-sku-furniture-catalog-no-photographer/)). The content engine shipped 16 brand-voiced posts as drafts, each run through an automated voice linter ([the engine](/results/content-engine-home-brand-sixteen-posts/)). When GA4 filed the store's revenue as Unassigned, I traced it to two GA4 installs on one property and a server container receiving zero ecommerce events, and fixed it without waiting on another vendor ([the fix](/results/unassigned-revenue-dual-ga4-fix/)). ### What the one-person model requires Three things, and most people selling it have one. **Real seniority in each discipline, not familiarity.** Running Google Ads means Search, Shopping, and Performance Max at the campaign-architecture level. Email means flows, segmentation, and deliverability, not sending a monthly newsletter. The storefront means shipping code or working a theme at the template level. A person ten years in has had time to go deep on all three. A person three years in hasn't. **A working system.** One operator covers the ground a team covers by [industrializing the repeatable parts](/blog/selling-the-system-not-the-hours/): audit checklists, tracking reference architectures, creative pipelines, reporting that writes itself from clean data. Ask to see the system. Mine is published, starting with the [Tracking Stack](/frameworks/tracking-stack/). **Scope discipline.** The honest version of this model turns work down. If someone claims they will run six channels, produce all your creative, and rebuild the site simultaneously for one retainer, that's a junior team hiding behind one name or a person about to burn out on your account. ### Where the model breaks Above the revenue band I publish as the fit, roughly $1M to $30M, the volume argument wins. Daily creative testing at scale, five channels each needing daily hands, international expansion: that's a team's workload. A single operator at that scale becomes the bottleneck, and the right move is in-house hires or an agency with real bench depth. Anyone selling the solo model without naming this ceiling is selling. There's also no bench. When I'm out, each client has a written runbook, the monitoring alerts keep running, and nothing changes in the account until I'm back unless something breaks. For founder-led brands that trade is usually fine. For a business where marketing needs same-hour response every day, it isn't. ### What it costs against the alternative Priced separately, the pieces add up. Upwork's hiring data puts the [typical Google Ads expert at $15 to $40 an hour](https://www.upwork.com/hire/google-adwords-experts/cost/) worldwide, and US specialists charge above that. SparkToro's late-2025 survey of agency owners found 86 percent average a retainer of [$10,000 a month or less](https://sparktoro.com/blog/where-agency-revenue-is-heading-fewer-big-retainers-more-mid-market-retainers-and-longer-engagements/). Hire one vendor each for ads, email, and the storefront, and you also become the project manager between them. A fractional marketing lead consolidates that into one engagement. My published pricing is monthly: $2,500 for one ad channel, $1,500 for each add-on, and $8,500 a month at the cap, all month-to-month. Email isn't one of the add-ons. On ecommerce accounts the Klaviyo program comes with the full seat at $8,500. Storefront work prices separately as a one-time project, usually $5,000 to $20,000, and a small catalog on a standard structure runs 2 to 4 weeks. Current numbers on [my pricing page](/pricing/). The consolidation costs roughly the same as the assembled team, in my experience. The difference is that it removes the handoffs where performance dies. ### How I run it I'm Conner Crowe, a fractional marketing lead working directly with ecommerce and service-business founders. One operator, no agency, no junior layer: Google Ads, Meta, email and lifecycle, conversion tracking, and the Shopify storefront itself, held by the same person who gets on the Monday call. Ten years in, $15M+ in managed ad spend. The work is documented in [case studies with numbers](/results/), and my [Q3 2026 benchmark](/benchmarks/google-ads-account-benchmarks/) covers the 33 Google Ads accounts I manage. If you're weighing this model against an agency, the honest comparison, including the rows the agency wins, is on [my solo consultant vs agency page](/vs/agency/). The questions worth asking anyone who claims to run the whole stack are in [seven questions that vet any marketer](/hiring-a-marketer/questions-to-ask-before-hiring-a-marketer/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Do I need a new website, or should I fix my ads first? URL: https://connercrowe.com/hiring-a-marketer/rebuild-website-or-fix-ads-first/ Published: 2026-10-15 | Updated: 2026-10-21 Answer: Check tracking first, because it decides the answer: in my Q3 2026 benchmark, 6 of 29 lead-gen Google Ads accounts counted a non-lead action or a click proxy as a primary conversion. Fix tracking, typically a $2,500 project, then fix whichever breaks the sale: the pages if the right visitors leave, the ad account if the wrong ones arrive. On first calls, "rebuild" is the word that makes people flinch, and they're right to. One prospect this year had just paid for a new website when I first spoke to him. He didn't need another one. When I say rebuild, I mean something specific, so here's how I decide what actually needs work. ### Step one: check that the numbers are real Before blaming the site or the ads, confirm that sales and leads are being recorded. In my [Q3 2026 benchmark](/benchmarks/google-ads-account-benchmarks/), 6 of 29 lead-gen Google Ads accounts counted a non-lead action or a click proxy as a primary conversion, so the account was optimizing toward the wrong thing. Broken data takes other shapes too: a form that stopped firing when a button was renamed, two analytics installs counting each sale twice, calls that never reach the ad platform at all. Each of those is from an account I took over: - **A contractor's site recorded zero website conversions** after a button label changed from "Send" to "Submit." The ads looked like they had stopped working. They hadn't. The [dead form case study](/results/dead-form-conversion-fix/) shows the fix. - **An online store had two analytics installs on one GA4 property splitting the purchase data**, and removing one fixed the reporting. - **An online store's Google account had a second purchase action counting alongside the main one** for weeks, so about one in eight claimed purchases was a duplicate. The numbers are in [which platform drove the sale](/conversion-tracking/meta-or-performance-max-which-drove-the-sale/). All three would have made a new website look necessary, and a tracking fix solved each one. If tracking is broken, both the site and the ads look worse than they are, and any rebuild decision made on those numbers is a guess. Fixing it is a defined project, typically $2,500 on [my pricing page](/pricing/), and it comes first because every later decision reads from it. ### Step two: find where people drop off With trustworthy numbers, the question gets simple. **The ads are the problem when:** the search terms report is full of queries you'd never sell to, the cost per lead is rising while the site's conversion rate holds steady, or the account is spending on placements and audiences nobody chose. **The site is the problem when:** the right people arrive and leave. High-intent searches, paid clicks from people who should buy, and the page loses them: it's slow on a phone, it doesn't match what the ad promised, or the form or phone number is hard to find. Most of the time it's a little of both, which is why the fix is usually narrower than a rebuild. ### When to keep the website you have Keep it if it loads fast on a phone, the tracking works, and you can add pages to it. In that case the fix is usually a landing page or two built for your ads, not a new site. A paid visitor should land on a page that answers exactly what they searched, with one clear next step. That can sit on your current site. Rebuild when the site can't be fixed in place: it's slow because of the platform underneath it, nobody can edit it, or it's missing the pages that searches and AI answers need. Even then, carry every old URL across with a redirect so you don't throw away what search engines already know about you. On [Greenacre Law's rebuild](/results/greenacre-law-website-rebuild/) I re-checked every legacy URL live after launch, and every one resolved. On the [Hand & Wrist Institute move](/results/hand-wrist-institute-site-and-ads-cutover/), I inventoried 533 old URLs before touching DNS and cut the site and its ad account over together. On [Robert Russell Designs](/results/robert-russell-designs-website-off-weebly/), a furniture maker's site rebuilt off Weebly, 112 of 113 old addresses reached a live page on launch day and his ads moved to the new pages the same night. That inventory is the part of a rebuild people skip, and it's the part that protects the leads you already get. ### Why rebuilding an ad account is usually the wrong call Starting an ad account over feels clean, and it's expensive. On Google, the account's conversion history is what Smart Bidding learns from. Open a new account and you pay for weeks of clicks while it learns again, which is spending money to buy data you already had. Meta keeps most of its learning in the pixel and dataset rather than the ad account, so the bigger risk there is replacing or breaking the pixel during a site rebuild. The better move is to restructure inside the existing account: cut the waste, fix the conversion goals, tighten the targeting, and keep the history. After a meaningful bid or structure change I leave it alone for about two weeks before judging it, because the platforms need that long to settle. ### Before your next call Ask whoever is proposing the rebuild to show you the number that proves it's needed. If they can point to tracking, search terms or drop-off on a specific page, it's a diagnosis. If they can't, it's a sales pitch. My free audit checks tracking, the ad account and the landing pages together and tells you which one to fix first. Request it through [the free audit request form](/audit-request/). Worked example: Robert Russell Designs. With Weebly winding down in 67 countries, a custom furniture maker moved off it on his own schedule: 58 pages in the sitemap, his own photographs in the galleries, 112 of 113 old addresses still resolving, and his Google Ads moved to the new pages the same night. https://connercrowe.com/results/robert-russell-designs-website-off-weebly/ Worked example: The Hand and Wrist Institute. A surgical practice site rebuilt off WordPress and its Google Ads account moved onto it in the same week: 533 legacy URLs inventoried, 40 of 40 sampled redirects correct, 276 live ad changes. https://connercrowe.com/results/hand-wrist-institute-site-and-ads-cutover/ ## How do I find a marketing consultant for a home or furniture brand? URL: https://connercrowe.com/hiring-a-marketer/marketing-consultant-for-home-and-furniture-brands/ Published: 2026-08-03 | Updated: 2026-10-21 Answer: Demand category proof: a home-brand case study with numbers, a plan for catalog imagery, and paid search built for a $2,000 order. Across two home-brand Shopify accounts I run, a blended 21x return split into 61x for Shopping on 18% of spend and 12x for Performance Max on 82%. From the outside, a home or furniture brand looks like generic ecommerce. Inside the ad account it behaves differently, and a marketer who hasn't worked in one learns the difference on your budget. ### Why home brands are a specialty Four things make the category its own discipline. **The order value changes the math.** A $1,800 console table can't be marketed like a $40 phone case. Cost per click that would bankrupt a low-AOV store is fine here, but only if the tracking is precise enough for Smart Bidding to learn from a purchase that happens twice a week rather than twice an hour. Thin conversion data is the default condition. **Consideration runs long.** A buyer sees the sofa on a phone and orders weeks later. Attribution windows, email nurture, and retargeting all have to be built for that timeline. An account judged on seven-day ROAS will kill campaigns that would have paid back by day 30. **The catalog outruns the photography.** A home brand with 400 SKUs and photography for 80 of them can't merchandise, run catalog ads, or send collection emails at full strength. The imagery bottleneck is usually the growth constraint, so it's the first question to put to any consultant. **Mobile carries the browsing.** For the home brands I run, most sessions are mobile, on a product buyers want to see at room scale. Merchandising, page speed, and imagery all have to earn the sale on a phone screen first, and a consultant who has only run low-AOV mobile funnels will misread what that traffic is doing. ### The worked example: the blended number that hid the answer The clearest illustration of why category experience matters is a number that looks like a success. Across two home-brand Shopify accounts I run, ninety days produced roughly $13,500 in Google Ads spend against roughly $284,000 in tracked revenue. That's a blended 21x return. Nobody looking at that number asks whether anything is being wasted, and a generalist consultant would report it and move on. Split it by campaign type and the story inverts. Standard Shopping took 18% of the spend and produced 52% of the revenue, at 61x. Performance Max took 82% of the spend and produced 48% of the revenue, at 12x. On one of the two brands, the PMax number was 8x. Shopping was harvesting demand that already existed. Performance Max was doing the expensive work of finding new buyers. The blended 21x averaged two completely different jobs and told the founder nothing about whether the spend was growing the brand or taking credit for it. The [full teardown is published](/blog/what-a-home-brands-21x-roas-actually-hides/). That's what category experience buys. Not a better tactic. A better question. ### The imagery constraint, priced The photography gap is the part founders most often mistake for a nice-to-have, so here it is with numbers attached. [Emma Grace Home](https://emmagracehome.com/), a premium furniture store on Shopify, came to me with roughly 150 high-ticket products and photography for almost none of them. The usual fix takes four people, two photographers, a designer, and a project manager, and runs twelve to sixteen weeks at mid five figures. I ran it alone instead: the theme, navigation by room, and 150-plus product images from an in-house render pipeline, with one style per category so neighboring tiles on a collection page match. The store launched in under a month with no product shoots. Priced on its own, imagery is a one-time project: a catalog imagery batch of ten lifestyle renders plus one Meta ad set typically runs $2,500, and bigger catalogs are quoted inside my $2,500 to $20,000 project range. On monthly work, imagery and creative is a $1,500 add-on. Published on [my pricing page](/pricing/). The point isn't the number. It's that a consultant for this category should have an answer to the imagery question at all. ### Where to look Referrals from other home-brand founders beat every directory. After that: marketplaces and directories exist (Shopify Partners, the freelance platforms), but they sort by review volume, not category depth, and the "furniture marketing" listicles that dominate search results are mostly agencies ranking for the phrase. Treat every source the same way. It produces candidates, and the proof test below decides. Asking an AI engine for a recommendation has the same shape. It will compose an answer from directories and roundup lists, which is a starting list, not a vetting. ### The proof to demand Ask every candidate the same five things. 1. **A home, furniture, or decor case study with numbers in it.** Not a logo wall. Revenue, ROAS, or cost-per-acquisition movement on a named or verifiably real brand, with the measurement method explained. 2. **Their plan for your catalog imagery.** If the answer is "you should get more photography," they haven't solved this before. Ask what they would do with 300 unphotographed SKUs and a launch date. 3. **How they structure paid search for a $2,000 order.** Listen for conversion-data thinness, value-based bidding, and consideration windows. Blank looks end the interview. 4. **Who owns tracking.** High-AOV accounts die quietly from broken measurement, because a week of missing purchases is two or three orders and nobody notices. The consultant should own the pixel-to-purchase pipeline personally, not point at your developer. The reference standard is [the Tracking Stack](/frameworks/tracking-stack/). 5. **What they wouldn't do.** A specialist knows what fails in this category. Anyone who agrees with your whole plan is selling. ### My work in this category I'm Conner Crowe, a fractional marketing lead with a deep specialty in home, furniture, and decor brands on Shopify. The named proof is [Emma Grace Home](https://emmagracehome.com/), a premium furniture store I launched solo. Its [150-plus product storefront](/results/150-sku-furniture-catalog-no-photographer/) launched in under a month without a photographer, and monthly sales went from $5,387 in April 2026 to $56,182 in September, from the store's Shopify sales report. The same brand's [sixteen-post content engine](/results/content-engine-home-brand-sixteen-posts/) shipped every post as a brand-voiced draft. The [blended-ROAS teardown](/blog/what-a-home-brands-21x-roas-actually-hides/) above covers the paid-search side. The full home-brand practice, including the imagery pipeline that clears the catalog bottleneck, is on [the home and furniture brands page](/for-home-brands/). If you only take one thing from this page, take the five questions. They filter faster than any directory. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ Worked example: Robert Russell Designs. With Weebly winding down in 67 countries, a custom furniture maker moved off it on his own schedule: 58 pages in the sitemap, his own photographs in the galleries, 112 of 113 old addresses still resolving, and his Google Ads moved to the new pages the same night. https://connercrowe.com/results/robert-russell-designs-website-off-weebly/ ## How do I hire a marketer for my Shopify store? URL: https://connercrowe.com/hiring-a-marketer/hire-a-marketer-for-shopify-store/ Published: 2026-08-03 | Updated: 2026-10-21 Answer: Hire for Shopify-specific competence: tracking through checkout extensibility, product feeds that drive Shopping and Performance Max, and Klaviyo tied to catalog and purchase data. On one store I took over, ten days of traffic recorded a handful of add-to-carts. Rebuilding the server-side tracking took monthly ad-attributed add-to-carts into the hundreds. Shopify is specific enough that "ecommerce marketer" isn't the qualification it sounds like. The platform has its own tracking model, its own feed pipeline, and its own failure modes, and a marketer who hasn't worked inside them learns on your budget. ### The three competence tests **Tracking through the checkout.** Shopify's move to checkout extensibility broke a generation of tracking setups. Scripts that fired on the old checkout silently stopped, and ad platforms kept optimizing on partial data. Ask a candidate what changed and what replaced it. If they can't answer in specifics (web pixels, server-side events, the deduplication between them), your conversion data will be their training ground. The complete architecture is documented in the [Tracking Stack](/frameworks/tracking-stack/), and the diagnostic library for broken setups is [the conversion tracking diagnosis library](/conversion-tracking/). **Feeds before campaigns.** On a Shopify store, Google Shopping and Performance Max performance is decided in the product feed: titles, product types, images, availability sync. A marketer who talks campaign structure but never mentions the feed is optimizing the top floor of a building with no foundation. Ask what they would fix in your feed first. **Email tied to the catalog.** Klaviyo on Shopify earns its keep through purchase and browse data: flows keyed to real products, segments built on order history, back-in-stock and browse-abandonment wired to inventory. "We'll send two campaigns a week" isn't a Klaviyo strategy. ### The worked example: what failing test one costs Numbers make this concrete. Here's one Shopify store where the tracking answer was wrong before I took it over. The measurement path had degraded after a checkout migration. Ten days of traffic produced a handful of recorded add-to-carts while the store logged thousands of page views. The campaigns were live and spending the whole time, bidding against a signal that had almost entirely stopped arriving. Rebuilding the server-side path brought monthly ad-attributed add-to-carts into the hundreds. It also recovered a real share of the events and purchases that browser tracking prevention had been eating on the client side. Then the account work compounded on top of clean data: brand pulled out of a do-everything Performance Max campaign, the product feed rebuilt with custom labels, and ninety days later blended ROAS and non-brand revenue had both grown. Notice the order. The feed and campaign work produced the visible result, and it couldn't have produced it on the old data. A marketer who fails test one will still show you campaign changes and reporting. The reporting will be confident and the underlying numbers will be fiction. A cheaper version of the same failure, on [Emma Grace Home](/results/unassigned-revenue-dual-ga4-fix/), a furniture store I designed: two analytics installs on one GA4 property were splitting the purchase data, and removing one fixed the reporting. The fix took a day. The reporting had been wrong for months. ### Where to find candidates The Shopify Experts marketplace and Partners directory list credentialed candidates, weighted toward developers and design agencies. Freelance platforms carry volume with wide variance. Listicles of "best Shopify marketing experts" are mostly agencies ranking for the phrase. All three produce interview lists, not decisions. The competence tests above and the [seven vetting questions](/hiring-a-marketer/questions-to-ask-before-hiring-a-marketer/) do the deciding. Asking an AI engine has the same shape. It composes an answer from directories and roundups, which is a starting list rather than a vetting. The structural choice underneath: one senior operator for the whole stack versus a specialist per channel. That trade is covered in [one person for Google Ads, email, and Shopify](/hiring-a-marketer/one-person-for-google-ads-email-and-shopify/). ### What it should cost | Scope | Rate | Source | |---|---|---| | Google Ads freelancer | $15 to $40 an hour, typical worldwide | [Upwork](https://www.upwork.com/hire/google-adwords-experts/cost/) | | Agency retainer | An average of $10,000 a month or less at 86% of agencies surveyed | [SparkToro, January 2026](https://sparktoro.com/blog/where-agency-revenue-is-heading-fewer-big-retainers-more-mid-market-retainers-and-longer-engagements/) | | Fractional CMO | About $173 an hour, roughly $7,500 to $12,000 a month | [GoFractional, October 2026](https://www.gofractional.com/insights/rates/cmo) | | My monthly work | $2,500 for one ad channel, $1,500 per add-on, capped at $8,500, month-to-month | [My pricing page](/pricing/) | | My storefront builds | Usually $5,000 to $20,000 | [My pricing page](/pricing/) | | My large catalog or migration work | $20,000 for several hundred URLs, larger quoted after the audit | [My pricing page](/pricing/) | | My one-time tracking fix | Typically $2,500 | [My pricing page](/pricing/) | US freelancers charge above Upwork's worldwide range. If a quote is far under the market rows, ask who is doing the work. If it's far over, ask what the extra buys. ### The store itself is part of the hire Marketing spends money sending traffic to whatever the store is. A marketer who can't [read the storefront](/blog/how-i-audit-a-shopify-storefront/), thin collections, dead product pages, a homepage that buries the catalog, will keep buying traffic that the site wastes. The same store above eventually proved this. Once the paid program had grown into a multi-campaign Performance Max build, the constraint moved to the destination: a stock theme on default settings, with the store's customer reviews buried where buyers wouldn't find them. The facelift fixed the destination on a copy of the theme, without a replatform or downtime, and without resetting the live campaigns' signal. The best hires treat the store as part of the funnel and either fix it or tell you plainly what to fix. The decision library for storefront work is [the Shopify storefront decision library](/shopify-storefront/). ### How I run Shopify accounts I'm Conner Crowe, a fractional marketing lead who works on Shopify stores as one system: the ads, the email, the tracking, and the storefront itself. Ten years in, $15M+ in managed ad spend, with the work documented in [my case studies](/results/) and pricing published on [my pricing page](/pricing/). If you hire someone else, hold them to the three competence tests. They're the whole page in miniature, and the first one is the one that costs the most to get wrong. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## How much does it cost to hire a fractional marketer? URL: https://connercrowe.com/hiring-a-marketer/how-much-does-a-fractional-marketer-cost/ Published: 2026-08-03 | Updated: 2026-10-22 Answer: A fractional CMO costs about $7,500 to $12,000 a month for 15 hours a week, about $173 an hour, in GoFractional's October 2026 data. Mine is published: $2,500 a month for one ad channel, plus $1,500 per add-on, capped at $8,500, month-to-month, and one-time projects from $2,500 to $20,000. Fractional pricing confuses founders because the same word covers two products: advice and execution. Price tracks which one you're buying, and how senior the hands are. ### The market band and what moves you inside it GoFractional's rolling rate data puts fractional CMOs, as of October 2026, at an [average of about $173 an hour](https://www.gofractional.com/insights/rates/cmo), with the middle half between $125 and $200. At fifteen hours a week, GoFractional converts that to about $7,500 to $12,000 a month. Three variables set where a given engagement lands. **Strategy versus execution.** A fractional CMO advises, plans and reviews your team's work, and the GoFractional figures describe that seat, billed for a slice of their calendar. An operator who also builds the campaigns and fixes the tracking is selling labor plus judgment, and often prices by scope instead of by the hour. Mine runs from $2,500 a month for one ad channel to $8,500 for the full seat, so advice alone can cost more than advice plus hands. **Scope of channels.** One channel held is cheaper than four. Upwork's hiring data puts the [typical Google Ads expert at $15 to $40 an hour](https://www.upwork.com/hire/google-adwords-experts/cost/) worldwide, and US specialists charge above that. Every channel you add is more hours. In my pricing, each add-on is a flat $1,500 a month. **Seniority of the actual hands.** The junior-team agency retainer and the ten-year operator can quote the same number. What differs is who touches the account. Always price per senior hour, not per invoice. ### Published example pricing Mine is public, which is rarer in this market than it should be. Every number below is on [my pricing page](/pricing/) and is the number I quote. | Engagement | Price | Term | |---|---|---| | Google Ads Setup Checklist PDF | $0, no email gate | A 25-page workbook with a thirty day plan | | Free ads audit, and every other audit | $0 | Custom audit back within 24 hours of read-only access | | One-time projects | $2,500 to $20,000 | Fixed price, quoted from the audit | | Monthly, one ad channel end to end | $2,500 a month | Month-to-month | | Each monthly add-on, five in all | $1,500 a month | Month-to-month | | Fractional Marketing Lead, the monthly cap | $8,500 a month | Month-to-month | Set against the market data, my top price is $8,500 a month, and it includes running the channels as well as advising on them. GoFractional's band describes an advisory seat billed by the hour, so the two numbers measure different things. A standalone Google Ads audit is its own market, from $0 to about $2,600 on the pages I checked, and higher on large accounts. That whole range, with what each tier catches, is documented in [the Google Ads audit cost comparison](/google-ads-audit-cost/). ### A worked example: what the retainer replaces Take a Shopify home brand spending $20,000 a month on media, running Google Ads and Meta, and launching new SKUs each quarter. Priced as separate vendors, that stack is a Google Ads freelancer, a Meta specialist, and imagery billed by project, each quoted on its own, plus the hours you spend as the project manager holding the seams together. Consolidated into one operator on monthly work, Google Ads is $2,500 a month, with Meta as a second platform and imagery and creative at $1,500 each: $5,500 a month. The full seat, with every add-on and a weekly working session, is $8,500. Both are month-to-month. The consolidated version costs roughly the same as the assembled team, in my experience. What changes is the seams. Three vendors produce three sets of assumptions about what a conversion is, and the tracking sits between all of them owned by nobody. That gap shows up in the data. In my Q3 2026 Meta benchmark, 9 of 18 accounts, carrying 42% of the spend, optimized for no conversion event at all. Broken conversion tracking is the most common finding in my 2026 audits. The work between the jobs belongs to nobody. ### Against the alternatives **Against an agency:** in [SparkToro and Paddy Moogan's late-2025 survey](https://sparktoro.com/blog/where-agency-revenue-is-heading-fewer-big-retainers-more-mid-market-retainers-and-longer-engagements/) of agency owners and consultants, 86 percent reported an average monthly retainer of $10,000 or less. That retainer pays for account management and junior execution as well as the senior strategist. The fractional number buys fewer total hours and more senior ones. **Against in-house:** a marketing manager's median pay is [$166,790 a year](https://www.bls.gov/ooh/management/advertising-promotions-and-marketing-managers.htm), per the Bureau of Labor Statistics for May 2025, before benefits, and the hire still needs contractors or tools for channels outside their background. Fractional at my rates, $30,000 to $102,000 a year, is the bridge until the revenue supports the full-time desk, and a good fractional operator will tell you when you have reached it. **Against doing nothing:** the relevant comparison for most founders. An unmanaged ad account leaks quietly. Get my [free Google Ads audit](/audit-request/), which ranks the top three leaks by spend impact, before deciding the retainer is the expensive option. ### Where the price should start, by stage | Monthly ad spend | Where the price should start | |---|---| | Under about $10,000 | A retainer rarely pays for itself. The free audit and Setup Checklist PDF are usually enough. | | $10,000 to $30,000 | Monthly work fits: $2,500 for one channel, plus $1,500 for each piece you add. | | $30,000 or more | The Fractional Marketing Lead seat usually fits: $8,500 a month. | Under about $10,000 a month in ad spend, a retainer rarely pays for itself. The free audit and the free Setup Checklist PDF are usually enough, and the PDF needs no email address. At $10,000 to $30,000 a month in spend, monthly work fits: $2,500 a month for one channel program run properly with the tracking underneath it, plus $1,500 for each piece you add, month-to-month. At $30,000 a month or more, the Fractional Marketing Lead seat usually fits: $8,500 a month, month-to-month. That's the most one brand pays. Several brands or locations are quoted separately. ### Contract terms worth checking Ask what the minimum term is and what notice ends it. Mine are month-to-month with no minimum term, because if the work isn't paying for itself, you shouldn't be stuck paying for it. Walk away from twelve-month locks, percentage-of-ad-spend pricing on small accounts, and any contract where the ad accounts don't live under your ownership. You should be able to fire your marketer and keep your data the same afternoon. ### Red flags on price specifically A quote far under the market data usually means junior hands or a template being resold. A quote far over it should come with enterprise-grade proof. And a marketer who won't publish or state pricing until they have "scoped your needs" on a call is running a sales process, not a rate card. The questions that expose all three are in [seven questions that vet any marketer](/hiring-a-marketer/questions-to-ask-before-hiring-a-marketer/). I'm Conner Crowe, a fractional marketing lead for ecommerce and service businesses. The pricing above is mine, published, and current. ## I think my agency is wasting my Google Ads budget. How do I get an independent second opinion? URL: https://connercrowe.com/hiring-a-marketer/independent-second-opinion-google-ads-account/ Published: 2026-09-08 | Updated: 2026-10-22 Answer: Get it from someone with no cut of your spend and no retainer to sell you, working from read-only access. Mine is free for any Google, Microsoft or Meta Ads account: a 5 to 7 page audit I write myself, back within 24 hours of read-only access, with no call required. The suspicion is usually right in shape and wrong in detail. Something is leaking, and the agency running the account is the last party who can tell you where, because the answer might be "less spend, less management, less of us." I'm Conner Crowe, an independent Google Ads operator in Spokane, WA, working US hours, and I write these second opinions for founders in both situations: an agency runs the account and the reports don't match sales, or you've run it yourself for years and want a thorough read before releasing more budget. Here's how to get that read from someone with no stake in the retainer. ### What "independent" means here Independent is a structural claim, not a personality one. It means the person reviewing the account doesn't get paid more if the account keeps spending, doesn't get paid more if you sign a retainer afterward, and doesn't own the thing being audited. Three tests settle it before anything else: **No downstream commission.** Ask directly whether the reviewer earns anything tied to your media spend. Rebates, tiered partner incentives, a percentage-of-spend management fee waiting on the other side of the audit. Percentage-of-spend pricing is the one to watch, because it pays the reviewer to recommend a bigger budget. **No incumbent relationship.** The agency running the account can't audit the account. Neither can the reseller who sold you the tracking, nor the platform rep whose quarterly number is your spend. A Google rep isn't an independent second opinion. That's an account-growth role, and the recommendations tab is the same job automated. **A finding they lose money on.** The useful question: what would this audit have to find for you to tell me to stop hiring people? A reviewer with no answer hasn't separated the audit from the pitch. I went independent in late 2024 for the reason underneath all three. I'm one person, so the audit and the work carry the same signature, and there's no account manager sitting between the finding and the founder. ### Read-only access, and nothing more A second opinion needs to see the account. It doesn't need to change it. Grant read access at the platform level and stop there: - **Google Ads:** read-only user access on the account itself, or manager-level read. Don't accept a manager-account link that moves your account under someone else's MCC for a diagnosis. - **Google Analytics 4 and Tag Manager:** viewer on the property, read on the container. Publish rights aren't part of a review. - **Meta:** a Business Manager partner add at analyst role. Not admin, not an asset ownership transfer. - **Microsoft Advertising:** an agency link with read permission. Conversion access is the part most founders skip, and it's the part that matters. If the reviewer can't see how conversion actions are configured, which one bidding optimizes against, and what values are being sent, the audit catches symptoms and never reaches the cause. It's also where the problems are: broken conversion tracking is the most common finding in my 2026 audits. In my Q3 2026 benchmark, 6 of 29 lead-gen Google Ads accounts I manage counted a non-lead action or a click proxy as a primary conversion. In my Meta benchmark, 9 of 18 accounts, carrying 42% of the spend, optimized for no conversion event at all. Two rules hold regardless of who you pick. Nothing gets written, paused, or restructured during a diagnosis. And every grant stays revocable by you, from your own login, without asking anyone. ### No retainer required A second opinion that arrives attached to a proposal is a proposal. Say the condition out loud before the work starts: no retainer required, and the audit is complete whether or not anything gets hired afterward. That's a scope term, so put it in writing with the rest of the scope. Mine is published rather than negotiated: free for any account, no management retainer attached, and if the right call is to keep your current setup, the memo says that. How that compares with the paid market is on [the audit cost page](/google-ads-audit-cost/). Watch for the softer version of the same trap: the audit that withholds the fix. A finding described precisely enough to act on is a finding you can hand to your current agency. A finding described as "structural issues in campaign architecture" is a sales hook wearing a diagnosis. ### If you run the account yourself A self-managed account fails differently. Nobody is hiding anything. The settings have gone unquestioned since the day they were set, and the budget has grown around them. A thorough audit of a self-run account reads the conversion actions first: which ones are primary, whether calls and form fills are counted once, and what value each one sends to Smart Bidding. Then it reads the search terms against the match types and the landing pages against the ads pointing at them. What comes back is a ranked list of what to change before the next budget increase, with the dollars at stake on each item. The reviewer gets read access, you keep admin, and the account is unchanged when the memo lands. If you want help with the fixes but don't want to give up the logins, I lay out [the ways to get expert help and keep control of the account](/hiring-a-marketer/keep-control-of-google-ads-with-expert-help/), from a written plan you carry out yourself to month-to-month help. ### How to vet the auditor before you hand over access Four checks, in the order that saves the most time. **Ask for a sample report.** Not a case study, not a deck. A real audit with the client details removed. You're looking for named fixes, the click path or the change that implements each one, and a number attached: what this is costing per month, and what recovering it's worth. Anything that arrives as red, yellow, and green dots on a slide is a sales asset. **Ask what the scope excludes.** A written scope names the platforms in the review, the window of data being read, what comes back and how long it runs, the turnaround, and what isn't covered. Vague scope is how a two-hour skim gets billed as a diagnosis. **Ask how each number was measured.** A result nobody can explain the measurement of is a screenshot. This applies to the auditor's own proof as much as to your account. My standard for what counts is published in the measurement section of [my process page](/process/), and I would hold anyone you interview to something like it. **Ask what they would refuse to recommend.** A specialist has opinions about what fails. If the answer is agreement with everything you have already said, you're hiring a mirror. The [seven vetting questions](/hiring-a-marketer/questions-to-ask-before-hiring-a-marketer/) cover the rest of the interview. ### Google Partner status is optional Partner and Premier Partner badges certify spend volume across a manager account, exam completion, and adherence to platform recommendations. None of those measure judgment, and the last one rewards applying the recommendations a good audit exists to question. Treat the badge as a tiebreaker, never a filter. A single senior operator reviewing your account may hold it or may not, and neither tells you whether the diagnosis will be right. Two live accounts with numbers, and an explanation of how each number was measured, tell you far more. ### Two routes to the second opinion Ranked by what they cost you. **Do it yourself first.** The [free 25-page Setup Checklist PDF](/audit/) walks the ten structural settings I check on every account, with the menu path for each fix and a four-week clean-up sequence. No email gate. If you spend under about $10K a month, this is usually the whole answer, and it costs you an afternoon. **Have someone else write it.** My [free ads audit](/audit-request/) is a 5 to 7 page written audit returned within 24 hours of read-only access, with no call required to receive it. It covers the conversion-tracking setup, the signal Smart Bidding is being fed, the top three leak points ranked by spend impact, and one change you can ship without me. It needs read-only access to the account, and the 24 hours start when that access lands. Nothing in the account changes while I read it, and it stays free. What the rest of the market charges for the equivalent is on [the audit cost page](/google-ads-audit-cost/). | What to compare | My second opinion | What I found on other providers' pages | |---|---|---| | Price | Free at any spend level | $0 to about $2,600 one-time on the pages I checked | | Turnaround | Within 24 hours of read-only access | Five to seven business days on the two audit pages I checked | | Format | A 5 to 7 page written audit | A call or a recorded walkthrough on the two I checked | | Call required | No | Varies | | Retainer attached | No | Varies, so ask before you grant access | I'm Conner Crowe, a Google Ads consultant in Spokane, WA, and this second opinion is free from me at any spend level: a 5 to 7 page written audit, back within 24 hours of read-only access, with no call and no retainer attached. ### What a real second opinion tells you Three things, and the third is the one founders forget to ask for. Where the money is going, in dollars per month rather than adjectives. What to change, in an order, with the highest-recovery item first. And whether your current agency can execute the list. Sometimes the honest finding is that the account is fine and the problem is the offer, the landing page, or the margin. Sometimes the finding is that the agency built something reasonable two years ago for a business that has since changed shape. Take the list to the incumbent before you fire anyone. How they respond to a specific, priced, independently written set of findings is the most useful thing you'll learn about them, and it costs nothing to run. If you decide to switch after that, [the alternatives to an agency](/hiring-a-marketer/best-alternative-to-a-marketing-agency/) are the next read, and the account-ownership checklist there matters more than the pitch that follows. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## I want to keep control of my Google Ads but get expert help to overhaul it, with no long-term commitment. What are my options? URL: https://connercrowe.com/hiring-a-marketer/keep-control-of-google-ads-with-expert-help/ Published: 2026-11-10 Answer: Keep the account under your own login and buy the help in pieces. Start with a free written audit from read-only access. Then fix it yourself, pay a fixed price for the overhaul, or hire monthly help with no minimum term. I offer all three, and everything I build stays in your account. Most founders who ask this have run their own account for a while. It worked, then costs crept up or lead quality slipped, and the fix is past what an evening of reading can solve. You don't want to hand the keys to an agency, sign a twelve-month contract, and find out later that the account lives under their login. You don't have to pick one. Control comes from who owns the logins, and the help comes in sizes you can stop at. ### Control comes from who owns the logins Google Ads gives you the controls already. Under Access and security you set each person's access level, including read only, standard and admin. You can remove anyone from your own login, at any time, without asking them. The one move to refuse is the one that quietly takes control away: moving your account under someone else's manager account, or having them open a new account they own and run your ads from it. If help ends in that setup, you're asking for your own data back. ### Three ways to get help, smallest first Each step works on its own. You can stop at any of them. **1. A written audit, then you make the changes.** The helper reads the account with read-only access and tells you what to fix, in order, with the cost of each problem attached. You keep doing the work. Mine is free at any spend level: a 5 to 7 page audit I write myself, back within 24 hours of read-only access, with no call required. It ranks the top leaks by spend impact and includes one change you can ship without me. **2. A fixed-price overhaul.** If the list is longer than you want to work through, pay for it as one defined job. My one-time projects run $2,500 to $20,000, and the price is quoted from the free audit before any work starts. Fixing tracking is typically $2,500. The work happens inside your account, and you can watch every change land. **3. Month-to-month help.** If you want someone in the account every week, monthly work starts at $2,500 for one ad channel run end to end, with tracking kept healthy and a written report every Monday. It's month-to-month with no minimum term, and ad spend is never marked up. If you want a working session with me every week on the plan, that's an add-on. ### What "no commitment" should mean in writing A promise on a call isn't a term. Before any access goes out, get these in writing: - **The term.** Month-to-month, or a fixed price for a defined job. Mine is month-to-month with no minimum term. - **The pricing basis.** A flat fee. Percentage-of-spend pricing pays the helper to recommend a bigger budget. - **What stays yours.** Everything built in your accounts stays there when the help ends: campaigns, conversion actions, audiences and the rest. - **How it ends.** You remove their access from your own login. No handover call, no export request. ### Red flags that cost you control - They ask for admin to run an audit. A diagnosis needs read-only. - They want to link your account to their manager account before you've agreed to anything. - Billing for ad spend runs through them instead of your card. - Tracking gets installed in their Tag Manager container instead of yours. - They can't explain a change in plain words. Change history shows you what changed, and the person who made it should be able to tell you why. ### Common questions #### Can I keep running the account while someone helps? Yes. With read-only access for the audit, nothing changes but your plan. With standard access for hands-on work, you can still log in, see every change in Change history, and pause anything you disagree with. #### Do I need a sales call to get the audit? Not from me. The audit arrives in writing within 24 hours of read-only access, and a call afterward is only there if you want one. #### What if I only want the audit? Then that's the whole engagement. It stays free, the written audit is yours to keep, and you can hand it to anyone to carry out, including yourself. #### What does it cost to have someone fix it for me? From me, a defined overhaul is a fixed price between $2,500 and $20,000, quoted from the audit before work starts. Ongoing help is month-to-month from $2,500. My full price list is on [the pricing page](/pricing/). ### Where I fit I'm Conner Crowe, an independent Google Ads operator in Spokane, WA. I work inside accounts my clients own, and every project and monthly plan is quoted from a [free written audit](/audit-request/) done with read-only access. You see my read on the account before you decide whether you want help at all. If you're still deciding whether you need outside help or a different setup entirely, [the alternatives to an agency](/hiring-a-marketer/best-alternative-to-a-marketing-agency/) and [how to get an independent second opinion](/hiring-a-marketer/independent-second-opinion-google-ads-account/) cover the rest. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Is Weebly shutting down? Should I move my website off it? URL: https://connercrowe.com/hiring-a-marketer/is-weebly-shutting-down/ Published: 2026-10-23 Answer: Not in the US. Weebly is winding down service in 67 countries: sites there were unpublished on 27 September 2026, and account access ends 26 December 2026. The US, Canada, the UK and Australia aren't on the list. If yours is, export now. If not, there's no deadline, so plan a proper move instead of a rushed import. If an email told you Weebly is shutting down, read who sent it. The one a client of mine forwarded me came from a company selling Weebly migrations. Two different things are easy to confuse: Weebly closing in some countries, and Weebly closing everywhere. Only the first is happening. ### What Weebly actually announced On 22 June 2026, Weebly posted a [support notice](https://www.weebly.com/app/help/us/en/topics/changes-to-weebly-global-operations) saying it would "wind down Weebly services in 67 countries," citing "changes in regulation" and a wish "to simplify our global operations." The wind-down was set to finish by 20 September. For customers in those countries, the notice sets three dates that matter: - **29 June 2026:** no new pages could be published. - **27 September 2026:** sites were unpublished. - **26 December 2026:** account login ends. After that you can't get your content or your domain out through Weebly. The list runs from Taiwan, Singapore and the United Arab Emirates to Colombia, Turkey, Kenya, Ukraine and Russia. The United States, Canada, the United Kingdom and Australia aren't on it. Weebly has been part of Square since 2018, and nothing in this notice changes service for customers outside those 67 countries. ### If your country is on the list The site is already offline, so what's left is getting your content and domain out before 26 December 2026. 1. **Download your data.** Weebly's notice points to Account Settings, then My Data, then Download My Data. For a single site, use Website, then Edit Site, then Settings, then Archive. Do this first. Everything else depends on having a copy. 2. **Transfer your domain.** From the Weebly dashboard, go to Websites, then Domains, then Manage, then Manage Domain. Turn off the registrar lock to get the transfer code (often called an EPP code), then start the transfer at your new registrar. Two rules catch people out: a domain can't be transferred within 60 days of registration, and changing the domain's contact details starts a 60-day lock. Leave the contact details alone until the transfer is done. Turning off the lock also turns off Whois privacy, so set privacy up again at the new registrar. And if your domain is a country-code domain, like .sg or .co, Weebly says those transfers work differently, so contact its support before you start. 3. **Rebuild the site somewhere else** from the export. The export is a copy of your content and pages. Anything that ran on Weebly's own servers, like forms, a store checkout or members' areas, won't work until it's rebuilt on the new host. If you use Google Workspace email through Weebly, the notice asks you to contact Weebly support so the account can move directly to Google. Sort that out before you touch the domain, because email that stops arriving is harder to notice than a website that's down. ### If you're in the US, should you move anyway? There's no deadline, which is the best time to move a site. A Weebly site in the US is still working, so nothing forces your hand this month. The question is whether you want to depend on a platform whose owner is shrinking it country by country. **Move when the site is part of how you make money.** If customers call or inquire from it, if you run ads to it, or if its old pages still bring in search traffic, a forced move on someone else's 90-day notice is the worst way to do it. Doing it now, on your schedule, means you get to do it properly. **Stay a while longer when the site is a brochure.** If it's five pages, nobody finds you through it, and you don't run ads, the risk of waiting is small. Export a copy of your data once so you have one, and revisit it next year. **Move soon either way if the site is holding you back.** If you can't add the pages you need, can't install the tracking your ads depend on, or it's slow on a phone, the Weebly news is just a good reason to fix what was already broken. ### What a one-click import doesn't do The migration pitch my client received led with an automatic import that brings your pages, posts and images across. That's useful for the content. It isn't the hard part of a move. The hard part is the addresses. Every page on your old site has a URL that search engines, AI assistants, old emails and your own ads point to. If an import keeps every one of those URLs exactly as it was, there's nothing to redirect. Most moves don't: Weebly's addresses often end in .html, and a rebuild usually merges or renames pages. On the site in the example below, 78 of the 113 old addresses needed a permanent redirect. When a URL changes and nothing redirects it, the visitors it used to bring land on an error page, and the search history behind it goes with it. Someone has to decide, page by page, where each old address should now go. ### How I move a site off Weebly The order matters more than the platform you move to. 1. **Crawl the old site before building anything.** Every URL, every image, every page title, saved, so nothing gets forgotten. 2. **Make a redirect map.** One row per old address: keep it, send it to its new page with a permanent redirect, or retire it on purpose. No address is left to chance. 3. **Check the addresses that actually get traffic.** Search Console shows which old URLs have earned impressions. Test every one of them against the new site after launch, not just the ones in the map. 4. **Move only the website records.** When the domain points at the new site, leave the email records alone. A website can be rolled back in minutes. Mail that stops arriving can cost you a customer before you notice. 5. **Move the ads the same night.** If you run Google Ads, every ad and sitelink that points at an old page should point at its new page the night the site goes live. On [Robert Russell Designs](/results/robert-russell-designs-website-off-weebly/), a furniture maker in Illinois who moved off Weebly after the wind-down was announced, the redirect map ended at 113 rows once the Search Console check had added the addresses it turned up after launch. Of those 113, 112 reach a live page in one redirect or fewer. The one that doesn't was a broken video link, removed on purpose. His 22 ads and 39 sitelinks moved onto the new pages the same night. ### Before you sign up for anything Ask whoever is moving your site one question: what happens to each of my old URLs? If the answer is a redirect map you can read, row by row, they're doing the part that protects your traffic. If the answer is "the import handles it," it doesn't. Worked example: Robert Russell Designs. With Weebly winding down in 67 countries, a custom furniture maker moved off it on his own schedule: 58 pages in the sitemap, his own photographs in the galleries, 112 of 113 old addresses still resolving, and his Google Ads moved to the new pages the same night. https://connercrowe.com/results/robert-russell-designs-website-off-weebly/ ## Should I hire a freelancer, an agency, or a fractional marketer? URL: https://connercrowe.com/hiring-a-marketer/freelancer-vs-agency-vs-fractional-marketer/ Published: 2026-08-03 | Updated: 2026-10-21 Answer: Hire a freelancer for a defined single-channel task, an agency when scale needs a bench, and a fractional marketer when a founder-led business needs one senior owner across strategy, channels and tracking. In SparkToro's 2025 agency survey, 86% of agencies averaged $10,000 a month or less per retainer. My fractional work is $8,500 a month, month-to-month. The three labels get used interchangeably, and they describe three different products. Buying the wrong one costs a year. ### The three shapes, defined **A freelancer sells a skill.** One channel or one craft, billed hourly or by project: a Google Ads manager, an email designer, a Shopify developer. The good ones are excellent inside their lane. Nobody in this shape owns your marketing; you do, and you coordinate everyone in it. **An agency sells a team.** Account manager, strategist, channel specialists, creative, all behind one retainer. The structural trade: the person who sold you is rarely the person doing the work, and the retainer funds the layers between them. In exchange you get bench depth, parallel capacity, and continuity if any one person leaves. **A fractional marketer sells ownership.** One senior person who takes the whole problem: strategy, the channels, the tracking, and accountability for the number. Part-time by design, senior by definition. A fractional CMO usually directs and oversees. They set strategy and manage the team or vendors who execute it. A fractional marketing lead does both: the strategy and the execution in the same seat. For a founder-led company without an internal marketing team to direct, the lead model fits better, because there's no one underneath to hand the work to. What the seat costs is on [the pricing page](/pricing/). ### The comparison, with numbers | | Freelancer | Agency | Fractional marketer | |---|---|---|---| | What you buy | A skill | A team | An owner | | Typical price | $15 to $40 an hour for a Google Ads expert, worldwide | An average of $10,000 a month or less at 86% of agencies surveyed | About $7,500 to $12,000 a month for a fractional CMO at 15 hours a week | | Who does the work | The person you hired | Usually a coordinator or junior specialist | The person you hired | | Strategy layer | You | Included, above the execution | Included, doing the execution | | Tracking ownership | Rarely | Sometimes, in their container | The core of the engagement | | Bench if someone leaves | None | Yes | None | | Best fit | A defined task | Enterprise scale, five-plus channels | Founder-led, roughly $1M to $30M | The freelancer figure is [Upwork's typical worldwide range](https://www.upwork.com/hire/google-adwords-experts/cost/), so US rates run higher. The agency figure comes from [SparkToro and Paddy Moogan's late-2025 survey](https://sparktoro.com/blog/where-agency-revenue-is-heading-fewer-big-retainers-more-mid-market-retainers-and-longer-engagements/) of agency owners and consultants. The fractional figure is [GoFractional's October 2026 rate data](https://www.gofractional.com/insights/rates/cmo): fractional CMOs average about $173 an hour. My own numbers are on [my published pricing page](/pricing/): $2,500 a month for one ad channel, up to $8,500 for the full fractional seat, month-to-month. ### Decision rules that hold up **Buy a freelancer when the task is defined.** "Rebuild my Search campaigns" or "design a welcome flow" is freelancer work. If you can write the spec, you can hire the skill. The failure mode: handing a freelancer an undefined "grow my revenue" mandate that nobody is steering. **Buy an agency when you need parallel hands.** Multi-region, five channels needing daily attention, procurement and security review, a marketing team that already exists and needs an execution arm. At that scale the coordination the retainer funds is real work. Under that scale, you're paying for meetings. **Buy a fractional marketer when you need an owner.** Founder still making every marketing decision, revenue roughly $1M to $30M, channels underperforming because nobody senior holds them together. This is the gap the shape exists for: too big for task-by-task freelancers, too small to feed an agency's process. ### The cost math founders run wrong The sticker comparison misleads. A freelancer at $15 to $40 an hour looks cheapest, an agency retainer under $10,000 a month looks mid, and a fractional CMO at $7,500 to $12,000 a month looks expensive for one person. The real comparison is cost per senior hour spent on your account. The agency retainer buys you a few senior hours and many junior ones. Three coordinated freelancers cost three invoices plus your own hours as the project manager. The fractional model is all senior hours with no markup layer. Run the math on that axis and the ranking usually inverts. There's a second line nobody prices: the cost of the seams. ### A worked example of what the seams cost A Shopify store running Google Ads, Meta, and Klaviyo across three vendors. Each vendor competent inside their lane. The Google Ads freelancer optimizes toward a purchase conversion. The developer who last touched the theme renamed the event. The email agency has never seen the search terms report, so the flows discount products the ads are already discounting. Nobody owns measurement, because measurement isn't any one of their jobs. My own benchmarks show how often measurement goes unowned. In Q3 2026, 6 of 29 lead-gen Google Ads accounts I manage counted a non-lead action or a click proxy as a primary conversion. In my Meta benchmark, 9 of 18 accounts, carrying 42% of the spend, optimized for no conversion event at all. Every day an account runs like that, the ad platform bids on wrong data with real money. The invoices get paid on time. The work between the invoices is the problem, and no vendor is wrong to skip it, because none of them was hired to hold it. That's the argument for the ownership shape stated as a number instead of a philosophy. The real version without seams is [Emma Grace Home](https://emmagracehome.com/). I launched the Shopify store solo, with 150-plus products and every product image rendered in-house, and took it from kickoff to launch in under a month. The theme, the catalog and the imagery had one owner, so there was no vendor boundary for anything to fall through. [The build is written up here](/results/150-sku-furniture-catalog-no-photographer/). ### The failure mode of each **Freelancers fail at the seams.** Three specialists, no shared strategy, tracking nobody owns. **Agencies fail at altitude.** Strategy decks above, junior hands below, and reporting built to defend the retainer rather than to change a decision. **The one-operator model fails at scale.** No bench, one calendar, a ceiling on volume. Past roughly $30M in revenue or $50,000 a month in spend across many channels, the volume argument wins and the right move is in-house hires or an agency with real depth. The comparison table, including the rows where the agency legitimately wins, is on [my solo consultant vs agency page](/vs/agency/). I keep that page honest because I lose the enterprise rows on purpose. ### How to decide this week Write down the three things you most need done in the next ninety days. If all three are specific tasks in one channel, hire a freelancer. If all three need daily hands across five channels and a procurement process sits between you and every decision, hire an agency. If one of them is "someone should own this," you're looking for the third shape. I'm Conner Crowe, a fractional marketing lead. One senior operator across Google Ads, Meta, conversion tracking, landing pages and Shopify storefronts for ecommerce and service businesses, with ten years in and the results published as [case studies](/results/). If your business fits the freelancer or agency shape better, the sections above already told you so. If it fits this one, the engagement models and current pricing are on [my pricing page](/pricing/), and the seven questions to interrogate any of the three shapes are in [seven questions that vet any marketer](/hiring-a-marketer/questions-to-ask-before-hiring-a-marketer/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Should I pay a marketer a monthly retainer or for a one-time project? URL: https://connercrowe.com/hiring-a-marketer/retainer-vs-one-time-marketing-project/ Published: 2026-10-15 | Updated: 2026-10-21 Answer: Buy a fixed-price project when the work has a finish line, like broken tracking, a new site or a storefront; a tracking fix typically runs $2,500. Pay monthly when the work never finishes, because ad accounts need weekly changes. Monthly starts at $2,500 for one ad channel with a Monday report, month-to-month, no minimum. This one comes up on first calls, usually phrased as "wait, what was that price for?" The confusion is fair. Marketers quote both kinds of price, often in the same proposal, and rarely say which problem each one solves. ### The test: does the work have a finish line? **A project ends.** You can write down what done looks like before it starts. Tracking that records every sale. A site with every old URL redirected. A storefront live on your domain. When it ships, the invoice is over. **Monthly work doesn't end.** An ad account drifts every week: search terms change, competitors move bids, creative wears out, a platform rolls out a new default that quietly spends your money. Somebody has to look, change things and report what changed. That's ongoing labor, so it's priced by the month. If a marketer quotes a retainer for something with a finish line, ask what happens in month four. If they quote a project for something that never ends, ask who runs it after launch. ### What each price should cover **A project price should name what you get and when.** Mine are fixed prices, quoted after a free audit and agreed before work starts. Typical figures from [my pricing page](/pricing/): - Fixing tracking: typically $2,500 - A Shopify storefront: usually $5,000 to $20,000 - [A product launch](/for-ecommerce-brands/product-launch/): $8,000 - A service-business site, up to about 40 pages: from $10,000 - A 200+ page site, or a migration of several hundred URLs: $20,000 **A monthly price should name the channels and the cadence.** Mine starts at $2,500 a month for one ad channel run end to end, with tracking kept healthy and a written report every Monday. Each add-on is $1,500: a second ad platform, landing pages and site upkeep, imagery and creative, lead-quality and call scoring, or the strategy seat. Four add-ons bring it to $8,500, which is the cap: at the cap every add-on is included, and it's the full Fractional Marketing Lead seat. Ad spend is never marked up. ### When a project comes first Running ads on broken tracking means paying the platform to learn from bad data, so when the foundation is broken, a fixed-price project to fix it can come first. In practice, few of my clients started that way. Most came straight to monthly work, which already includes keeping the tracking healthy. That doesn't mean you have to buy both. Some businesses take the tracking fix and run their own ads well. Some already have clean tracking and only need the monthly work. The free audit is where I tell you which one you need, and sometimes the answer is neither. ### The term matters more than the price My work is month-to-month, with no minimum term. The reason is simple: if the work isn't paying for itself, you shouldn't be stuck paying for it. You also own every account, tag and file from day one, so leaving costs you nothing but the next invoice. Whoever you hire, ask two things before you sign. What exactly is done when this project ends? And what do I get every week for this monthly fee? If either answer is vague, the price doesn't matter yet. I'm Conner Crowe, a fractional marketing lead for ecommerce and service businesses. The full price list is on [my pricing page](/pricing/), and [the free audit](/audit-request/) tells you which kind of work you need. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## What are the best alternatives to a marketing agency? URL: https://connercrowe.com/hiring-a-marketer/best-alternative-to-a-marketing-agency/ Published: 2026-08-03 | Updated: 2026-10-22 Answer: Five options can replace an agency retainer: a fractional CMO, about $7,500 to $12,000 a month for 15 hours a week (GoFractional); specialist freelancers, $15 to $40 an hour on Upwork; an in-house specialist, about $113,000 a year with benefits; a virtual assistant; or software plus your time. Past $1M in revenue, a fractional lead is usually closest. Founders start this search after months of reports that look fine while revenue doesn't. What replaces the agency depends on what failed, and on what you can afford. ### The five alternatives, ranked for a founder-led business **1. A fractional marketing lead.** One senior marketer owns the whole system part-time: strategy, channel execution, tracking and reporting. It replaces what founders thought they were buying from the agency, the senior person from the pitch call, except that person does the work. GoFractional's rolling rate data puts fractional CMOs, as of October 2026, at an [average of about $173 an hour](https://www.gofractional.com/insights/rates/cmo), with the middle half between $125 and $200, which GoFractional converts to about $7,500 to $12,000 a month at fifteen hours a week. Operators who also run the channels hands-on often price by scope instead. It fits businesses from roughly $1M to $30M in revenue where marketing needs an owner rather than a vendor. **2. Two or three specialist freelancers.** A Google Ads freelancer and an email freelancer put expert hands on each channel for less than most retainers. Upwork's hiring data puts the [typical Google Ads expert at $15 to $40 an hour](https://www.upwork.com/hire/google-adwords-experts/cost/) worldwide, so expect US specialists to charge above that. The hidden cost is that you become the strategist and the project manager, and the gaps between freelancers have no owner. Tracking falls into those gaps first. This fits founders who like running marketing and only need execution. **3. An in-house hire.** A marketing specialist (BLS groups them with market research analysts) costs a [median $78,760 a year](https://www.bls.gov/ooh/business-and-financial/market-research-analysts.htm) and a marketing manager [$166,790](https://www.bls.gov/ooh/management/advertising-promotions-and-marketing-managers.htm), per the Bureau of Labor Statistics for May 2025. Benefits make up about 30 percent of total compensation in private industry, [per BLS employer cost data](https://www.bls.gov/news.release/ecec.nr0.htm), so budget about 1.4 times salary. A junior hire paired with a part-time senior consultant is the common middle path: the hire handles daily execution and the consultant sets direction. This fits businesses building a team on purpose, with a year to let the hire compound. **4. A virtual assistant, with the founder as strategist.** A VA handles scheduling, posting, list hygiene and reporting pulls. Upwork puts the [typical VA at $10 to $20 an hour](https://www.upwork.com/hire/virtual-assistants/cost/) worldwide. It works when the founder already knows what to do and lacks the hours. It fails when the problem is strategy, because a VA carries out instructions and doesn't write them. **5. Software plus founder time.** Smart Bidding, Shopify's native tools and Klaviyo's default flows cover the basics for tool fees only. The founder's calendar sets the ceiling. This fits pre-$1M businesses where a retainer would eat the margin it's meant to grow. ### Which alternative fits your situation | Your situation | Best fit | Why | |---|---|---| | Under $1M revenue, one channel | Software plus founder time | Fees would outrun the return | | Founder knows the plan, lacks hours | VA or freelancers | Execution is the gap | | $1M to $30M, no marketing owner | Fractional marketing lead | One owner across channels and tracking | | Building a permanent team | In-house hire plus senior consultant | The hire learns the business while the consultant steers | | Five-plus channels needing daily hands, procurement review | Stay with an agency | Volume is the real need | ### When the agency is still the right call Enterprise scale, five or more channels needing daily attention, multi-region builds, procurement and compliance review, or a CMO who needs an execution arm rather than a thinking partner all favor a team with a bench. If your business looks like that, a one-person model trades a coordination problem for a capacity problem. The full comparison, with the rows each side wins, is in [my solo consultant vs agency comparison](/vs/agency/). ### Who owns the accounts when you leave an agency? Check this before you give notice, because some of it gets harder after. Every item below should sit under a login your business controls, with an admin user on your own email: - **Google Ads.** Your business should hold admin access to the ad account itself, not only view access through the agency's manager account. History stays with the account, and Smart Bidding runs on that history. - **Meta.** The ad account, the Page, and the pixel or dataset should belong to your Business Manager, with the agency added as a partner. - **Google Analytics 4 and Google Tag Manager.** Agencies often build these inside their own accounts. Ask for administrator access on the property and the container, or for a transfer. - **Domain and hosting.** The registrar login, DNS and the Shopify or site admin. - **Email platform and list.** The Klaviyo or Mailchimp account and the subscriber list. - **Call tracking numbers.** If the agency's call-tracking account owns the numbers on your ads and site, ask about porting before you cancel. - **Creative and copy.** Read the contract for who owns the ads, images and landing pages they made. The Google Ads screen further down this page shows where to check who has access to your account. ### Switching costs to plan for **Tracking usually belongs to them too.** Tags, conversion events and feeds often live in agency-owned containers. Whoever comes next should audit and rebuild measurement under your ownership in the first month. The reference for what complete looks like is [the Tracking Stack](/frameworks/tracking-stack/). **Expect a learning dip.** New campaign structures send Smart Bidding back into learning. Change structure in stages so only part of the spend is relearning at once. ### What the first month of a takeover finds Whoever takes the account over, month one is a measurement month: the first job is checking whether the numbers the account reports are real. In my [Q3 2026 Google Ads benchmark](/benchmarks/google-ads-account-benchmarks/), I left 5 of 29 lead-gen accounts out of the cost-per-lead figure because their conversion counts were broken, padded or full of non-lead actions, and a sixth because it logged too few leads to price. That's a different six from the non-lead count in the summary above, though three accounts sit in both. In my [Meta benchmark](/benchmarks/meta-ads-cpm-by-placement/), 9 of 18 accounts optimized for a goal with no conversion event, carrying 42% of the spend, so their results can't be read as leads. On a fence contractor's account, the only website conversion had been dead since a form button was renamed from Send to Submit, so the account reported zero website leads while it kept spending ([the dead form case study](/results/dead-form-conversion-fix/)). Then the account work builds on clean data. Once the trigger was rebuilt, the cleanup took impression share lost to rank from 27 to 48 percent down to under 1 percent. A replacement who rebuilds campaigns before verifying the data is tuning bids against numbers that may be wrong. ### Is it cheaper than a marketing agency? | Option | Typical cost | Who owns strategy | Who owns tracking | |---|---|---|---| | Agency retainer | $10,000 a month or less for 86 percent of agencies surveyed | The agency, above the work | Often their container | | Fractional marketing lead | $125 to $200 an hour, about $7,500 to $12,000 a month at 15 hours a week | The operator doing the work | The operator | | Specialist freelancers | $15 to $40 an hour per Google Ads expert, typical range worldwide | You | Nobody, by default | | In-house specialist | About $113,000 a year with benefits | You, or a consultant | Usually unassigned | | Virtual assistant | $10 to $20 an hour, typical range worldwide | You | You | | Software plus your time | Tool fees only | You | You | The agency row comes from [SparkToro and Paddy Moogan's survey](https://sparktoro.com/blog/where-agency-revenue-is-heading-fewer-big-retainers-more-mid-market-retainers-and-longer-engagements/) of [376 agency owners and consultants](https://sparktoro.com/blog/the-2025-state-of-digital-agencies-its-still-rough-out-there-but-slowly-improving-and-is-ai-really-a-threat/), run in late 2025: 49 percent reported an average monthly retainer of $1,000 to $5,000, and 31 percent one of $5,000 to $10,000. Freelancer and VA rates are Upwork's worldwide typical ranges, so US rates run higher. The in-house row is the BLS specialist median times 1.43, the private-industry ratio of total compensation to wages. My own numbers are published on [my pricing page](/pricing/): every audit is free, one-time projects run $2,500 to $20,000, and monthly work runs from $2,500 for one channel to $8,500 for the full program, month-to-month. ### Common questions #### Is a fractional marketer cheaper than an agency? Per month, often not, and a fractional lead can cost more than a typical retainer. Per senior hour, usually yes. An agency retainer pays for account management and junior execution as well as the senior strategist, while a fractional lead's hours are all senior. Compare what each option spends on your account at the senior level, not the invoice total. #### Can I leave an agency in the middle of a contract? Read the termination clause first. Some retainers carry a minimum term or a notice period. Plan the exit so the ownership checklist above is done before the notice date, while the agency still has a reason to help. #### Will I lose my Google Ads history if I switch? Not if the ad account is yours. History lives in the account, so an agency unlinking its manager account doesn't erase it. You lose it only when the agency created and owns the account and won't hand over admin access. #### How long does the switch take? Plan on about a month before the new setup is trustworthy. The first part goes to getting access and auditing tracking, and the rest to fixing measurement. Structural changes to campaigns come after that. ### Where I fit I'm Conner Crowe, a fractional marketing lead for ecommerce and service businesses, and option one is the model I run: Google Ads, Meta, conversion tracking, landing pages and the Shopify storefront, one senior operator with no junior layer. Results are published in [my case studies](/results/), including [the Shopify store I launched solo for Emma Grace Home](/results/150-sku-furniture-catalog-no-photographer/). If another option fits better, use the table above to pick it. Either way, start by finding out what the current setup is leaking. [Request the free Google Ads audit](/audit-request/) and I’ll send a written read within 24 hours of read-only access. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## What happens if my one-person marketer gets sick or quits? URL: https://connercrowe.com/hiring-a-marketer/what-if-my-solo-marketer-is-unavailable/ Published: 2026-10-15 | Updated: 2026-10-21 Answer: Set up right, the account keeps running: you own every ad account and tag container, and the work is written down. Each of my clients has a written runbook, monitoring alerts keep running, and nothing changes while I'm out unless something breaks. My work is month-to-month, and a missed first-month commitment means your first month refunded in full. Someone at a medical practice put it to me plainly this year: if something happened to you, who would they call? It's the best question a buyer can ask a solo operator, and anyone who brushes it off shouldn't get the account. ### The honest trade An agency has a bench. If your account manager leaves, someone else gets assigned, usually someone who knows less about your business but can keep things running. A one-person operator has no bench. That's the real cost of the model. What you get in exchange is the person on the call being the person doing the work, with no junior taking over after the sale. The question is whether the risk can be made small enough that the trade is worth it. It can, if the setup follows a few rules. ### Five rules that make one operator safe **1. You own every account.** Google Ads, Meta, Google Analytics, Tag Manager, the call tracking and the website all sit under your business, with the marketer added as a user. If your marketer's account is the owner, you're renting your own data. **2. Every change is written down.** I send a written report every Monday that says what changed, why, and what happened. Over a year that becomes the manual for your account. Anyone you hire next can read it. **3. Campaigns run without daily hands.** A well-built account keeps running for a few weeks if nothing in the market shifts. Bidding, budgets and tracking keep working on their own. The spending keeps going. What stops is the improvement. **4. Tracking lives in your containers.** Tags, conversion goals and offline imports sit in your own Tag Manager and analytics, documented, so nothing breaks when a person changes. **5. There's no term to escape.** My work is month-to-month with no minimum. If I can't do the job, you stop paying and keep everything. Nothing about the exit depends on me. ### What happens when I'm out Each client has a written runbook for their account. Monitoring alerts keep running while I'm away, so a broken tag or a stalled conversion still raises a flag. Nothing changes in the account until I'm back unless something breaks, because a quiet week is the wrong time for a restructure. ### The first month is the riskiest, so I put it in writing A new engagement is when a solo operator's absence would hurt most, because nothing is documented yet. So the first month of my monthly work carries a written commitment. From the day I confirm working access, your written plan arrives within 7 days. Within 30 days, a test conversion shows as recorded in each ad platform I run for you, and at least one change I made is live in your ad account or on your site. If any of those is late, I refund the first month in full within 7 days, without you asking. If a step is waiting on something only you or a third party can do, like a developer publishing a tag, a platform review or your sign-off on a change, I email you that day and that step's clock pauses until it clears. It covers my work, not your ad results, which nobody can honestly promise. That commitment exists for the same reason the five rules do: you shouldn't have to take one person's reliability on faith. ### What a handoff looks like With those rules in place, there are three options if your marketer is out for good: - **Keep running.** The account can keep spending sensibly for a few weeks if nothing in the market shifts. Watch the Monday numbers and leave the structure alone. - **Pause.** You hold admin, so you can pause any campaign yourself in a minute. - **Hand it off.** Give the next person access and the written history. They start from a documented account instead of guessing. ### Questions to ask any solo marketer Ask who owns the ad accounts today. Ask to see a sample of the weekly reporting. Ask what happens to your tracking if they leave. Ask whether there's a minimum term. Four clear answers mean the risk is managed. A vague one means it isn't. I run a one-person operation by design, so I'd rather answer this question than dodge it. How an engagement runs is on [the process page](/process/), and the full comparison with agencies, including the rows where an agency wins, is on [my solo consultant vs agency page](/vs/agency/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## What questions should I ask before hiring a marketer? URL: https://connercrowe.com/hiring-a-marketer/questions-to-ask-before-hiring-a-marketer/ Published: 2026-08-03 | Updated: 2026-10-21 Answer: Ask seven questions: who touches my account, show me two live results with numbers, how you measure success, who owns my tracking, what happens when something breaks, what you'd refuse to do, and what I keep if the engagement ends. My answers are published: a written Monday report, month-to-month terms, and every account stays yours. Every marketer interviews well. The portfolio is hand-picked, the deck is polished, and the discovery call is a sales instrument. These seven questions cut under that, in order of how much they reveal. ### 1. Who personally touches my account? The single most predictive question in the set. At agencies, the senior person on the pitch call hands your account to a coordinator after signing. Ask for names: who builds the campaigns, who writes the copy, who reviews the data weekly. If the answer is a role ("your account team") instead of a person, you have your answer. The follow-up that closes the loophole: how many other accounts does that person hold? Expect a number. Mine is the 33 Google Ads accounts I manage, the count in my [Q3 2026 benchmark](/benchmarks/google-ads-account-benchmarks/). A role offered in place of a number is the warning. ### 2. Show me two live results, with numbers Not a logo wall and not "we grew a brand like yours." Two accounts, stores, or campaigns you can look at, with the before-and-after numbers and what specifically changed. Real operators keep these ready because the work produces them. A useful shape to expect: a ninety-day window, a structural change described in one sentence, and a number attached. "Account rebuilt around search intent, conversion tracking fixed: ad spend cut in half while qualified signups doubled, inside ninety days." That's one of mine, from a law firm account, and the format is the point. Change, window, number, method. Ask how each result was measured while you're at it. A result nobody can explain the measurement of is a screenshot, not a result. ### 3. How do you measure success, and how will I see it? Listen for revenue, qualified leads, cost per acquisition: business numbers. Be wary of impressions, engagement, and "brand lift" arriving in month one. Then ask what the reporting looks like and how often. A weekly written summary in plain language beats a monthly dashboard nobody opens. Mine is a written report every Monday. Push once more on the hard version: what would make you tell me [a result isn't claimable](/blog/44-agents-fact-checked-my-case-studies/)? A marketer who has never declined to claim a number hasn't thought about measurement. My own standard for what counts is published on [my process page](/process/), and I would hold any marketer to something like it, including me. ### 4. Who owns my conversion tracking? The quiet killer. Ads optimize against tracking, and when tracking is nobody's job, the account learns from wrong data for months before anyone notices. The right answer: the marketer owns pixel-to-purchase measurement personally, or names exactly who does. If they point vaguely at "your developer," the most important input to your ad spend is unowned. Here's what unowned looks like with numbers on it. On a fence contractor's account, the only website conversion had been dead since someone renamed a form button from Send to Submit, so the account reported zero website leads while campaigns kept spending. [The case study](/results/dead-form-conversion-fix/) has the fix. On an online store, a second purchase action counted alongside the main one for weeks, so about one in eight claimed purchases was a duplicate ([the numbers are here](/conversion-tracking/meta-or-performance-max-which-drove-the-sale/)). Both were invisible in reporting. Both were expensive. What a complete setup looks like is documented in the [Tracking Stack](/frameworks/tracking-stack/). ### 5. What happens when something breaks in month two? Launch month has everyone's attention. Ask about the Tuesday in month two when conversions flatline: who notices, how fast, and what the response time commitment is. Operators with monitoring name the system. Everyone else improvises, and you find problems by watching your own revenue dip. Ask for the worst example they have. A real operator has one, and the answer tells you both how bad things get and how fast they get caught. ### 6. What would you refuse to do for me? A specialist has opinions about what fails. Ask what they would push back on if you demanded it: a channel they think is wrong for you, a tactic they consider spend-burning, a timeline they call unrealistic. Someone who agrees with everything you have said is mirroring, not advising, and will keep mirroring after you have hired them. My own version of this answer is a published list of who shouldn't hire me, which sits under the prices on [my pricing page](/pricing/). ### 7. If the engagement ends, what do I keep? The clean answer: everything. Ad accounts under your ownership, admin access to your own analytics and tag manager, documentation of what was built, and a handoff without hostage negotiations. Any hesitation here, any mention of "proprietary account structures" living in their business manager, is a reason to walk regardless of how good the rest of the interview went. Accounts inside someone else's manager account can mean starting bidding history from zero when you leave, which is a switching cost nobody quotes. ### What the answers should cost Vetting is easier when you know the market. Upwork puts the [typical Google Ads expert at $15 to $40 an hour](https://www.upwork.com/hire/google-adwords-experts/cost/) worldwide, and US rates run higher. SparkToro's late-2025 survey found 86 percent of agencies average a retainer of [$10,000 a month or less](https://sparktoro.com/blog/where-agency-revenue-is-heading-fewer-big-retainers-more-mid-market-retainers-and-longer-engagements/). GoFractional's October 2026 rate data puts fractional CMOs at [about $173 an hour](https://www.gofractional.com/insights/rates/cmo), roughly $7,500 to $12,000 a month. My monthly work is $2,500 for one ad channel plus $1,500 per add-on, capped at $8,500. One-time audits run from $0 to about $2,600 on the pages I checked, mapped in [the Google Ads audit cost comparison](/google-ads-audit-cost/). A quote far under the market means junior hands. A quote far over it needs proof attached. The [full pricing breakdown](/hiring-a-marketer/how-much-does-a-fractional-marketer-cost/) has the rest. ### Checking the answers afterward Ask for one client reference and call it, with one question: "what went wrong, and how did they handle it?" Every engagement has a wrong moment, and how it was handled is the only part of the reference that predicts your experience. Then verify a number from question two against anything public: a case study page, a review, a screenshot with context. I'm Conner Crowe, a fractional marketing lead. My answers to all seven live on this site: the results in [my case studies](/results/), the measurement standard on [my process page](/process/), the pricing on [my pricing page](/pricing/), a written report every Monday, month-to-month terms, and every account stays yours. Bring the list to whoever you interview, including me. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ --- # Answers: Product Photography Hub: https://connercrowe.com/product-photography/ ## Do I need a photographer for a 100+ SKU Shopify catalog? URL: https://connercrowe.com/product-photography/product-photography-large-shopify-catalog/ Published: 2026-06-18 | Updated: 2026-10-22 Answer: Partly. Shoot the two or three definitive hero photos per flagship SKU traditionally, then render the lifestyle and multi-angle library in-house, held to a 99% fidelity bar. A full-catalog studio shoot, which runs into the mid five figures over several months and gets redone every time the catalog turns over, is the wrong tool at 100-plus SKUs. The answer splits in two. At 100-plus SKUs, a traditional photographer for the whole catalog is the wrong tool, and a photographer for the few images that carry each purchase is the right one. The mistake is treating it as all-or-nothing. ### What a full-catalog shoot costs Booking a studio to shoot every SKU sounds thorough. At scale it's a trap. A full-catalog shoot for a home or furniture brand runs into the mid five figures and ties up months of calendar. The worse part is what happens next. The catalog turns over, you add a collection, you drop a slow seller, and the whole shoot has to be redone. You pay the bill again to keep the photos matching the products you sell. That model fights the business. A growing catalog refreshes constantly. A shoot that takes months and costs the price of a car can't keep pace, so the photos drift out of date and the storefront starts showing a catalog that no longer exists. ### The split that works Hire the photographer for the work only a photographer can do. Shoot the two or three definitive hero photos per flagship SKU traditionally. The straight-on, true-color, every-detail-honest images a buyer studies before spending real money. That's where a camera and a trained eye earn their fee. Then build the rest in-house. The lifestyle library and the multi-angle packshots get rendered and held to a 99% fidelity bar, days per batch instead of weeks, regenerated on demand when a new collection lands. The render has to match the ship-to-customer product on material, color, hardware, and silhouette, but above that threshold it carries the contextual and angle layers that don't need a studio. What matters most at scale isn't any single hero image. It's a consistent, normalized set across all 100-plus SKUs, the one-shoot look, every product reading like it came from the same brand on the same day. That consistency is the thing a piecemeal approach never delivers, and it's what an in-house pipeline holds by design. [Emma Grace Home](/results/150-sku-furniture-catalog-no-photographer/), a furniture store I launched solo, sits at the far end of that split: it went fully rendered. All 150-plus products were rendered in-house with zero product shoots, under one style lock per category, so neighboring tiles on a collection page read like one shoot. Kickoff to launch took under a month. ### The real bottleneck On most large home-brand catalogs the constraint isn't the ad budget. It's the photo budget. The photo budget gates the catalog refresh, the catalog refresh gates the conversion lift, and the whole chain stalls behind a line item nobody questions. Brands keep pouring money into traffic while the pages that traffic lands on show stale or thin imagery. Breaking that bottleneck is the point of the split. Keep the photographer for the heroes, take the library in-house, and the refresh stops being a $50K event you dread once a year. It becomes a batch you run whenever the catalog changes. The broader method lives in [the product-photography library](/product-photography/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## How do I get lifestyle product photos without a studio shoot? URL: https://connercrowe.com/product-photography/lifestyle-product-photos-without-a-studio-shoot/ Published: 2026-06-18 | Updated: 2026-10-22 Answer: For a full catalog, the render pipeline wins. I stage a brand's real product photos into styled rooms, held to a 99 percent fidelity bar, batch them in days, and keep the marginal cost under four dollars a scene. The non-negotiable on any path is that the product in the scene matches what ships, or returns eat the savings. ### The three paths, and when each one fits A studio booking is one way to get lifestyle photos. It isn't the only way, and for a catalog past a hundred SKUs it's the slowest and most expensive way. Three production methods get a product into a styled room without a photographer, a set, or a shoot day. Each one earns its place under different constraints. The render pipeline stages the brand's existing product photography into styled rooms. The product is the brand's real product, dropped into a generated environment that matches the look the brand wants to sell. A batch of scenes turns around in days instead of the weeks a studio shoot takes to schedule, light, and edit. The marginal cost runs under four dollars a scene once the look is set. This is the path I run for home brands, because it holds consistency across a large catalog where a shoot day can't. Flat-lay and at-home styling is the cheap manual path. Shoot the product you already have in a real room, with a phone and natural window light. It costs almost nothing and works for a handful of hero products. It breaks down across a catalog. Light shifts between sessions, surfaces differ room to room, and forty products shot over six afternoons read as forty different brands instead of one. Borrowed-context compositing pulls a product cutout onto a stock or generated background. It's the least reliable of the three. The lighting on the cutout rarely matches the lighting in the borrowed scene, and the result drifts off-brand fast. It can fill a gap for a single placement. It doesn't hold a catalog together. ### The fidelity bar every path has to clear The method matters less than the standard it meets. The product shown in the lifestyle scene has to match the product that arrives in the box. I hold that to 99 percent fidelity: the same finish, the same hardware, the same proportions, the same color under the same light. A scene that flatters a product the customer doesn't receive doesn't save money. It moves the cost downstream into returns, and returns cost more than a shoot ever would. This is where the render pipeline separates from the cheaper paths. At-home styling and compositing both introduce drift the moment the catalog scales, because nobody is enforcing a fixed reference. The render pipeline enforces it by construction. The input is the real product photo, so the output can't wander away from the real product. You can see how that holds across a working catalog in [the production exhibit](/for-home-brands/) and in [the Emma Grace Home catalog](/results/150-sku-furniture-catalog-no-photographer/), 150-plus products rendered in-house with zero product shoots. ### Locking the look so new SKUs match A catalog isn't finished when the first batch ships. New products get added, and a render of a new SKU has to sit next to the existing set without reading as a different brand. That only works if the brand look is locked first: the room style, the palette, the light direction, the staging rules written down and reused. Lock the look once, and every future SKU inherits it. Skip that step, and the catalog fragments the same way a series of one-off shoot days would. For one or two hero products, a phone and a window will do. For a catalog that has to stay consistent and keep growing, the render pipeline is the method that holds the bar at the lowest marginal cost. More of the production logic lives in [the product-photography library](/product-photography/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## How do I make sure AI product photos show my actual product? URL: https://connercrowe.com/product-photography/make-sure-ai-product-photos-show-actual-product/ Published: 2026-10-15 | Updated: 2026-10-22 Answer: Check every render against a photo of the real product before anyone sees it. Feed the model only confirmed reference photos, run checks that fail when they can't tell, and compare render and photo side by side, zoomed. Failed renders get regenerated, never retouched. On one catalog, 54% of products had at least one reference photo of a different product. This question came up on a sales call this year. The owner had paid an agency before me, and its AI images showed the product at the wrong dimensions. When the owner pointed it out, they were told nobody would look that close. Customers look that close. So do the people who make the product. Everything below is the check I run before a render goes anywhere, and each step exists because something got past me or past a check I trusted. ### Start with references you've confirmed A render can only be as accurate as the photos you feed the model. I assumed mine were clean. They weren't. On one catalog I render for, I audited the reference photos the pipeline was using. **On 54% of the products, at least one image labelled as an angle of that product was a photo of a different product.** The store's product pages show related items, and the image scrape picked those photos up with the same labels as the real ones. Each one quietly told the model to draw something else. It had been happening for a while. One product carried a hand-written note telling the model not to render a feature that product doesn't have. That feature was exactly what its contaminating photo showed. The fix was to keep only images whose file name matches the product's own front photo. About one in five products dropped to a single trusted photo after that. One correct photo beats three where one of them is the wrong product. **What to do:** before any render, open every reference image for that product and confirm it's the product. If a vendor can't tell you where their references came from, they haven't checked. ### Automate what you can measure, and make it fail closed Some defects are measurable. Overall proportions, which side a feature sits on, whether something is mirrored, how many of a repeated element appear. I write those as automated checks against a spec file that holds the product's real geometry. Two lessons cost me real work here. **A check has to fail when it can't answer.** On the same catalog, an automated vision check asked whether a render had a feature most of the products don't have. It returned "no" whenever the QA model errored out. Since "no" was the passing answer for nearly every product, an overloaded model didn't stall the gate. It passed it. A render with the defect plainly visible passed five runs out of five, and every one of those answers turned out to be a server error. With retries added, the same question caught that render three times out of three. An error now counts as a failed check. **Test a new check on a known-bad image first.** The first version of one check passed the exact render it was written to catch. It was averaging in pixels from a neighbouring feature. I now keep a stress test that takes a known-good render and breaks it on purpose: mirrored, flipped, squashed, stretched, tapered, a hole filled in, a hole made too big, an empty frame, a second part in frame. Fourteen deliberate defects, and each one has to trip the right check. I re-ran it on 15 October 2026: | Defect introduced on purpose | What the gate has to do | Result | |---|---|---| | None (the clean render) | Pass it | Passed | | Image mirrored | Catch the feature on the wrong side | Caught | | Mounting slot flipped | Catch the slot direction | Caught | | Part squashed | Catch the length-to-width ratio | Caught | | Part stretched | Catch the length-to-width ratio | Caught | | Part tapered | Catch the width changing along its length | Caught | | Hole filled in | Catch hole size and shape | Caught | | Hole made too big | Catch hole size and shape | Caught | | Empty frame | Refuse to grade nothing | Caught | | Low-contrast background | Still find the part | Found it | | Second part in frame | Still find the part | Found it | | Background the same color as the part | Fail rather than pass blind | Failed it | | A real earlier render with the slot on the wrong side | Catch it | Caught | | Wide scene shot from an angle | Hand angle-dependent checks to a person | Handed off | All 14 behaved correctly. The last row matters as much as the others. In a scene shot from an angle, perspective makes outline measurements unreliable, and on its first run the gate confidently reported a slot on the wrong edge purely because of the camera. A check that's confidently wrong is worse than one that admits it can't tell. On the catalog with the contaminated references, the three vision checks that run on every render each take three votes, fail closed, and were validated 15 out of 15 against a labelled test set that includes a known-bad image. ### The side-by-side is the part that can't be skipped Automated checks measure. They can't judge shape. On a small hardware part, a render passed all three of the automated checks I had at the time and went live. It was wrong in three places none of them looked at: a curved lip rendered flat, a mounting hole drawn too large and too low, and a fastener standing proud of the surface instead of seated in it. Later that day a mounting slot came out flipped. That one wasn't cosmetic. With the slot upside down, the part can't hang. Every one of those was obvious the moment the render sat next to a real photo at the same scale. I tried to automate the curved-lip check anyway, and the numbers are why I stopped. Measured from the outline, the good render scored 0.05, the rejected render scored -0.01, and **a photo of the real product scored 0.01.** The real thing looked like the bad render, because the lip points toward the camera and barely shows in an outline. That check would have failed the actual product, so I didn't ship it. The gate prints its own blind spots on every run instead, and the side-by-side stays mandatory. How I run the side-by-side: - Render and real photo on one sheet, at matched scale, with gridlines. - Zoom in. On another job a render added a third wall light that the source scene doesn't have. At thumbnail size it was invisible. A cropped, upscaled band with gridlines at 25%, 50% and 75% showed it immediately. - Count discrete things out loud: holes, handles, drawers, lights. Judging the overall look is how extra parts get through. ### Checks I built and threw out Most of what I've built to check renders didn't survive testing. I'm publishing the failures because they're the reason the side-by-side stays manual. | Check | What it was meant to catch | Why I pulled it | |---|---|---| | Lip profile from the outline | A curved lip rendered flat | The real product scored 0.01, the rejected render -0.01, the good render 0.05. It would have failed the real product. | | Camera angle | Renders shot from the wrong angle | Two of five test images graded wrong and a third flipped between runs. On one product the question compared the front with a side panel the product doesn't have, so no answer could be right. | | Color cast from the wall | Renders that came out too warm | One render read as a strong warm cast. The wall in that scene is specified as a warm taupe, so the check was measuring the room's own color. On a surface that's actually white, that render was the most neutral of the five. | | "Has this feature?" vision check, original version | A feature most products don't have | Returned a pass on every server error, so an overloaded model passed everything. Rebuilt to fail closed. | The pattern is the same each time. A number that seems to measure the defect is really measuring something else: the outline instead of the shape, a side panel that isn't there, the paint instead of the light. Compute what's genuinely measurable, and put a person on the rest. ### Know how the product is used before you put it in a scene Getting the part right isn't enough once it's in a scene. I once rendered a product in use at the wrong scale, because I had a spec for the part and had never looked at how people use it. The client's own site had real photos of every step of the job the whole time. Now I pull real in-use photos first and write scale anchors into the spec, like how big the product is next to a hand or a person. In the prompt I name the geometry with numbers ("exactly eight times longer than it is wide") instead of describing it ("slender"). Where a rule sits in the prompt matters too. On one product, four renders in a row failed the gate. Those rules lived about 20,000 characters down, in the guardrails. I moved them to the top, next to the geometry, and the next render passed clean. That's the one prompt change I've measured, so it's the one I'll vouch for. There's a limit worth knowing before you pay anyone for this. A generative model won't reliably hold a feature a quarter inch wide on a ten-inch part in a wide scene. Ten rounds of generate-and-correct kept breaking it, and each fix to one detail broke another. Renders are dependable for the studio shot of the product on its own, where the geometry can be checked. For detailed in-use images of small precise parts, the owner's own photography beats any render, and I ask for it first. ### The rules when a render fails - A failed render gets regenerated, never retouched. - Never mirror an image to fix a feature on the wrong side. Mirroring moves every other asymmetric detail too. - A render that fails doesn't go live "for now." If one is already live, it comes down. - Passing every automated check isn't a pass. The side-by-side decides. ### What to ask whoever makes your images - Can I see the side-by-side sheet for every image, render next to the real product? - Where did the reference photos come from, and who confirmed each one shows this product? - What does your automated check do when the QA model errors or can't tell? - What do you do when a render fails: regenerate, retouch, or ship it? If the answers are vague, the images will be too. For when AI imagery is worth using at all, see [whether AI product photography is good enough for a Shopify store](/product-photography/ai-product-photography-good-enough-shopify/). ## How many product photos should a product page have? URL: https://connercrowe.com/product-photography/how-many-product-photos-per-pdp/ Published: 2026-06-18 | Updated: 2026-10-22 Answer: Roughly 6 to 8 photos for a considered home purchase: one lifestyle hero, four white-background angles, and one or two detail or scale shots. Each image answers a specific buyer question about depth, edge, joinery, surface, or scale. A cheap commodity SKU needs fewer. ### The working minimum is 6 to 8 images For a considered home or furniture purchase, the floor is roughly six to eight images per product page. That breaks down as one lifestyle hero, four white-background angles, and one or two detail or scale shots. The number isn't arbitrary. Each image exists to answer a question the buyer would otherwise have to ask, and on an ecommerce page there's no one to ask. Here's what each slot earns: - **Lifestyle hero.** One scene that places the piece in a room and sets the tone. It answers the first question a buyer has, which is whether the thing fits the life they imagine. - **Four white-background angles.** A three-quarter hero that shows form, a straight front elevation that shows true proportion, a top-down that shows surface and layout, and an end profile that shows depth and edge. These strip out styling and let the buyer read the object itself. - **One or two detail or scale shots.** A texture close-up for the surface and material, and a hand or a known object in frame so the buyer can size the piece without doing math against a spec line. Every unanswered question is a reason to bounce, or worse, a reason to buy and return. A return on a $1,400 table costs far more than the photo that would have prevented it. ### The four angles are per-product, not a template The four-angle set is a starting checklist, not a stamp you apply blind. A feature only exists for the customer from the angle that reveals it. A hidden drawer is invisible until a shot shows it open. A cable channel on the back of a desk doesn't exist on the page until a photo finds it. A specific leg join that signals the build quality is just a claim in the copy until an angle proves it. So the rule is to shoot the angles the product has, not the angles the template lists. Walk the piece, find the features a buyer is paying for, and make sure one frame reveals each one. Sometimes that means a fifth or sixth angle. Sometimes the top-down adds nothing and the budget moves to a second detail shot. The checklist is a prompt to think, not a quota to fill. ### When fewer is the right call | Purchase | Photos | |---|---| | Considered home or furniture purchase, such as a $1,400 table | Roughly six to eight | | Commodity SKU, such as a $40 accessory | Two or three clean shots | The 6-to-8 floor is for a considered purchase, where the buyer is spending real money and weighing the decision. A $40 commodity SKU doesn't earn that production. The buyer isn't zooming into the joinery on an impulse item, so two or three clean shots carry the page. Spending studio time on the cheap SKU is waste, the same way under-shooting the expensive one is. Match the image count to the stakes. A $1,400 table needs every angle it can use. The $40 accessory doesn't. More on building the full set the right way is in [the product-photography library](/product-photography/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## How much does ecommerce product photography cost in 2026? URL: https://connercrowe.com/product-photography/ecommerce-product-photography-cost-2026/ Published: 2026-06-18 | Updated: 2026-10-22 Answer: In the 2026 quotes I’ve seen, traditional studio product photography runs $25 to $150 per image. A full lifestyle shoot for a 150-SKU furniture catalog quotes in the mid five figures over a multi-month timeline. An in-house render pipeline now produces lifestyle scenes held to a 99 percent product fidelity bar for under $4 per finished scene, in days. ### The numbers, by photography type | Type | Cost, going by the quotes I’ve seen | | --- | --- | | Traditional studio, per finished image | $25 to $150 | | Full lifestyle shoot, 150-SKU furniture catalog | Mid five figures, with a multi-month timeline | | My in-house render pipeline, per finished scene | Under $4 marginal cost, in days | Traditional studio product photography in 2026 runs $25 to $150 per finished image, going by the quotes I've seen. The spread depends on retouching depth, the number of angles, and whether a stylist is on set. Freelancers quote lower, but the consistency drops with them. Lighting shifts between shoots, white balance drifts, and a 300-SKU catalog ends up looking like it was shot by four different people across six months. Lifestyle photography is the expensive line item. A full lifestyle shoot for a 150-SKU furniture catalog quotes in the mid five figures, and the bid comes with a multi-month timeline. That covers set builds, props, location or studio rental, a crew, and post. For a furniture brand on Shopify launching a season, the dollar figure stings less than the calendar. The catalog can't go live until the photography lands. The 2026 shift is the render pipeline. An in-house pipeline produces lifestyle scenes held to a 99 percent product fidelity bar for a marginal cost under $4 per finished scene, in days instead of months. The product geometry, material, and finish hold true. The room around it gets generated. That changes the budget conversation from a five-figure shoot to a per-scene unit cost that rounds to nothing. [Emma Grace Home](/results/150-sku-furniture-catalog-no-photographer/), a furniture store I designed, sits at the render-only end of that range: all 150-plus products rendered in-house, zero product shoots, launched in under a month. ### The real cost question is per-catalog, not per-image Per-image pricing is the wrong frame. A $40 image looks cheap until you multiply it across 150 SKUs at four angles plus a lifestyle scene each. The catalog is the unit that matters, and the timeline is the cost most founders underprice. The studio path's delay is usually the disqualifier, even when the dollar cost is equal. A multi-month shoot means the catalog is dark for a quarter. Products that could be selling are sitting in a queue waiting for a set to get built. I have watched a brand miss a full selling season because the lifestyle photography was the bottleneck, not the inventory. That lost quarter doesn't show up on the photographer's invoice. It shows up on the P&L. Hero product photography still gets shot traditionally. Two or three definitive shots per SKU, the ones a buyer zooms into on the product page, earn the studio cost. Those are the images where fidelity has to be perfect and trust is on the line. Everything past that, the lifestyle scenes that set context and fill out a collection page, is where the render pipeline replaces the five-figure shoot. The catalog-photo bottleneck I have documented for [home brands on Shopify](/product-photography/product-photography-large-shopify-catalog/) is almost always the lifestyle layer, not the hero shots. ### How to budget it in 2026 Split the budget in two. Fund traditional photography for the hero shots, two or three per SKU, and price that against the $25 to $150 per image quote range. Fund a render pipeline for the lifestyle layer, and price that at the per-scene marginal cost, which lands under $4 once the pipeline is calibrated to the product. The math favors the split on both axes. The dollar cost drops because the expensive lifestyle shoot gets replaced. The timeline collapses from weeks to days, which is the variable that decides whether a catalog ships this quarter or next. For a 150-SKU catalog, that's the difference between a mid-five-figure line item with a multi-month wait and a hero-shot budget plus a render run that finishes in days. More on the full approach lives in [the product-photography library](/product-photography/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Is AI product photography good enough for a Shopify store? URL: https://connercrowe.com/product-photography/ai-product-photography-good-enough-shopify/ Published: 2026-06-18 | Updated: 2026-10-22 Answer: Yes, above a 99% product fidelity threshold, and no below it. The render has to match the real product on material, color, hardware, stitching, and silhouette. Use AI for lifestyle and contextual scenes. Shoot the two or three definitive hero shots traditionally. Below that threshold, AI imagery costs more in returns than it earns. The honest answer has a line in it. AI product photography is good enough for a Shopify store above a fidelity threshold, and it's a liability below it. The deciding question isn't whether the image looks impressive. It's whether the rendered product matches the product that ships to the customer. ### The 99% fidelity threshold I hold AI product imagery to 99% product fidelity. That means the render matches the ship-to-customer item on material and color, hardware, stitching, silhouette, and the brand's design language. A linen-weave render of a cotton chair fails. A walnut tone rendered as oak fails. A handle in the wrong place fails. Early generative tools missed this constantly. I watched a render change the fabric weave, shift the wood tone a few shades, and slim the proportions. The catalog looked great. Then the customer opened the box, saw something different from the page they bought, and felt cheated. Returns spiked. The image earned the click and lost the order. That's the cost of being below the threshold. Not a softer brand. Real money in return shipping and a dent in trust you don't get back. ### Three controls get you there Hitting 99% isn't a prompt. It's a set of controls. The bar is enforced, not assumed: the checks fail closed, so an image that doesn't clear them doesn't ship. First, a per-SKU product spec lock. Before anything renders, I write down the material, the color hex, and the dimensions for that SKU, then check the output against it. The spec is the referee, not my eye. Second, provenance metadata on every render. Each image carries a record of what it is, which SKU it represents, and how it was made. When something looks off later, I can trace it instead of guessing. The check itself, with what it has caught on real catalogs, is in [how to make sure AI product photos show your actual product](/product-photography/make-sure-ai-product-photos-show-actual-product/). Third, a brand constitution. A short document that defines what an on-brand scene looks like and what an off-brand one looks like. It keeps a batch of fifty images reading like one brand instead of fifty moods. ### Where AI belongs, and where it doesn't AI rendering belongs on the lifestyle and contextual layer. The styled room, the morning-light kitchen, the product in a believable setting. That's where it saves the most time and where small interpretive freedom does no harm. The two or three definitive hero shots still get shot traditionally. The straight-on, true-color, every-detail-honest images a buyer studies before spending real money. Those carry the purchase decision, so they get a camera. ### Where it held up [Emma Grace Home](https://emmagracehome.com/), a premium furniture store I launched solo, sits at the render-only end of that range: it went fully rendered. All 150-plus product images came out of my in-house pipeline with zero product shoots, on statement furniture priced $999 to $7,500. The store did $154,878.84 in sales from April to September 2026, from its Shopify sales report. The [full build](/results/150-sku-furniture-catalog-no-photographer/) has the month-by-month numbers. On most catalogs I'd still put a camera on the hero shots. This is the same discipline I cover across [the product-photography library](/product-photography/), and it's the fix for [the catalog bottleneck on a 100+ SKU store](/product-photography/product-photography-large-shopify-catalog/) that stalls so many home brands. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Lifestyle photos or white-background photos: which does a product page need? URL: https://connercrowe.com/product-photography/lifestyle-vs-white-background-product-photos/ Published: 2026-06-18 | Updated: 2026-10-22 Answer: A product page needs both. White-background packshots answer the spec question, proving edge, depth, and joinery from every angle. Lifestyle scenes answer the desire question, showing the piece at home. Lifestyle earns the click, packshots close the deliberation, and skipping either one stalls the cart. The two photo types fight different battles. When a store picks one and drops the other, it loses the half of the sale that photo was built to win. So the answer to "which does my product page need" is both, and the reason is that they do separate jobs. ### Packshots answer the spec, lifestyle answers the desire White-background packshots exist to prove what the buyer is getting. Clean backdrop, no styling, no distraction. The buyer scrolls them to confirm the piece is real and to read it. Multiple angles show the edge profile, the depth, the joinery, the back panel nobody photographs and everyone worries about. A packshot is the digital version of walking around the piece in a showroom and running a hand along the seam. Lifestyle scenes do the opposite job. They put the piece in a room and let the buyer picture it in theirs. A styled shot answers the question the spec photo can't: what does this feel like in my space. For anything above roughly a $500 average order value, that lifestyle scene is the highest-leverage conversion asset on the page. It's what turns a browser into someone who wants the thing. ### Skip either one and the cart stalls Drop the packshots and you have desire with no proof. The buyer falls for the piece, then reaches for the detail that closes the decision and finds a styled photo where the spec answer should be. The doubt creeps in. The cart sits. Drop the lifestyle and a $1,200 sofa becomes a cutout floating on white. Technically accurate, emotionally dead. The buyer can't picture it at home, never builds the wanting, and never gets far enough to care about the joinery. You closed an objection nobody had reached yet. The sequence is the point. A buyer scrolls the lifestyle shot to fall for the piece, then scrolls the packshots to confirm it's real and it fits. ### How to order them on the PDP Lead with the lifestyle hero. It earns the attention and starts the wanting. Then run the packshots below it, ordered to answer questions in the sequence a careful buyer asks them: full piece, then the angles that prove depth and edge, then the construction detail. Desire first, proof second. That ordering mirrors how the decision gets made. For more on the full set, see [the product-photography library](/product-photography/). The packshot side, four white-background angles and why each one is doing work, is in [how many product photos a product page needs](/product-photography/how-many-product-photos-per-pdp/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ --- # Answers: Medical Practice Marketing Hub: https://connercrowe.com/medical-practice-marketing/ ## Can a medical practice run remarketing or Customer Match ads? URL: https://connercrowe.com/medical-practice-marketing/can-a-medical-practice-run-remarketing-or-customer-match/ Published: 2026-09-15 | Updated: 2026-10-21 Answer: Mostly no, and by two different routes. Google blocks the audiences an advertiser builds, including Customer Match and lookalikes, for advertisers promoting restricted health content. Meta's Custom Audiences Terms bar audience criteria built on health information. Search intent, geography, in-market and custom segments all remain available to a practice. I'm a marketer, not a lawyer, and nothing here is legal advice. Everything below describes published platform policy, which is a contract question rather than a legal one, and I have linked the policy text so you can read it yourself. ### The two platforms get to the same place by different routes People describe this as "health is a sensitive category" and leave it there. The two platforms don't actually work the same way, and the difference changes what you can do. **Google.** The [personalized advertising policy](https://support.google.com/adspolicy/answer/143465) says advertisers promoting products and services in sensitive interest categories can't use audiences the advertiser builds, and names Customer Match, your own data segments, audience expansion, and lookalike segments. The block is on the audience tools, applied to advertisers in the restricted categories. **Meta.** Meta took a different route. Starting [January 19, 2022](https://developers.facebook.com/blog/post/2021/12/08/updating-metas-detailed-targeting-options/), it removed detailed targeting options referencing causes, organizations or public figures that relate to health, so most condition-based interest options are gone for every advertiser rather than gated by category. On audiences, the [Custom Audiences Terms](https://www.facebook.com/legal/terms/customaudience) require that audience criteria not include "health information, financial information, consumer report information, or other categories of sensitive information." That's a data rule about what you upload, not a category ban on healthcare advertisers. Separately, the [Personal Attributes ad standard](https://transparency.meta.com/policies/ad-standards/objectionable-content/privacy-violations-personal-attributes/) governs ad copy rather than targeting: it bars an ad implying it knows your condition. "Do you have diabetes?" is out. "New diabetes treatment available" is expressly permitted. Same practical outcome for a patient list. Different rule, different wording in an appeal, and worth knowing which one you're actually arguing about. ### What "restricted health" covers on Google The policy is narrower than the shorthand, and reading it matters because a practice may sit outside it. Google's [health sub-policy](https://support.google.com/adspolicy/answer/16701855) restricts a defined list: physical or mental health conditions including diseases, sexual health and chronic conditions; products, services or procedures to treat or manage chronic conditions; health issues associated with intimate body parts or functions; invasive procedures including surgery, cosmetic surgery and injections; and disabilities. A surgical practice is squarely inside that. So is anything managing a chronic condition. General wellness, routine dental and routine vision advertising aren't on the list, and a practice in one of those categories should read the policy rather than accept a blanket "you can't do audiences." ### What it closes for a practice inside the category **Remarketing to condition-page visitors.** A list of people who read the sleep apnea page is a list of people who probably have sleep apnea. The list is the inference. **Customer Match from a patient list.** Uploading patient contact details to target or exclude them means handing the platform a set defined by being patients of a medical practice. Google signs no business associate agreement covering Google Ads, so the disclosure question arrives alongside the policy one. **Lookalikes and similar audiences**, because they're seeded from one of the two above. Two things people find counterintuitive. It applies to exclusions as well as inclusions, since excluding current patients still requires uploading who they are. And I read it as covering a list you assembled entirely legitimately, because the policy language is about the category rather than about provenance. That reading is mine rather than a quote, so check it against the text if the exclusion case matters to you. ### What's still open, including the parts usually left out The "you lose less than you think" argument only works if it's honest about what remains, and most write-ups on this understate it. **Search intent.** Somebody typing "knee surgeon near me" has declared everything you need. You're answering a question that was asked rather than inferring a condition from behavior. This is the engine for most practices. **Google's own predefined audiences.** The same health policy that blocks the ones you build explicitly preserves predefined Google audiences, in-market segments, affinity segments, demographics, life events, and custom segments. In-market and custom segments in particular are real targeting tools, they're documented as available, and a practice told "no audiences at all" was told something the policy doesn't say. **Geography.** Radius targeting per office. For a practice this is a genuinely strong lever, because a patient who won't drive to you isn't a patient regardless of intent. **Time.** Dayparting around when the phones are staffed. Worth more than most audience features, and the setting I find misconfigured most often. Serving at 8pm to a practice whose phones close at 5pm is buying voicemail. **Offline outcome import.** A different mechanism from an audience, and the one that lets bidding learn which searches become appointments. Covered in [how to track booked patients without sending PHI](/medical-practice-marketing/how-to-track-booked-patients-without-sending-phi/). ### On the risk, stated accurately I've seen this argued with a worst case presented as the expected case, so here's the actual shape. Google's [enforcement policy](https://support.google.com/adspolicy/answer/14848200) gives at least seven days' notice before account suspension for ordinary violations, with immediate suspension reserved for egregious ones. Meta's reported behavior is more often blocking a flagged audience at setup than pausing live campaigns. So the honest version isn't "you wake up on Tuesday with everything off." It's that you're running something the policy text says you can't, you'll probably be warned first, and the remedy involves an appeal and downtime you didn't schedule. That's still a bad trade for a single-specialty practice. Remarketing's upside here is modest, because the consideration window is short and the demand is already in search, and it reliably takes credit for conversions that were going to happen anyway. I wouldn't run it, and I'd want an agency proposing it to explain in writing why they read the policy differently. ### The general-service workaround, and its limits Some practices remarket only from non-condition pages: homepage, about, locations. The argument is that a homepage visit implies nothing about anyone's health. I understand it and I don't build on it. The practice is a single-specialty clinic, so everyone on its homepage is there for one category of medicine. Reaching for a page-level technicality against a category-level rule isn't a position I'd want to defend in an appeal. If a practice wants to pursue it, that's a conversation for counsel rather than a marketing judgment. ### What to check on your account today Read what audiences already exist. Practices that changed agencies inherit lists nobody remembers creating, and an old remarketing list still attached to a campaign is live exposure whether or not anyone is reviewing it. Check observation settings too. An audience attached in observation mode rather than targeting mode is still an audience the account built. Then check whether anything uploads a customer list on a schedule. Those integrations get installed once and run for years. The [free 25-page Setup Checklist PDF](/audit/) covers the account-level version of this pass, or [request a free ads audit](/audit-request/) and I’ll read the account myself. The PDF isn't healthcare-specific, and the audience and conversion sections apply to a practice unchanged. Worked example: The Hand and Wrist Institute. A surgical practice site rebuilt off WordPress and its Google Ads account moved onto it in the same week: 533 legacy URLs inventoried, 40 of 40 sampled redirects correct, 276 live ad changes. https://connercrowe.com/results/hand-wrist-institute-site-and-ads-cutover/ ## Can a medical practice use Google Ads conversion tracking without violating HIPAA? URL: https://connercrowe.com/medical-practice-marketing/can-a-medical-practice-use-google-ads-conversion-tracking/ Published: 2026-09-15 | Updated: 2026-10-22 Answer: Yes, with three changes to the default install: vendors receiving a patient identifier sign a business associate agreement, nothing sent out says why someone sought care, and health audiences stay off. Google signs no BAA for Ads, so send a click id and check that no redirect strips it on the way to your page. I'm a marketer, not a lawyer, and nothing here is legal advice. What your practice is permitted to disclose is a decision for your own counsel or compliance officer. This page is written to give that conversation something specific to start from, because "is Google Ads HIPAA compliant" isn't a question anyone can answer as asked. ### The question is wrong, and the wrong version is what gets practices in trouble There's no such thing as a HIPAA-compliant advertising platform. HIPAA regulates covered entities and their business associates. It doesn't certify software. Google isn't a covered entity, and it doesn't sign business associate agreements for Google Ads. Its own documentation says so for the adjacent measurement product: Google "does not offer Business Associate Agreements in connection with this service" ([Google Analytics help](https://support.google.com/analytics/answer/13297105)). Google does sign BAAs for Workspace and Cloud, which is why the scoping matters and why "Google signs no BAA" is too broad a thing to say. Ads isn't on the covered list. There are contractual data-processing terms, so it isn't a lawless void, but there's no business-associate obligation. So the real question is narrower: what can leave the practice, who can receive it, and under what agreement. Asked that way, conversion tracking stops being a yes-or-no and becomes a list of specific decisions. ### What makes the data protected, and where that line moved The statutory definition at [45 CFR 160.103](https://www.law.cornell.edu/cfr/text/45/160.103) is individually identifiable health information: information, including demographic information, that relates to health, health care, or payment for health care, and identifies the individual or gives a reasonable basis to identify them. So an identifier alone isn't protected. An identifier plus a relation to the seeking of care is where it gets interesting. Here's the part most write-ups on this are two years out of date on. OCR's [tracking-technologies bulletin](https://www.hhs.gov/hipaa/for-professionals/privacy/guidance/hipaa-online-tracking/index.html) of December 2022 (revised March 2024) treated an IP address combined with a visit to an unauthenticated public page about a specific condition as protected. In June 2024 a federal court vacated that specific combination in [*American Hospital Association v. Becerra*](https://www.govinfo.gov/content/pkg/USCOURTS-txnd-4_23-cv-01110/pdf/USCOURTS-txnd-4_23-cv-01110-0.pdf), and HHS dropped its appeal that August. HHS has since put a notice at the top of the bulletin naming the vacated part. The rest of the bulletin, including authenticated patient portals, still stands. That matters for how strongly anyone can argue the page-view case. It doesn't touch the case this page is actually about. Somebody who calls the practice, or submits a form, has handed you an identifier in the act of seeking care, and no court vacated that. Build for the strong version and the weak version takes care of itself. ### The three changes to a default install **Every vendor touching an identifier is one I'd expect your compliance officer to want covered.** Call tracking, the form handler, the scheduling widget, analytics, and the mailbox the lead notification lands in. Each receives a name or a number attached to somebody contacting a healthcare provider. The test isn't whether the tool is reputable. It's whether the vendor signs a business associate agreement and whether the plan you're actually paying for is the one that agreement covers, which is frequently a higher tier than the one a practice signed up on. **Nothing leaving your systems says why.** The platform receives a click id or a hashed identifier, a timestamp, and a value. Not a condition, a procedure, a department, or a page. This extends further than people expect: a conversion action named after a procedure carries the same information as a field containing it, and so does an account where the only campaign importing bookings is the one named for a single condition. **Health-based audience features stay off.** Remarketing, Customer Match and lookalikes all work by assembling a group around a behavior, and when the behavior is reading about a condition, the audience is the disclosure. That's a platform-policy question with its own answer in [whether a practice can run remarketing](/medical-practice-marketing/can-a-medical-practice-run-remarketing-or-customer-match/). ### What you can still measure Almost everything that matters, which surprises people who assumed compliance meant flying blind. You can count conversions. A call over a qualifying duration, a form submission, a booking click, all countable with no identifier reaching the platform at all. Standard conversion counting requires only the event. You can attribute those to keywords, ads and campaigns, because the click identifier does that work and a click identifier isn't a patient identifier. It's a random string Google generated when it served the ad. Check that the click id survives the trip. A redirect between the ad and your landing page can drop the `gclid` parameter while the page still loads fine, and then no booking can be tied back to its click. Open your final URL with `?gclid=test` added (or `&gclid=test` if the URL already has a `?`) and confirm it's still in the address bar once the page settles. You can import the outcome. If the practice knows which callers became patients, that can go back as an offline conversion. The version that sends the click id and no personal identifier at all is the one a practice should reach for first, and the mechanics are in [how to track booked patients without sending PHI](/medical-practice-marketing/how-to-track-booked-patients-without-sending-phi/). What you give up is the audience layer and the granular ecommerce-style reporting where the platform knows what somebody looked at. In a practice, that layer was never worth what it costs. ### The order to do it in Vendors first. Until the agreements are in place, every improvement to the measurement increases the volume moving through a tool that shouldn't have it. I've watched a practice spend a month perfecting a tracking setup while every patient enquiry routed through an unsigned vendor the whole time, and the month of work made the exposure larger rather than smaller. Then the event layer: count what happens, attribute it to the click, send nothing personal. Then the outcome import, once the practice can actually tell you which enquiries became patients. That last part is usually a front-office workflow problem rather than a technical one, and it's worth solving before it gets automated. ### Where practices get caught The tracking pixel on the confirmation page is the classic. A scheduling confirmation firing an analytics or advertising tag while the URL still carries the appointment type is sending the condition in the page path, and nobody looks at the page path. Session recording and heatmap tools are second. They're almost never covered by an agreement, they capture form contents by default, and they're usually installed by whoever built the website rather than by anyone thinking about patients. A Microsoft Ads tag can also [load Clarity session recording by itself](/blog/bing-ads-tag-loads-clarity-session-recording/), so check the ads tag too. Third is the notification email. A form handler that emails the enquiry to the front desk has transmitted the whole thing, including whatever the patient typed in the message box, through whatever mail path that vendor uses. Delivery working isn't the same as the disclosure being permitted, and the two get conflated constantly because one is visible and the other isn't. ### What to do this week Write down every tool that receives a name, an email or a phone number from somebody contacting the practice. Include the ones nobody thinks of as medical software, because those are the uncovered ones. For each, find the agreement or find out there isn't one. That list is the actual position, and it's nearly always different from what the practice believes it is. The measurement architecture underneath all of this, for any business that books by phone rather than by checkout, is [The Lead Quality Stack](/frameworks/lead-quality-stack/). To see what your site's tags already send before a visitor consents, [the five-minute browser test](/pre-consent-tracking/check-what-your-website-sends-before-consent/) shows it in your own browser. ### Common questions #### Does Google sign a BAA for Google Ads? No. Google signs business associate agreements for some Workspace and Cloud services, and Google Ads isn't one of them. #### Did the court ruling make tracking pixels safe? No. In June 2024 a federal court vacated one part of OCR's tracking guidance: an IP address combined with a visit to a public, unauthenticated page about a health condition or provider. A call or a form from someone seeking care still hands you an identifier tied to care, and the ruling didn't touch that. #### Can I still import booked patients? Yes, with a click id and no personal identifier. That's the version to build first. Worked example: The Hand and Wrist Institute. A surgical practice site rebuilt off WordPress and its Google Ads account moved onto it in the same week: 533 legacy URLs inventoried, 40 of 40 sampled redirects correct, 276 live ad changes. https://connercrowe.com/results/hand-wrist-institute-site-and-ads-cutover/ ## Can I record and score patient phone calls for marketing? URL: https://connercrowe.com/medical-practice-marketing/can-i-record-and-score-patient-calls-for-marketing/ Published: 2026-09-15 | Updated: 2026-09-15 Answer: Practices do this, and it takes two separate permissions: a business associate agreement covering the vendor that holds the recording, and consent that satisfies your state's recording law. Score the transcript rather than the audio, keep the scoring model inside your own environment, and store no identifier you do not need. I'm a marketer, not a lawyer, and nothing here is legal advice. Recording law varies by state and the consent question is genuinely a legal one. Take it to counsel before you switch anything on. ### Why a practice would want this at all Because the phone is where the booking happens, and the phone is a black box. A practice can see that a keyword produced forty calls. It can't see that eleven were records requests, nine were existing patients, six were the wrong specialty, and only fourteen were people trying to become patients. Two accounts with the same call count look identical in the reporting and are worth completely different amounts. Scoring closes that. Each call gets read against a rubric the practice wrote, and the ones that were genuine new-patient enquiries get counted as such. The bidding then learns which searches produce those and which produce noise. It's the same job as [importing booked patients](/medical-practice-marketing/how-to-track-booked-patients-without-sending-phi/), one step earlier in the funnel and available much faster, because a call can be scored the next morning while a booking may take weeks. ### Two permissions, not one These get conflated and they're unrelated. The vendor that stores the recording needs a business associate agreement, on the plan you're actually on. Recording consent is a separate question of state law, and some states require every party on the call to agree. Both are covered in [the BAA question](/medical-practice-marketing/do-i-need-a-baa-with-my-call-tracking-provider/). A practice can clear one and not the other, and clearing the vendor agreement is the one that creates false confidence. ### Score the transcript, not the audio This is the design decision that makes the whole thing tractable, and I arrived at it the hard way. Vendors implementing a covered configuration commonly restrict access to call audio while leaving the machine transcript available as ordinary text on the lead record. When I confirmed that in writing with one vendor's support team, it changed the architecture: the audio path came out entirely and the pipeline was rebuilt to score text. That turned out to be better on every axis. Text is cheaper to process, it can be scored by a model running locally, and the residual data at rest is smaller. Verify it with your own vendor in writing rather than assuming, because the exact split between what's withheld and what remains reachable differs between products and between plan tiers. ### Where the model runs matters more than which model If scoring goes to a hosted AI provider, that provider is receiving patient call content and becomes another vendor needing an agreement, on a plan that offers one. Running the model locally removes the question rather than answering it. A small open-weights model on a machine the practice or its operator controls, with constrained decoding so the output is always a valid record, handles a rubric like this without difficulty. No vendor, no API key, no per-call cost, and one fewer agreement to negotiate. Two things I wouldn't skip. First, keep a human reading the output for a long time before anything automated acts on it. A small model's stated confidence isn't reliable; I have measured one scoring correctly on nine of ten rubric cases while reporting maximum confidence on the one it got wrong. You can't build a threshold on a number like that, so the control is human review, not a cutoff. Second, run in observe mode first, and make sure observe mode genuinely writes nothing. An observation run that marks records as processed will cause the live run to skip them forever, which is a silent, permanent undercount that looks like the scoring simply found nothing. ### Store less than you think you need The instinct is to keep the reasoning so a human can audit a score. Then you look at what the log contains and it's the caller's number beside a description of their complaint, sitting in a file on a machine, which is a small clinical record you didn't mean to create. Keep the decision. Keep a keyed fingerprint instead of the raw identifier, so duplicates can still be detected without the number being readable. Drop the free-text reasoning once a human has reviewed it, or don't write it at all. Set a retention period and schedule the deletion job at the same time you schedule the scoring job, because a purge script that exists but is never scheduled is how a system quietly accumulates months of transcripts. Check the schedule, not just the script. A purge that never runs leaves a pipeline built to hold days of data holding months of it. ### Expect it to be quiet The volume surprises people. A single-specialty practice can run a meaningful ad budget and still produce a handful of scoreable paid calls a month, because most calls aren't from ads and most ad clicks don't call. Two implications. Value-based bidding needs far more volume than a practice this size generates, so the honest answer for most is that scoring informs human decisions rather than feeding an algorithm, at least for a long while. And a quiet night has to be distinguishable from a broken one. If the pipeline is silent when it works and silent when it fails, nobody finds out it stopped. Make it report that it ran and found nothing, rather than saying nothing at all. ### The order I would build it in Get the vendor agreement and the covered configuration working first, and find out what that configuration breaks. Settle the recording consent question with counsel. Then score in observe mode with a human reading every output against the rubric, and only after the rubric is signed off by whoever owns intake does anything get written back to the ad account. Most of the value arrives at the observation stage, when a practice finds out what its paid calls actually consist of. That alone usually changes the keyword list. Worked example: The Hand and Wrist Institute. A surgical practice site rebuilt off WordPress and its Google Ads account moved onto it in the same week: 533 legacy URLs inventoried, 40 of 40 sampled redirects correct, 276 live ad changes. https://connercrowe.com/results/hand-wrist-institute-site-and-ads-cutover/ ## Do Healthgrades, Zocdoc and Vitals profiles affect AI answers, and how do I claim them? URL: https://connercrowe.com/medical-practice-marketing/do-healthgrades-zocdoc-vitals-profiles-affect-ai-answers/ Published: 2026-10-27 | Updated: 2026-10-28 Answer: AI answers draw on doctor directories, so a profile with an old phone number or the wrong insurance list can end up in the answer. Healthgrades and Vitals profiles are free to claim. Zocdoc is a booking marketplace that charges per new-patient booking. Fix each provider's NPI record first, because directories compile from public records. I'm a marketer, not a lawyer, and nothing here is legal advice. Replying to reviews on these sites raises HIPAA questions, so agree the rules with your compliance officer before anyone at the practice replies. This page covers the profile facts, which hold no patient information. ### Do they affect what AI says? In my [AI answers field report](/benchmarks/ai-answers-field-report/), pooled across 9 businesses in different industries, not medical practices specifically, directories, review sites and social profiles turned up in the answers for almost every business. The most-cited domains across 8 businesses were facebook.com, google.com, linkedin.com, reddit.com, bbb.org, yelp.com, youtube.com. Doctor directories aren't on that list. The report only charts domains cited for at least 5 businesses across different fields, so no field's own directory can appear. I haven't measured a lift from claiming Healthgrades or Vitals. I'd claim them anyway, because a stale profile can feed a wrong fact. The accuracy side I can measure. Between 15% and 35% of the answers that named a business got at least one fact wrong (per-business range across 5 businesses in different industries, not medical practices specifically, in my [field report](/benchmarks/ai-answers-field-report/)). An unclaimed directory profile is a listing nobody at the practice controls. ### Start with each provider's NPI record Directories build doctor profiles from public records. The NPI registry is the record a HIPAA-covered provider must keep current: changes must be reported "within 30 days of the change" ([45 CFR 162.410](https://www.ecfr.gov/current/title-45/subtitle-A/subchapter-C/part-162/subpart-D/section-162.410), checked October 27, 2026). Updates go through NPPES, the CMS system behind the [NPI Registry](https://npiregistry.cms.hhs.gov/) (checked October 27, 2026). For each provider, check the practice location address, the phone number and the specialty taxonomy. Then the organization's own NPI, if the practice has one. Do this before the directories, so the record they compile from is already right. ### Healthgrades - **What it is:** a doctor directory with ratings, profiles listed whether or not the doctor claimed one. - **To claim:** go to [Healthgrades for Professionals](https://update.healthgrades.com/) (checked October 27, 2026), find your profile and join for free. Expect to verify that you're the provider. - **Then:** correct the practice locations, phone numbers, hours, accepted insurance and specialties, and add a photo and bio. Healthgrades says that when you update your profile, "your data syncs across all of our partner brands," so one correction travels. - **Paid options** exist. The free profile is enough to get the facts right. ### Vitals - **What it is:** a directory that says it "lists all actively licensed physicians in the United States" from public records, and that physicians "may not decline listing" ([Vitals FAQ](https://www.vitals.com/faq), checked October 27, 2026). - **To claim:** use the "claim this profile" link on your profile page, or [Vitals' claim page](https://go.vitals.com/vitals/doctor/claim/) (checked October 27, 2026). Only physicians or their staff may claim. ### Zocdoc - **What it is:** a booking marketplace, not a free directory. Zocdoc charges "a one-time booking fee" when "a new patient books through the Zocdoc Marketplace or affiliated partners," with "no upfront or subscription costs." The fee "varies by specialty and location," and existing patients don't trigger it ([Zocdoc](https://www.zocdoc.com/blog/facts/pay-per-booking-fees-explained/), published December 17, 2025, checked October 27, 2026). - **What's free:** Zocdoc says bookings through its button on your own website and through your Google Business Profile don't carry the fee. - **Is it worth it for AI answers?** I can't tell you that from data. Join Zocdoc if the per-booking cost works for your schedule. If you list there, keep the profile's locations, insurance and doctor list matching everything else. ### Insurer directories and the general listings Patients ask assistants which practices take their insurance, and insurer directories are one place that answer can come from. Check your listing in each plan you take, and in the directory of any hospital where your doctors have privileges. Your Google Business Profile, Bing Places and Apple Business Connect come next. They feed maps for every business, medical or not. Google's guideline is to "represent your business as it's consistently represented and recognized in the real world" ([Google Business Profile guidelines](https://support.google.com/business/answer/3038177), checked October 27, 2026). One name, one address and one phone per location, the same as your NPI record. ### Keep one version of every fact The free [What AI says about you](/what-ai-says-about-you/) template has a ledger with one row per fact and one column per source, including a column for your field's directory. Add a row per provider. In the shorter snapshot checks, part of the same data, pooled across 9 businesses in different industries, not medical practices specifically, 4 of 4 businesses had facts that disagreed across their own site and listings. The ledger is how you find yours. When the facts agree, ask the assistants about the practice by name and compare. The [guide to why ChatGPT doesn't name a practice](/medical-practice-marketing/why-doesnt-chatgpt-name-my-practice/) covers the questions to ask. ## Do I need a BAA with my call tracking provider? URL: https://connercrowe.com/medical-practice-marketing/do-i-need-a-baa-with-my-call-tracking-provider/ Published: 2026-09-15 | Updated: 2026-09-15 Answer: If the calls come from patients, assume yes until your compliance officer says otherwise. Call tracking receives the caller's number, the recording, and often a transcript, all attached to somebody contacting a healthcare provider. Most providers offer a business associate agreement only on a specific paid tier, so being a customer is not the same as being covered. I'm a marketer, not a lawyer, and nothing here is legal advice. Whether your practice needs a particular agreement is a decision for your counsel or compliance officer. This is what I check before wiring call tracking into a practice's ad account. ### Why call tracking is the exposure people miss A practice will interrogate its electronic records vendor for months and then install call tracking in an afternoon, because call tracking reads as a marketing tool rather than a clinical one. Look at what it receives. The caller's phone number. The time they called. Which ad and which keyword brought them. Frequently the recording, and increasingly a machine transcript of the whole conversation. A patient describing their symptoms to a receptionist, transcribed and stored on a vendor's servers, indexed and searchable. That's the combination the HIPAA definition turns on, an identifier alongside the seeking of care, and then a verbatim account of the reason. It's among the most sensitive data the practice's marketing stack touches, and it arrives through the tool nobody reviewed. ### Being a customer isn't being covered This is the part that catches practices, and it catches them quietly. Most call tracking vendors do offer a business associate agreement. Almost none of them offer it on the entry plan. It sits on a healthcare tier or an enterprise tier, usually at a meaningfully higher monthly cost, and it has to be requested, executed, and countersigned. A practice that signed up on the standard plan two years ago and assumed the vendor "does healthcare" has no executed agreement to point to. The failure mode is silent. Nothing in the product changes. The dashboard looks the same, the calls still record, no warning appears. The only way to know is to go and look for the executed agreement, and if nobody can produce it, it doesn't exist. Three things to establish, in writing: **Does this vendor sign a BAA at all.** Some won't, and that's a straight answer you can act on. **Is the plan you're on the one it covers.** Ask explicitly. "We're on plan X, does the BAA apply to plan X" is the question, not "are you HIPAA compliant." **What does enabling it change.** This is the one people skip, and it has real consequences. ### What a healthcare mode usually turns off Vendors implement their covered configuration by restricting features, and the restrictions are the part that breaks an existing setup. Expect some or all of these: personal data stripped out of notification emails, so the alert your front desk relies on arrives with the fields blank. Outbound webhooks to arbitrary URLs disabled, because an unsecured endpoint is an uncontrolled disclosure. Some integrations turned off entirely. Data retention shortened. Access to call audio restricted, sometimes while the transcript stays available through the API, sometimes not. That last distinction matters more than it sounds. A practice that planned to review recordings may find the audio withheld while the text remains reachable, or the reverse. It decides whether any downstream review or scoring is possible at all, and it's worth getting in writing from the vendor's support team before you design anything on top of it. The notification-email change is the one that reliably causes a fight two weeks after go-live. The front desk was reading the enquiry details out of that email. Now the email says a lead came in and nothing else, and someone has to log into a portal instead. Plan for it, or you'll get a complaint that reads like the tracking is broken when it's working exactly as intended. ### Recording adds a second question that has nothing to do with HIPAA Call recording is governed separately by state wiretapping law, and the rules differ by state. Some require only one party to consent. Others require all parties. A practice with two offices can sit in one regime, and a patient calling from across a state line can sit in another. The practical answer is an announcement at the start of the call, every call, both directions. That's a decision for the practice rather than for whoever configures the tool, and it should be made before recording is switched on rather than discovered afterwards. ### The second hop nobody maps The agreement with the call tracking vendor covers the call tracking vendor. It doesn't cover whatever the data flows into next. Trace it. The vendor receives the call. Who else gets it? A CRM. A spreadsheet. An email to three staff members and, in most practices I have looked at, at least one person who is a contractor rather than an employee. A reporting tool. Possibly a transcription service the vendor subcontracts to. Each hop is its own question, and the one that surprises practices is the staff list. Whether an outsourced scheduler or a virtual receptionist sits inside the practice's workforce for policy purposes is a real question with a real answer, and it's worth resolving before their address is on a notification list rather than after. ### What this means for the ad account None of this stops you running call tracking, and I wouldn't run paid search for a practice without it. In the practices I have looked at, most bookings arrive by phone rather than by form. An account that can't see calls is optimizing against the minority of its own results. What it changes is sequencing. Get the agreement executed and the covered configuration enabled before the campaigns scale, not after, because the covered configuration will break something downstream and you want to find that out at low volume. Then build the measurement on what survives. The architecture for a business that books by phone, including how the call outcome gets back to the ad platform without carrying anything it shouldn't, is [The Lead Quality Stack](/frameworks/lead-quality-stack/), and the practice-specific version is in [how to track booked patients without sending PHI](/medical-practice-marketing/how-to-track-booked-patients-without-sending-phi/). Worked example: The Hand and Wrist Institute. A surgical practice site rebuilt off WordPress and its Google Ads account moved onto it in the same week: 533 legacy URLs inventoried, 40 of 40 sampled redirects correct, 276 live ad changes. https://connercrowe.com/results/hand-wrist-institute-site-and-ads-cutover/ ## How do you track which ads produce booked patients without sending PHI? URL: https://connercrowe.com/medical-practice-marketing/how-to-track-booked-patients-without-sending-phi/ Published: 2026-09-15 | Updated: 2026-09-15 Answer: Match booked appointments back to enquiries inside your own systems, then send the platform the click id, a timestamp and a value, with no condition, procedure or department attached. The click-id path sends no personal identifier at all, which is why a practice should reach for it before the hashed-data one. I'm a marketer, not a lawyer, and nothing here is legal advice. The design below is built to narrow what your compliance officer has to rule on, not to substitute for their ruling. ### The gap this closes A practice optimizing to form fills and phone calls is optimizing to the wrong thing, and the account will get worse at the real job while its reported numbers improve. Here's the mechanism. Smart Bidding buys more of whatever you tell it counts. Tell it a phone call counts, and it will find you phone calls: records requests, existing patients rescheduling, billing questions, people who wanted a different specialty. The conversion count climbs, the cost per conversion falls, the report looks like progress, and the appointment book doesn't move. The searches that produce appointments and the searches that produce call volume aren't the same searches. Until the outcome gets back into the account, the bidding can't tell them apart. ### The shape of the fix | | Click-id import | Enhanced conversions for leads | |---|---|---| | What it sends | The click identifier, a timestamp and a value | A hashed email or phone | | Personal identifier | None at all | Hashed email or phone | | Upload deadline | 90 days after the last click | 63 days | | Where I start | The one to reach for first | Only where you cannot hold the click id through to booking | Do the matching inside your own systems, and send out only a result. **Enquiries land with an identifier and a click id.** Somebody calls or submits a form. Your call tracking or form handler holds their phone number or email, plus the click identifier that brought them. That click identifier is a random string the ad platform generated. It isn't a patient identifier. **The practice knows who booked.** Appointments exist in the practice management system. Some of those appointments correspond to enquiries from the previous step. **Match the two, on your side.** Normalize phone numbers to a single format and lowercase the emails, then join booked patients to enquiries. This is the step where all the patient data lives, and it never leaves the practice's own environment. **Send the platform the outcome and nothing else.** One conversion action, named for the event rather than the reason. "Booked new patient" is a fine name. "Booked sleep apnea consult" is a disclosure wearing a conversion action's clothes. Two mechanisms exist here and the difference matters more for a practice than for anyone else. [Offline conversion import keyed to the click id](https://support.google.com/google-ads/answer/7012522) sends the click identifier, a timestamp and a value, and **no personal identifier at all**. Enhanced conversions for leads is the one that sends a hashed email or phone. Both are widely described as "the compliant way," and for a practice they aren't equivalent: the click-id path hands the platform nothing about the person, so that's the one to reach for first. Use the hashed path only where you can't hold the click id through to booking. The platform now knows a particular click eventually became an appointment. It doesn't know the person, the condition, the procedure, or the department. ### What must not be in the payload If you're using the hashed path, the hashing isn't the protection people assume it is. Hashing an identifier stops the platform reading the identifier; it does nothing about the fields next to it. And on the click-id path there's no identifier in the payload to argue about. So: no condition, no procedure code, no department, no provider name where the provider is a single-specialty signal, no free-text notes, and no conversion action or campaign name that encodes any of those. If your account has one campaign per condition and only one of them is importing bookings, the import itself is carrying the condition. Value needs the same discipline. A revenue figure that differs per procedure can encode the procedure. This is a real tension, because value-based bidding is the reason to do any of this, and the honest answer is that most practices shouldn't attach value at all in the first months. ### Why I start at count only A scored call isn't a booked patient, and a booked patient isn't a completed episode of care. Practices routinely have no measured idea what fraction of qualified enquiries become appointments, and attaching a revenue figure before that fraction is known is inventing a number and then bidding on it. So the first phase imports bookings at count only, with no value. That alone fixes the biggest problem, which is that the account was optimizing to calls rather than appointments. Then measure the booking rate against the practice's own intake records for a few months. When you know it, and when the value you would attach can't encode the reason for the visit, value-based bidding becomes a real option. On one practice I set this up for, the pipeline ships with no value map in the code at all and no method to set one, deliberately, so that nothing can start sending a number nobody has verified. ### The failure modes worth designing for **The front desk keys a different number.** A patient calls from a mobile and gives a home number at booking. No match. This is the single largest source of loss and there's no clever fix, only matching on more than one field. **A family member books.** A spouse calls, the patient attends. The identifier that clicked and the identifier that booked are different people. **The existing patient.** An established patient rebooking isn't a new patient, and importing them as one teaches the bidding to buy people who already know you. **The click window, and the upload deadline underneath it.** Two separate ceilings, and the second one is the one that bites. The booking has to fall inside the click window your conversion action allows. On top of that Google enforces an upload deadline: [conversions uploaded more than 90 days after the last click aren't imported](https://support.google.com/google-ads/answer/15081888), and enhanced conversions for leads is tighter still at 63 days. Healthcare consideration can be long, so run the match on a schedule tied to booking date rather than enquiry date, and check both numbers before concluding the pipeline is underperforming. Every one of these makes the import undercount rather than overcount, which is the right direction for an error to run. An account that undercounts bids conservatively. An account that overcounts buys more of something that isn't happening. ### Getting the data out of the practice management system Usually the least technical and most difficult part. The reliable path is a scheduled report the system already knows how to produce, delivered to a mailbox under a signed agreement, with the minimum fields: an identifier, the booked date, a new-patient flag, and nothing clinical. Most systems can email a scheduled report. Far fewer expose a usable appointment feed, and the vendor's real-time interface is typically a paid product with a procurement timeline measured in months. One scheduling note. Since 15 June 2026 the Google Ads API no longer accepts new adopters for offline conversion imports, including enhanced conversions for leads, and Google now names the Data Manager API as the primary path. Existing integrations got a transition window. Anything built now should be built against the Data Manager API rather than a tutorial written before the change. Start the vendor conversation early and run the scheduled export in the meantime. The export isn't elegant and it works, and a working undramatic pipeline beats an elegant one that's still in procurement when the quarter ends. The general architecture for any business that closes on the phone, without the healthcare constraints, is [The Lead Quality Stack](/frameworks/lead-quality-stack/). The vendor-agreement questions that sit underneath this are in [do I need a BAA with my call tracking provider](/medical-practice-marketing/do-i-need-a-baa-with-my-call-tracking-provider/). To see what your site's tags send before a patient makes any consent choice, run [the five-minute browser test](/pre-consent-tracking/check-what-your-website-sends-before-consent/) on your homepage and your booking page. Worked example: The Hand and Wrist Institute. A surgical practice site rebuilt off WordPress and its Google Ads account moved onto it in the same week: 533 legacy URLs inventoried, 40 of 40 sampled redirects correct, 276 live ad changes. https://connercrowe.com/results/hand-wrist-institute-site-and-ads-cutover/ ## Why do my healthcare keywords and ads keep getting restricted in Google Ads? URL: https://connercrowe.com/medical-practice-marketing/why-are-my-healthcare-keywords-getting-disapproved/ Published: 2026-09-15 | Updated: 2026-10-21 Answer: Four separate systems, not one. Condition keywords usually go Eligible (limited) under the personalized ads policy and keep serving: on one practice account I restructured, 11 of 30 exact keywords needed an exemption and none was disapproved. Phone numbers in ad text are prohibited, some verticals need certification, and your state medical board regulates copy Google approves. I'm a marketer, not a lawyer, and nothing here is legal advice. The medical board section describes rules set by your state and enforced against your license, so it belongs with counsel rather than with me. ### Four separate systems, routinely confused A practice launching search campaigns hits restrictions from four directions, and treating them as one problem is why they take weeks to clear. **The personalized advertising policy.** Health is a sensitive interest category, and keywords naming conditions interact with it. **The content policy.** Unrelated to health. A phone number in ad text is the one that catches practices. **Certification gates.** Some healthcare verticals can't run at all until a third party certifies the advertiser. **State medical board advertising rules.** Not enforced by the platform, and the only one attached to your license. Different causes, different remedies, different consequences. ### Condition keywords, and what the restriction does This is the one that gets described most inaccurately. A campaign goes live and some keywords show a limited status against the personalized advertising rules. What that usually means on Search is **Eligible (limited)**, and the important part is that the keyword still serves. It's a restriction on how the ad can be personalized, not a switch that turns your traffic off. A practice told its campaign is dead because half the keywords show a policy label has usually been told something wrong. Google's [health policy](https://support.google.com/adspolicy/answer/16701855) lists remedies in order, and the first two are editing your targeting and editing your content. Appeal is on the list too, and a [keyword-level exemption mechanism](https://developers.google.com/google-ads/api/docs/policy-exemption/keywords) exists. On one account I restructured, eleven of thirty exact-match keywords came back flagged under this policy and all eleven cleared. That's one account's experience rather than a rate you should plan around, and I'd treat the flag as a queue rather than a verdict. What isn't restricted is worth saying. On the accounts I've worked, bare condition names have been the cheapest and best-converting keyword family, and the waste sat in the informational modifiers around them rather than in the condition terms. So a policy label isn't a reason to cut them. ### Phone numbers in ad text Putting the practice's phone number in a headline or description is [not allowed](https://support.google.com/adspolicy/answer/14848200), and Google's own example is "Call 1-800-123-4567" in a description. There's no exemption route for an ordinary practice number. The one documented exception is a business whose name is itself a phone number, which can request review, and there's the standard appeal path that exists for everything. Neither helps a practice with a normal number in its copy. This one wastes real time. I've seen a complete set of prepared ads, every headline and description written, where each variant carried the practice's number in the copy. Every one would have been rejected at launch. The number belongs in a call asset, which is the feature built for it, renders as a tappable button, and reports its calls as their own conversion. Google won't policy-check a payload it can't accept, so a validation pass can come back clean on a batch that would be rejected on submission. Screen your own copy for a phone-number pattern before you send it. ### Certification, which stops some practices entirely Worth knowing even if it doesn't apply to you, because it's the difference between a restriction and a wall. Google requires [certification](https://support.google.com/adspolicy/answer/176031) before certain healthcare advertisers can run: LegitScript certification for addiction treatment and for telemedicine providers, pharmacy certification for prescription drugs, and a separate certification for health insurance. Without it the ads don't serve at all, and no amount of keyword editing changes that. A general medical or surgical practice advertising its own consultations doesn't need any of these. If you're in one of the gated verticals, certification is step one and everything else on this page is step two. ### The medical board, which Google will never mention This is the one with real consequences, and the platform gives you no signal at all. State medical boards set advertising rules for licensed practitioners. What they commonly do: **Restrict superlatives and comparative claims, with a substantiation test rather than a flat ban.** California's [Business and Professions Code 651(b)(6)](https://california.public.law/codes/business_and_professions_code_section_651) makes it unlawful to claim professional superiority "unless that claim is relevant to the service being performed and can be substantiated with objective scientific evidence." So a substantiable claim can survive. An unsubstantiated "best hand surgeon in Dallas" is the problem. **Reserve board-certification language.** California limits "board certified" to ABMS-member or ACGME-accredited boards or approved equivalents and requires the full board name. Texas rule 22 TAC 164.4 limits it to ABMS, AOA BOS or ABOMS and prohibits "board eligible" and "board qualified" outright. Note this is about the certification claim rather than the word "specialist" itself; California expressly permits a physician to state that they limit their practice to specific fields. **Constrain testimonials and before-and-after imagery.** An ad reading "the best orthopedic surgeon in Phoenix" will be approved by Google in minutes. In most states it's the board that has a problem with it, unless the claim can be substantiated, and the complaint doesn't arrive through the ad account. Two things to do. Screen copy against a banned list before it's written rather than after: superlatives, "#1", "premier", "guaranteed", "safest", "fastest", and any outcome claim. And audit the assets that already exist, because they accumulate. Callouts, structured snippets, sitelinks and account-level assets written years ago by a previous agency are still serving, and an account can carry over a hundred flagged assets nobody has read since. Do that pass with the physician rather than around them. They know what their board permits, and they're the one whose license carries the risk. ### How to launch cleanly Confirm whether your vertical needs certification. Screen copy for board language and phone numbers before anything is submitted. Submit the keyword set expecting some to carry a limited label, and read that label rather than reacting to it. Then check the assets you inherited, because those serve alongside every new ad you write. The account-level version of this pass is in the [free 25-page Setup Checklist PDF](/audit/), or [request a free ads audit](/audit-request/) and I’ll read the account myself. The PDF isn't healthcare-specific, and the sections on assets, conversion actions and account hygiene apply to a practice unchanged. The compliance constraints underneath all of it are in [whether a practice can use conversion tracking at all](/medical-practice-marketing/can-a-medical-practice-use-google-ads-conversion-tracking/). Worked example: The Hand and Wrist Institute. A surgical practice site rebuilt off WordPress and its Google Ads account moved onto it in the same week: 533 legacy URLs inventoried, 40 of 40 sampled redirects correct, 276 live ad changes. https://connercrowe.com/results/hand-wrist-institute-site-and-ads-cutover/ ## Why doesn't ChatGPT name my practice? URL: https://connercrowe.com/medical-practice-marketing/why-doesnt-chatgpt-name-my-practice/ Published: 2026-10-27 | Updated: 2026-10-28 Answer: Because patients rarely type your practice's name, and when they don't, AI answers lean on doctor directories, reviews and your Google profile more than your website. A practice whose details agree everywhere, starting with its NPI record, gives the assistants something consistent to repeat. Check what they say about you by name first, then fix the sources. I'm a marketer, not a lawyer, and nothing here is legal advice. Anything that touches patient information, including how you reply to reviews, is a question for your compliance officer. This page covers what AI assistants read about a practice and what you can check without touching patient data. ### A patient who knows your name and one who doesn't A returning patient asks "What are the hours for [practice]?" A new one asks "My knee has hurt for three weeks. What should I do, and who should I see near me?" The second question is where new patients come from, and it's the one where AI answers are least likely to name you. In my [AI answers field report](/benchmarks/ai-answers-field-report/), which pools graded answers about 9 businesses in different industries, not medical practices specifically, the business was named 90.2% of the time when the question named it. On problem questions that named nobody, it was named 2.6% of the time (16 of 623 answers), and 6 of 9 businesses were never named once. ### Where a named practice comes from A symptom question can get a health explanation and "see a specialist" with no name at all, so note in your log whether an answer names any practice. When an answer does name practices, it has to get them from places that list doctors: - **Doctor directories.** Healthgrades and Vitals, which list doctors whether or not they've claimed a profile, and booking sites like Zocdoc for practices that list there. - **Your Google Business Profile** and the map listing behind it. - **Reviews** on the platforms the answers quote. - **Insurer and hospital directories** that list which doctors take which plans. Your website explains your services. The assistants use it to describe what you do. They look elsewhere to decide who to name. ### The root record: your NPI Doctor directories compile profiles from public records, and the NPI registry is the record a HIPAA-covered provider has to keep current. Under HIPAA's administrative rules, a covered provider must report changes to its NPI data "within 30 days of the change" ([45 CFR 162.410](https://www.ecfr.gov/current/title-45/subtitle-A/subchapter-C/part-162/subpart-D/section-162.410), checked October 27, 2026). The record includes the practice location address and phone number ([NPPES data fields](https://npiregistry.cms.hhs.gov/help-api/json-conversion), checked October 27, 2026). Whenever a provider's address or phone changes, check that provider's NPI record before anything else. ### What to check first I don't have a measured medical effect. The figures below are pooled across 9 businesses in different industries, not medical practices specifically. From them, I'd check these in order: 1. **Facts that agree everywhere.** In the shorter snapshot checks, 4 of 4 businesses had facts that disagreed across their own site and listings, at least 6 each. For a practice, check phone numbers, hours, accepted insurance and provider credentials first. 2. **Accuracy when patients ask by name.** Between 15% and 35% of the answers that named a business got at least one fact wrong (per-business range). A wrong phone number or the wrong insurance list reaches a patient who was about to book. 3. **Reviews on the platforms the answers quote.** 4 of 4 snapshot businesses had fewer Google reviews than the median competitor the answers named instead. That's a pattern, not a proven cause. Before you ask patients for reviews or reply to one, settle with your compliance officer what you can say without confirming someone is a patient. 4. **Claimed directory profiles.** The [medical directory guide](/medical-practice-marketing/do-healthgrades-zocdoc-vitals-profiles-affect-ai-answers/) covers claiming Healthgrades, Vitals and Zocdoc. ### What doesn't explain it **Blocked crawlers, almost never.** In the field report data, pooled across 9 businesses in different industries, not medical practices specifically, all 530 fetches I made with AI and search crawler user agents across 9 sites came back HTTP 200. **One answer isn't a pattern.** The same question asked twice gets different practices. Ask three times and count. ### Check what AI says about your practice The free [What AI says about you](/what-ai-says-about-you/) template has ten questions, a log and a fact ledger. For a practice: - Fill questions 1 to 6 with the practice name, and repeat question 3 for each location. - Write question 9 the way a new patient describes the symptom, not the way a clinician names the condition. - Add a row to the ledger for each provider: name, credentials, locations and NPI address. Ask in a private window, three times on each assistant your patients use, and keep any patient detail out of the questions. If the practice runs Google Ads, the [conversion tracking guide](/medical-practice-marketing/can-a-medical-practice-use-google-ads-conversion-tracking/) covers measuring booked patients without sending health information. --- # Answers: Law Firm Marketing Hub: https://connercrowe.com/law-firm-marketing/ ## Are Google Local Services Ads worth it for lawyers? URL: https://connercrowe.com/law-firm-marketing/are-google-local-services-ads-worth-it-for-lawyers/ Published: 2026-10-21 | Updated: 2026-10-21 Answer: They can be. Compare cost per signed case for LSA and Search. LSA charges per lead, shows a Google Verified badge, and ranks on your bid, lead likelihood and reviews. I don't run LSA for a law firm now. On the home-service LSA accounts I've worked on, the costly failures were operational, like an unverified account that stopped the ads. I'm a marketer, not a lawyer, and nothing here is legal advice. Most of what's written about Local Services Ads for lawyers predates three big changes, so check the date on anything you read, including this. ### How LSA charges a law firm Search ads charge per click. Local Services Ads charge [per lead](https://support.google.com/localservices/answer/7195435). Google counts these as leads: - an answered call - a missed call during business hours where the caller stays on the line more than 20 seconds - a message - a booking, in the US and Canada Lead prices vary by location, practice area, lead type and your bidding mode, and you control the maximum through your bid. You don't pay for people who read the ad and leave. You do pay the same for every caller who reaches you, with a case or with a question. ### What the Google Verified badge means for a lawyer The badge was called Google Screened until October 2025 and Google Verified since. For lawyers, Google [verifies the active law license](https://support.google.com/localservices/answer/9376654) of each lawyer on the profile, and it also checks the identity of owners and featured lawyers. Older guides describe a separate background check on every attorney. Google's page says it relies on the license check instead, because states already background-check lawyers before licensing. ### How LSA ranks your firm Google lists [three things](https://support.google.com/localservices/answer/7527305): your bid, how likely you are to produce a lead (which includes your responsiveness and how relevant you are to the search), and profile quality, including rating and review count. Google says missed calls can hurt your responsiveness. For most firms, the cheapest LSA improvement is answering the phone during the hours you list. ### Lead credits are automatic now Many LSA guides still describe disputing a bad lead within 30 days. That process is gone. Since mid-2024, Google [credits invalid leads automatically](https://support.google.com/localservices/answer/15100654), usually within 30 days. Google also stopped crediting leads for a job type or area you don't serve. That second change puts the burden on setup. If your profile lists a practice area you rarely take, or a wider area than you serve, you pay for those calls. Trim both before raising the budget. You can still rate each lead in the inbox. ### LSA is moving into Google Ads Google is moving Local Services Ads into Google Ads as a [pay-per-lead Performance Max campaign](https://support.google.com/google-ads/answer/17213585). The first phase began in August 2026 for select US home and storefront service businesses. Other categories follow in late 2026 and 2027, and legal isn't named in the first phase. Three things change when your account moves: - Manual maximum cost-per-lead bidding goes away. - Conversions in the new campaign are charged leads. - Old LSA performance reports don't carry over. Lead history, including call recordings, does. Export your LSA performance reports now, before the move, so you have a baseline to compare against. ### Tracking LSA calls to signed cases Calls through the LSA forwarding number are [recorded by Google](https://support.google.com/localservices/answer/6224859) in the US and Canada, and in the US you can play them in the lead inbox. I haven't found a documented way for a firm to turn that recording off, so ask Google if your consent position requires it. Lead details and a [link to each call recording](https://developers.google.com/google-ads/api/fields/latest/local_services_lead_conversation) are also available through the Google Ads API. Pull the leads, match them to your intake records by phone number and time, and mark which became signed cases. I couldn't find Google documentation for sending signed LSA cases back as conversions, so this is for your own reporting, not for training LSA's bidding. Recordings like these can be scored. On one law firm account I manage, a scoring pass sorted 236 recorded paid calls from the firm's call tracking into 67 qualified, 110 not viable and 59 follow-ups. LSA recordings can go through the same process. ### What goes wrong on the LSA accounts I've worked on I don't run Local Services Ads for a law firm right now. The LSA accounts I've worked on are home-service businesses, and the problems I found weren't about bidding. One business's LSA profile had been paused by someone else, and I turned it back on. Another had a live LSA account generating leads and billing a card that nobody on the marketing side was tracking. A third couldn't launch: its business license had expired, it didn't hold a license Google asked for, and Google flagged the same business's ad account as unverified, which stopped its ads. So on every LSA account, I check five things monthly: - **Status.** Is the profile live, and is it still verified? - **Budget.** Does the spend match what the business approved? - **Leads.** Rate them in the inbox, and note which ones became jobs. - **Expiry dates.** Licenses and insurance on file, with renewal dates on a calendar. - **Billing.** Whose card is on the account, and who sees the invoices. A law firm's version adds one more: the bar license of every lawyer on the profile. ### Running LSA and Search together LSA usually shows above the Search ads, so the same person can see both. Track them as separate sources with separate phone numbers, and don't count one caller twice. Then compare cost per signed case, not cost per lead, once you have at least 90 days of data. The [cost per signed case](/law-firm-marketing/what-should-a-law-firm-pay-per-signed-case/) answer covers the calculation. My rule of thumb: if most of your matters start with a local call and a quick decision, LSA tends to pay off, and the per-lead price is easier to budget. If your best matters are rare and high-value, Search lets you target the specific situations behind them, and LSA leads skew toward general questions. Once you have 90 days of signed cases from each, go by those numbers. Worked example: Greenacre Law. A real estate law firm’s Google Ads rebuilt around WhatConverts-qualified calls and case-type ad groups: half the ad spend, twice the monthly qualified signups, and cost per qualified lead down about half in 90 days. https://connercrowe.com/results/real-estate-law-google-ads-growth/ ## Can a law firm record and score intake calls for marketing? URL: https://connercrowe.com/law-firm-marketing/can-a-law-firm-record-and-score-intake-calls/ Published: 2026-10-21 | Updated: 2026-10-21 Answer: It depends on your states and vendors; ask counsel. Federal law needs one party's consent, but California, Florida, Pennsylvania, Washington and others require everyone's, so most firms announce recording. Vet every vendor under Rule 1.18. Then test any score: my first rubric matched a firm's intake labels on 7 of 12 calls, and the redesign matched all six clear rejections. I'm a marketer, not a lawyer, and nothing here is legal advice. Whether your firm may record a call, and with which vendors, is a decision for your ethics counsel. I run these checks before I score a firm's calls. ### Consent: one party or all parties Federal law allows a call to be recorded with the consent of one party. A group of states requires the consent of everyone on the call. The list most often cited is California, Delaware, Florida, Illinois, Maryland, Massachusetts, Montana, Nevada, New Hampshire, Pennsylvania and Washington. Connecticut requires all-party consent for recorded phone calls under its civil statute, and Michigan and Oregon are treated differently depending on the situation. Confirm against your own state's statute rather than a list, this one included. California's rule is [Penal Code section 632](https://california.public.law/codes/penal_code_section_632), which bars recording a confidential communication without everyone's consent and carries a criminal fine of up to $2,500 per violation. [Section 637.2](https://california.public.law/codes/penal_code_section_637.2) lets the person recorded sue for $5,000 per violation, and [section 632.7](https://california.public.law/codes/penal_code_section_632.7) separately covers recording calls from cell phones. Counsel should confirm how each applies to your intake line. A firm that takes calls from more than one state, which includes many firms running Google Ads, should plan for the strictest one. The usual approach is an announcement at the start of the call that it may be recorded. Ask counsel whether that announcement is enough in your states. ### What Google records without asking you Two Google features record calls on their own: - **Call ads in Google Ads.** Call recording is [on by default](https://support.google.com/google-ads/answer/16913427) for most advertisers. Google's page names healthcare and financial services as examples of exceptions and doesn't mention legal, so assume it's on until you check the setting. Callers hear an automated notice. You can turn it off under Admin, then Account settings, then Call ads. - **Local Services Ads.** Calls through the LSA forwarding number [are recorded by Google](https://support.google.com/localservices/answer/6224859) in the US and Canada. I haven't found a documented setting to turn this off. Neither of these is the firm's own recording, but both create a recording of a conversation with a prospective client. Counsel should know they exist. ### Confidentiality duties to prospective clients Consent law decides whether you may record. Confidentiality is a separate professional duty, and it covers where the recording goes. ABA Model Rule 1.6 protects information relating to a representation, and [Rule 1.18](https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_1_18_duties_of_prospective_client/) extends that duty to prospective clients, which covers most people who call intake about a possible matter. The ABA's opinions on technology are the closest guidance I've found: - **Formal Opinion 477R** (2017) asks for reasonable efforts to secure client information when using technology and vendors. - **Formal Opinion 498** (2021) covers supervising vendors in a virtual practice. - **Formal Opinion 510** (2024) advises limiting the information gathered from prospective clients to what's needed to decide whether to take the matter. - **Formal Opinion 512** (2024), on generative AI, says some tools need the client's informed consent before confidential information goes into them. The New York City Bar's [Formal Opinion 2025-6](https://www.nycbar.org/reports/formal-opinion-2025-6-ethical-issues-affecting-use-of-ai-to-record-transcribe-and-summarize-conversations-with-clients/) addresses using AI to record and transcribe client conversations directly. I haven't found an ABA opinion on sending intake data to an ad platform, so counsel has to apply the general duty. Ask three questions of every tool in the path: who stores the recording, who can listen to it, and whether any of it trains someone else's model. ### What a marketing call score measures A call score for marketing answers one question: did a prospective client with a matter the firm takes call, and did the call move toward a consultation? It isn't an assessment of the case. Keep the rubric short enough that two people would score the same call the same way. On one law firm account I manage, the first rubric counted a call as good only if a consultation was booked. Checked against calls the firm's own team had labeled by hand, it agreed on 7 of 12. The disagreements showed the firm's own "yes" also covered callers retained on the call and some warm callers who needed a follow-up. So the rubric became three outcomes: booked or retained, warm but unclosed, not viable. On the next benchmark it matched all six clear rejections. The scoring ran on one machine, and the audio never left it. Write the rubric from how intake already labels calls, then test it against those labels before any score reaches Google. And don't score with keyword searches. In one intake analysis I ran, a search of transcripts for "call back" missed more than half of the calls where the prospect was left to make the next contact. ### What goes to Google afterward The score stays with the firm. Google receives a conversion: the caller's phone number or click id for matching, the call time, and a value for the matter tier. No transcript, no notes, no description of the matter. Value only goes on qualified calls, and one caller counts once a month, so a client who calls five times doesn't look like five leads. That setup is covered step by step on [tracking Google Ads leads to signed cases](/law-firm-marketing/how-law-firms-track-google-ads-leads-to-signed-cases/). What the website's tags send before a visitor consents is a separate question, covered in [law firm tracking on practice-area and intake pages](/pre-consent-tracking/can-a-law-firm-use-the-meta-pixel-on-practice-area-pages/). ### My checklist 1. Recorded-line announcement at the start of every intake call, approved by counsel. 2. Google's own call recording reviewed: off in Google Ads if counsel prefers, and LSA recording understood. 3. Every vendor that stores or hears calls listed, with where the data lives and who can access it. 4. A scoring rubric about marketing fit, not case merit. 5. Nothing but an identifier, a time and a value sent to Google. Worked example: Greenacre Law. A real estate law firm’s Google Ads rebuilt around WhatConverts-qualified calls and case-type ad groups: half the ad spend, twice the monthly qualified signups, and cost per qualified lead down about half in 90 days. https://connercrowe.com/results/real-estate-law-google-ads-growth/ ## Do Avvo, Justia, Super Lawyers and FindLaw profiles affect AI answers, and how do I claim them? URL: https://connercrowe.com/law-firm-marketing/do-avvo-justia-super-lawyers-findlaw-profiles-affect-ai-answers/ Published: 2026-10-27 | Updated: 2026-10-28 Answer: AI answers cite directories when a client asks who to hire, so an unclaimed profile with stale facts can end up in the answer. Avvo, Justia and FindLaw let you claim and edit a free profile. Super Lawyers can't be claimed or bought; it's a peer-selected list. Fix your state bar record first, because many directories copy it. I'm a marketer, not a lawyer, and nothing here is legal advice. A directory profile is advertising under most state bar rules, so the same rules that govern your website govern what you write there. The [bar rules guide](/law-firm-marketing/what-can-lawyers-say-in-google-ads/) covers "specialist" and "expert" claims. ### Do they affect what AI says? In my [AI answers field report](/benchmarks/ai-answers-field-report/), pooled across 9 businesses in different industries, not law firms specifically, directories, review sites and social profiles turned up in the answers for almost every business. The most-cited domains, across 8 businesses, were facebook.com, google.com, linkedin.com, reddit.com, bbb.org, yelp.com, youtube.com. Legal directories don't appear on that list. The report only charts domains cited for at least 5 businesses across different fields, so no field's own directory can appear, and it publishes no rate for any field. I haven't measured a lift from claiming Avvo or Justia. I'd claim them anyway, because a stale profile can feed a wrong fact. What I can measure is the accuracy side. Between 15% and 35% of the answers that named a business got at least one fact wrong (per-business range across 5 businesses in different industries, not law firms specifically, in my [field report](/benchmarks/ai-answers-field-report/)). A directory profile you never claimed is a listing you don't control. ### Start with your state bar record Several directories build profiles from bar records. The Cornell LII directory, run with Justia, says Justia's directory "provides a free listing to every lawyer in the United States, using information compiled from bar-association directories and the like" ([LII Lawyer Directory FAQ](https://lawyers.law.cornell.edu/content/about-lii-lawyer-directory), checked October 27, 2026). Avvo says its profiles are created from public records, and asks you to enter your details "exactly as they appear on your state bar registration" ([Avvo](https://www.avvo.com/for-lawyers/avvo-profile), checked October 27, 2026). So update the bar record first: firm name, office address, phone, and the email you want public. In Washington, that's the record behind the WSBA's public lawyer directory. If a directory shows the wrong license status or admission date, the fix is usually at the bar, not the directory. ### Avvo - **What it is:** a profile for nearly every licensed attorney, built from public records whether you claim it or not. - **To claim:** search for your name on [Avvo's claim page](https://www.avvo.com/for-lawyers/avvo-profile) (checked October 27, 2026), then verify you're the licensed attorney. Avvo says claiming is free. - **Then:** fix the office address and phone, list the practice areas you take, add a photo and a short bio, and answer any reviews. Avvo shows a rating on profiles; read [Avvo's support pages](https://www.avvo.com/support/categories/attorneys) (checked October 27, 2026) on how it's calculated before you spend time chasing it. ### Justia - **What it is:** a free listing for every US lawyer from bar data, mirrored into the Cornell LII directory once claimed. - **To claim:** from your profile page, or register at [Justia's lawyer account page](https://lawyers.justia.com/account) (checked October 27, 2026), which says you can "claim and complete a professional attorney profile for free." - **Then:** complete the practice areas, office locations and bio. Justia sells paid upgrades; the free profile is enough to get your facts right. ### FindLaw - **What it is:** the FindLaw Lawyer Directory. FindLaw says most sole practitioners already have a profile ([FindLaw help](https://profileupdate.findlaw.com/help/creating-a-profile-for-an-attorney/), checked October 27, 2026). - **To claim:** register in FindLaw's Profile Update tool and claim your profile. A firm can claim the organization profile and manage its attorneys' profiles from there ([FindLaw Profile Update FAQ](https://profileupdate.findlaw.com/help/frequently-asked-questions/), checked October 27, 2026). - **What's free:** with a basic listing you can update your name, address, phone, fax, email and practice area. Narrative text and other fields need a paid enhanced profile. FindLaw says most changes show immediately, and new profiles or firm changes can take up to two business days. ### Super Lawyers - **What it is:** a rating list, not a directory you join. Candidates come from peer nominations and the publisher's own research, then go through scoring and peer review. - **You can't claim your way on.** Super Lawyers says "lawyers cannot nominate themselves" and quotes an opinion that it doesn't "permit a lawyer to buy one's way onto the list." It says 5 percent of the lawyers in a state are selected, and 2.5 percent for Rising Stars ([selection process](https://www.superlawyers.com/about/selection-process/), checked October 27, 2026). - **What you can do:** if you're selected, check that the profile's firm, address and practice areas match your bar record. Bar rules on describing ratings apply. Other legal directories also publish profiles you didn't write. Search for your name on each one the answers cite for your practice area, and claim and correct them after the four above. Then the general listings that feed maps for every business: your Google Business Profile, Bing Places and Apple Business Connect. ### Keep one version of every fact Keep the same firm name, address, phone, practice areas and attorney list on your website, your bar record, your Google profile and every directory. The free [What AI says about you](/what-ai-says-about-you/) template has a ledger with one row per fact and one column per source. Fill it in once and every disagreement shows up as a row to fix. Then ask the assistants about your firm by name and see whether the answers changed. The [guide to wrong firm details](/law-firm-marketing/why-does-chatgpt-get-my-law-firms-details-wrong/) covers tracing a wrong fact back to its source. ## How do law firms track Google Ads leads to signed cases? URL: https://connercrowe.com/law-firm-marketing/how-law-firms-track-google-ads-leads-to-signed-cases/ Published: 2026-10-21 | Updated: 2026-10-21 Answer: Capture the click id, phone and email at intake, keep them on the CRM matter, and send qualified, consulted and signed milestones to Google Ads. On one law firm account I manage, only about 15% of paid calls carried a click id, so phone matching had to cover the rest. Send a value per matter type, never matter details. I'm a marketer, not a lawyer, and nothing here is legal advice. What your firm may send to Google or a vendor is a question for your ethics counsel. Below is the tracking setup I use on law firm accounts. ### Why the call isn't the conversion Google Ads learns from whatever the firm tells it a conversion is. On most law firm accounts that's a call over a minute or a form submission. Both measure contact. Smart Bidding then buys the cheapest contact it can find, which is often a caller with a question the firm doesn't handle. The signed case happens days or weeks later, inside a practice management system Google never reads. Getting it back into Google Ads is what lets the bidding favor cases over calls. The general mechanism is on [importing closed deals as offline conversions](/lead-quality/import-closed-won-deals-google-ads-offline-conversions/). The rest of this page covers the parts specific to a firm. ### What to capture at intake Three identifiers, stored on the lead and kept with the matter when it opens: - **The Google click id (GCLID).** A hidden form field picks it up from the landing page URL. For calls, a call-tracking platform with dynamic number insertion ties the call to the click. - **The phone number**, in full international format. - **The email address**, if the prospect gives one. The click id alone isn't enough. On one call-first law firm account I manage, only about 15 percent of paid calls carried a click id. For the other calls, the phone number and call time, matched through enhanced conversions, are the route back to Google. ### Send milestones, not only the signed case | Google window | Limit | |---|---| | Conversion keyed to a click id | Up to 90 days after the click | | Enhanced conversions for leads (hashed email or phone) | 63 days | | Retract or restate: adjustments bidding uses | Uploaded within seven days | | Retract or restate: outer limit | 54 days | Google sets hard windows. A conversion keyed to a click id can be uploaded up to [90 days after the click](https://support.google.com/google-ads/answer/10029210). Enhanced conversions for leads, which match on hashed email or phone, allow 63 days. Many matters sign later than that. So send the funnel in stages: 1. **Qualified lead.** The intake team, or a call score, confirms a prospective client with a matter the firm takes. This lands within days and is what Smart Bidding should optimize toward. 2. **Consultation held.** A second action, reported but not bid on. 3. **Signed case.** Uploaded when it happens, if it's still inside the window. It's the number you report to the partners, even when it arrives too late to train the bidding. When a qualified lead turns out not to be, Google lets you [retract or restate it](https://support.google.com/google-ads/answer/7686280). Bidding uses adjustments uploaded within seven days, and the outer limit is 54 days. ### Put a value on each matter type To Google, a contract review and a lawsuit are worth the same unless the call carries a value. Give each matter type a value from the firm's own fee data, and send it with the qualified-lead conversion. Then switch the campaign to Maximize Conversion Value, and the bidding starts favoring the larger matters. On the account in the case study, the firm priced its own work in three tiers, and the scoring system wrote that price onto 132 qualified leads in the call-tracking platform in its first backfill pass. Rejected calls carried no value, and repeat callers counted once a month. Those values are the firm's own price list on scored calls, kept in the call-tracking platform. The case study doesn't claim any bidding use or result from them. If you do send values, check that they keep arriving. Compare the count of qualified leads with the count carrying a value every month, because a qualified lead with no value set reaches Google worth nothing. ### How each legal CRM sends it back What the vendor documentation shows as of October 2026: | System | Captures the click id | Sends to Google Ads | |---|---|---| | Lawmatics | [Yes, automatically](https://help.lawmatics.com/en/articles/10738670-sending-offline-conversions-to-google-ads) | Through Zapier or a webhook plus your own code, not a native send | | Clio Grow | Not documented by Clio | Zapier templates exist for Clio; check whether they reach Grow and how they handle the click id | | Filevine, Litify | No native Google Ads integration documented | Middleware or a custom build | | CallRail | Yes, through its tracking script | Native integration for calls and forms with a click id | | WhatConverts | Yes | [Sends leads](https://www.whatconverts.com/help/docs/integrations/google-ads-integration/understand-the-google-ads-integration/) marked quotable, with a sales or quote value | In practice, the call-tracking platform often does the sending, because that's where the calls and values already live. The CRM's job is to tell it which leads qualified and which signed. ### What changed in 2026 Two changes matter if someone is building this now. From June 15, 2026, the Google Ads API stopped accepting offline conversion and enhanced-conversions-for-leads uploads from developer tokens that hadn't sent any between December 2025 and June 2026, and Google [points those workflows to the Data Manager API](https://support.google.com/google-ads/answer/16884284). Google also now describes [offline conversion import as a legacy feature](https://support.google.com/google-ads/answer/7014069) and recommends enhanced conversions for leads as the starting point. Integrations that were uploading before the cutoff keep API access for now, though Google recommends moving them to the Data Manager API. A firm commissioning a new custom build should confirm the developer is using the current path. ### What never to send Prospective clients are covered by the same confidentiality duty as clients (ABA Model Rules 1.6 and 1.18). Talk to counsel before any client data leaves the firm. In the setups I build, Google receives a click id or a hashed phone or email, a timestamp, a conversion name and a number: - **No matter descriptions**, and no case types in conversion action names. "Qualified lead, tier 2" rather than "Divorce consult". - **No notes or transcripts.** The value is the only thing that needs to travel. - **No client lists for targeting** if you handle criminal defense or divorce. Google treats [criminal defense](https://support.google.com/adspolicy/answer/16701956) and [divorce](https://support.google.com/adspolicy/answer/16700849) as sensitive categories and doesn't allow audiences built from the advertiser's own lists for them. Check the current policy for any practice area that deals with hardship, such as bankruptcy. Once signed cases come back, the partners can track cost per signed case. How to set that target is on [what a law firm should pay per signed case](/law-firm-marketing/what-should-a-law-firm-pay-per-signed-case/). The same discipline applies to the website itself: what the Meta Pixel and Google tags send from practice-area and intake pages is covered in [law firm tracking on practice-area and intake pages](/pre-consent-tracking/can-a-law-firm-use-the-meta-pixel-on-practice-area-pages/). Worked example: Greenacre Law. A real estate law firm’s Google Ads rebuilt around WhatConverts-qualified calls and case-type ad groups: half the ad spend, twice the monthly qualified signups, and cost per qualified lead down about half in 90 days. https://connercrowe.com/results/real-estate-law-google-ads-growth/ ## What can lawyers say in Google Ads under bar advertising rules? URL: https://connercrowe.com/law-firm-marketing/what-can-lawyers-say-in-google-ads/ Published: 2026-10-21 | Updated: 2026-10-21 Answer: Bar rules start from one baseline: ads must be truthful and not misleading, and each state adds its own. The ABA model rule allows 'certified specialist' only with the certifier named. Counsel decides. In one law firm account I audited, 16 enabled ads still said 'Top' or 'Expert', in ended campaigns one date change from serving. I'm a marketer, not a lawyer, and nothing here is legal advice. Your state bar's rules and your ethics counsel decide what your ads may say. I built the checklist below from the rule text. States amend their rules, so check the dates on anything here. Google reviews ads against its own policies, not against your bar's rules. An ad can be approved and serve for months while breaking a bar rule. ### "Specialist", "expert" and "certified" Under [ABA Model Rule 7.2(c)](https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_7_2_advertising/), a lawyer may say they're certified as a specialist only if a state-approved or ABA-accredited body certified them, and the ad names that body. The model rule's comments allow a lawyer to say they specialize in an area if that isn't misleading. States add their own limits. California's [Rule 7.4](https://www.calbar.ca.gov/Portals/0/documents/rules/Rule_7.4-Exec_Summary-Redline.pdf) handles certification through its Board of Legal Specialization, and Florida restricts both "specialist" and "expert" further. Search ads make this easy to get wrong, because the claim usually lives in a callout or a headline nobody rereads. The audit section below shows how much of it can survive in one account. ### Case results and testimonials ABA Rule 7.1's comments warn that past results can create unjustified expectations. States go further: - **California** presumes an ad misleading when it shows case results out of context or refers to money recovered ([B&P Code 6158.1](https://california.public.law/codes/business_and_professions_code_section_6158.1)), and electronic ads that show a result must state the facts or say results depend on the facts ([6158.3](https://california.public.law/codes/business_and_professions_code_section_6158.3)). - **Florida** treats past results as deceptive unless they're objectively verifiable (Rule 4-7.13). A 30-character headline has no room for the context those rules ask for. I keep dollar results out of ads entirely and let the landing page carry them, with the required disclosures, if the firm wants them at all. ### What every ad has to include ABA Rule 7.2(d) requires the name and contact information of at least one lawyer or firm responsible for the ad's content. California added requirements on January 1, 2026. [SB 37](https://legiscan.com/CA/text/SB37/2025) amended Business and Professions Code [section 6157.2](https://california.public.law/codes/business_and_professions_code_section_6157.2) so that every ad names a responsible California lawyer or firm and the city, town or county of a real office. The law treats websites, landing pages and paid search as advertising and bars paid or membership-based awards. In some cases, after a State Bar complaint, a misled consumer can sue for $5,000 to $100,000 per ad. California also bars guarantees and "quick settlement" promises, and requires a contingency-fee statement to say whether the client pays costs when there's no recovery. Firms must keep a copy of each ad for a year ([6159.1](https://california.public.law/codes/business_and_professions_code_section_6159.1)). Check the statute text with counsel. For a Search ad, the practical fix is the landing page and the assets: the responsible lawyer's name, the office location and the required disclosures on every page an ad points to. ### States that want to see your ads Some bars require ads to be filed: - **Texas** requires filing with the [Advertising Review Committee](https://www.texasbar.com/Content/NavigationMenu/ForLawyers/AdvertisingReview/default.htm) within 10 days of first use, with a $100 State Bar fee (Rule 7.04). Website pages are exempt except the homepage (Rule 7.05). - **Florida** requires non-exempt ads to be filed at least 20 days before use, with a $250 fee (Rule 4-7.19). Websites are exempt, but [internet ads must be filed](https://www.floridabar.org/ethics/etad/advertising-filing-requirements/). **New York's** Appellate Division adopted amendments, effective June 1, 2026, that largely follow the ABA model, according to the [NYSBA summary](https://nysba.org/amendments-adopted-to-advertising-rules/), with New York differences such as keeping Rule 7.5. Guides written before the change may be out of date. Confirm the current text in the court's order. Other states have their own filing or review rules. Look up yours before launch. ### Text Google writes for you Google can generate ad text you never wrote. Automatically created assets, AI Max text customization and Performance Max can all produce headlines and descriptions from your website. A generated line that calls the firm "the best" or promises an outcome is still the firm's ad. On law firm accounts I turn automatically created assets off, and leave text customization off unless someone at the firm reviews every variant in the asset report. I also turn off auto-applied recommendations, so new ads and keywords don't go live without a review. ### What I find when I audit a law firm's ads These claims tend to survive in ads nobody looks at anymore. In October 2026 I scanned a law firm account I'd taken over: - **The stored assets.** Of 1,675 text and callout assets saved in the account, 198 (12%) used words bar rules restrict or that read as unverifiable comparisons: "top" in 98, "expert" or "experts" in 66, "expertise" in 15, "top rated" in 4, and "specialist" or "specialists" in 3. - **The enabled ads.** 16 of the 46 enabled ads in enabled ad groups carried them, across 39 headlines and descriptions shaped like "Top Rated [City] Lawyers" and "[Firm]: [Practice] Experts". - **Why nobody had noticed.** None of the 16 had served in the last 30 days. All of them sat in three campaigns whose end date had passed, so Google listed the campaigns as ended while every ad inside stayed enabled. Changing one end date would have put them back in front of searchers. - **The ads that were serving.** The 22 ads that were serving were clean: none of those words in 411 lines. I paused the 16 rather than deleting anything, and wrote compliant replacements for any campaign that comes back. Pausing doesn't retire an ad, though, since a paused ad is one click from serving again, so I list every one for the firm to review with counsel. The lesson for any audit: search the enabled ads inside ended and paused campaigns, the stored asset library, and the default text in keyword insertion. A headline written as `{KeyWord:Top [Practice] Lawyers}` prints "Top [Practice] Lawyers" whenever the searcher's keyword is too long to fit. ### Ads that cross state lines A campaign targeted to a metro area can serve across a state line. ABA [Model Rule 8.5](https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_8_5_disciplinary_authority_choice_of_law/) covers which state's rules apply. Ask counsel before targeting outside the states where the firm's lawyers are admitted, and set the location option to people in the area rather than people interested in it. ### My checklist 1. My default: no "specialist", "expert" or "certified" without the certifying body named, unless counsel clears it for your state. 2. No dollar results or outcome promises in ad text. 3. Responsible lawyer, contact details and, in California, the office location on every landing page, confirmed with counsel. 4. Automatically created assets off, and text customization off unless reviewed. 5. Filing done where your state requires it. 6. A copy of every ad kept for as long as your state requires. 7. Enabled ads inside ended or paused campaigns, the stored asset library, and keyword-insertion default text all checked for the same words. Worked example: Greenacre Law. A real estate law firm’s Google Ads rebuilt around WhatConverts-qualified calls and case-type ad groups: half the ad spend, twice the monthly qualified signups, and cost per qualified lead down about half in 90 days. https://connercrowe.com/results/real-estate-law-google-ads-growth/ ## What should a law firm pay per signed case from Google Ads? URL: https://connercrowe.com/law-firm-marketing/what-should-a-law-firm-pay-per-signed-case/ Published: 2026-10-21 | Updated: 2026-10-21 Answer: Divide paid search spend by the signed cases it produced over the same period, and set the ceiling from your own fees, since published per-case benchmarks aren't audited. Then check intake before blaming the ads: at one firm I work with, 36 of 40 booked consultations in one practice area ended with the prospect told to call the firm back. I'm a marketer, not a lawyer, and nothing here is legal or financial advice. ### The formula Cost per signed case is paid search spend divided by signed cases from paid search, for the same period. Count each case against the month of the click that produced it, not the month it signed. Otherwise a slow month of signings makes a good month of advertising look bad. This is about ad spend. Paying a vendor per signed case raises separate questions under the rules on fee sharing and paying for referrals (ABA Model Rules 5.4 and 7.2(b)), and those are for counsel. You need two numbers most firms don't keep in one place: spend, which lives in Google Ads, and signed cases by source, which lives in intake or the CRM. Most of the work is joining them. The setup is on [tracking Google Ads leads to signed cases](/law-firm-marketing/how-law-firms-track-google-ads-leads-to-signed-cases/). ### Why cost per lead misleads Two campaigns, same budget: | | Campaign A | Campaign B | |---|---|---| | Cost per lead | $150 | $250 | | Leads that sign | 1 in 12 | 1 in 4 | | Cost per signed case | $1,800 | $1,000 | Campaign A looks cheaper on every report that stops at leads. It costs 80 percent more per client. Smart Bidding pointed at cost per lead will move budget toward A, because that's the number it was told to lower. The gap usually comes from search intent. Broad terms bring in callers researching a problem, people looking for free advice, and cases the firm doesn't take. Narrow terms written the way a client describes their situation bring fewer calls and more matters. ### Setting the target from your own fees Start with the average fee for the matter type, then decide what share of that fee the firm will spend to acquire the client. A firm earning $6,000 on an average matter that's willing to spend 20 percent of it on acquisition has a ceiling of $1,200 per signed case. The share is the partners' call. It depends on margin, how often clients return or refer, and how much capacity the firm has. Set a separate ceiling for each practice area. A contingency-fee injury case and a flat-fee estate plan don't belong under one number. ### What the benchmarks show The click side is well measured. WordStream and LocaliQ's [2026 Google Ads benchmarks](https://www.wordstream.com/blog/2026-google-ads-benchmarks) put attorneys and legal services at a $9.87 average cost per click, the highest of 23 industries, with a 5.55 percent conversion rate and an average cost per lead of $131.63. From there the funnel decides the cost. If a quarter of those leads qualify and 40 percent of qualified leads sign, one lead in ten becomes a client, and the cost per signed case is about $1,320. The case side isn't well measured. Published cost-per-signed-case figures for law firms come almost entirely from agencies quoting their own clients, with no method or sample size. I don't use them to set targets. Your own intake data, even three months of it, beats any of them. ### Turning the number into a bid target Smart Bidding can't see a signed case quickly enough to bid on it directly, because cases sign outside its learning window. Bid on the qualified lead instead, and derive the target from the case ceiling: **Target cost per qualified lead = ceiling per signed case x share of qualified leads that sign.** With a $1,200 ceiling and 40 percent of qualified leads signing, the qualified-lead target is $480. If the firm sends a value per matter type, use Maximize Conversion Value and let the values steer toward the bigger matters. Recheck the signing rate every quarter. If intake gets better at converting consultations, the qualified-lead target can rise, and the account can afford more of the competitive searches. ### When a separate practice-area campaign costs more Firms often split their best practice area into its own campaign to give it more budget. On that firm's account, a new campaign for one practice area was costing about seven times more per booked matter than the firm's general campaign. Budget wasn't the limit. It was losing about half its eligible impressions to ad rank and only about 15 percent to budget, most likely because a new campaign starts without the quality history the old ad groups had built. Raising its budget wouldn't have fixed that. Check the rank and budget split before moving money. ### Is it the ads or intake? Before blaming the ads for a high cost per case, split the funnel. Pull last month's paid calls and check how many were answered, how many reached a consultation, and how many signed. If calls go unanswered or callbacks take a day, the cost per case rises with no change in the ads. Fix the phone first. On one firm I work with, the partners saw almost no new cases in one practice area in August 2026 and assumed the ads had stopped working. The call data said otherwise. Inquiries in that practice area had roughly doubled since May, and intake was booking them at its best rate of the year. The gap came after the booking. I read the transcripts of the 40 booked consultations that month, and 36 ended with the prospect told to call the firm back at a set time. None had the firm committing to make the call. Clients retained on the first call fell from 12 in May to 2 in August. More ad spend would have bought more calls into the same gap. One warning from that analysis: don't measure this with a keyword search. A search of the transcripts for "call back" found 17 of the 40. Reading each one for who had to make the next contact found 36. ### What changed on one account On a law firm account I took over, every form submission counted as a conversion, spam included. Once the conversions counted qualified calls and the campaigns were split by case type, spend fell by half and cost per qualified lead was down about half by day ninety. Google Ads never saw a signed case on that account, so signed cases only show up in the firm's intake log. Later I scored those calls and put the firm's own price on each qualified one by matter type, held in the call-tracking platform. Worked example: Greenacre Law. A real estate law firm’s Google Ads rebuilt around WhatConverts-qualified calls and case-type ad groups: half the ad spend, twice the monthly qualified signups, and cost per qualified lead down about half in 90 days. https://connercrowe.com/results/real-estate-law-google-ads-growth/ ## Why does ChatGPT get my law firm's details wrong, and how do I fix it? URL: https://connercrowe.com/law-firm-marketing/why-does-chatgpt-get-my-law-firms-details-wrong/ Published: 2026-10-27 | Updated: 2026-10-28 Answer: AI assistants repeat what they find, and they often find more than one version of a firm: a wrong phone number on a directory, old hours on a map listing, the wrong admission state on a profile. Log each wrong fact, trace it to its page, and fix the source, starting with your state bar record and Google profile. I'm a marketer, not a lawyer, and nothing here is legal advice. If an AI answer says something about your firm that could mislead a client, your ethics counsel may have a view on what you need to correct and how fast. This page covers finding and fixing the source. ### How often AI gets the facts wrong Named doesn't mean described correctly. In my [AI answers field report](/benchmarks/ai-answers-field-report/), between 15% and 35% of the answers that named a business got at least one fact wrong (per-business range for the 5 businesses whose facts were graded, in different industries, not law firms specifically). The businesses' own records disagreed too. In the shorter snapshot checks, part of the same data, pooled across 9 businesses in different industries, not law firms specifically, 4 of 4 businesses had facts that disagreed across their own website, Google profile and listings, at least 6 facts each. When an assistant finds two versions of a fact, it picks one, blends them, or hedges. I'd fix accuracy before chasing new mentions. A wrong phone number or office reaches the client who already typed your name, which is the client closest to calling. ### Where the wrong versions come from For a law firm, the wrong versions usually come from a few places: - **The state bar record.** If the address or firm name on your bar record is old, directories that compile from bar data copy it. - **Directory profiles nobody claimed.** Avvo, Justia, FindLaw and others build profiles whether you claim them or not. - **Phone numbers and hours.** A tracking number, an old main line, or holiday hours nobody reset. - **The firm's own site.** A footer address that differs from the contact page, an old practice-area page still live, a bio that gives one number of years in practice and the About page another. - **Old names.** A firm that renamed or merged keeps its old name on listings for years. ### Trace each wrong fact to its source 1. **Ask by name and log it.** Use questions 1 to 6 in the free [What AI says about you](/what-ai-says-about-you/) template: what the firm is, reviews, address and hours, practice areas, who runs it, fees. Three runs on each assistant your clients use. Write each wrong fact down word for word. 2. **Ask where it came from.** In my audits, Perplexity, ChatGPT with search, Copilot and Google's AI Mode listed the sites they drew on for most answers, which is how I recorded each answer's sources. Open them. If the assistant shows no source, search the wrong fact in quotes. 3. **Fill the ledger.** One row per fact, one column per source: website, bar record, Google profile, Bing Places, Apple Business Connect, each directory, and what AI said. The row where the wrong value appears is your fix list. ### Fix in this order 1. **State bar record.** Directories copy it. Avvo, for one, asks you to enter your details "exactly as they appear on your state bar registration" when you claim a profile ([Avvo](https://www.avvo.com/for-lawyers/avvo-profile), checked October 27, 2026). 2. **Google Business Profile.** Google's guideline is to "represent your business as it's consistently represented and recognized in the real world," with "a precise, accurate address" ([Google Business Profile guidelines](https://support.google.com/business/answer/3038177), checked October 27, 2026). Name, address, phone, hours, categories and website should match the bar record. 3. **Bing Places and Apple Business Connect.** They feed maps outside Google. Claim both. 4. **Legal directories.** Claim and correct them. The [law directory guide](/law-firm-marketing/do-avvo-justia-super-lawyers-findlaw-profiles-affect-ai-answers/) walks through Avvo, Justia, FindLaw and Super Lawyers. 5. **Your own site.** One address and phone in the footer and on the contact page, a dated bio for each attorney, and old practice-area pages either updated or redirected. Mark up the firm as a LegalService in your structured data with the same name, address and phone. ### How long until the answers change I don't have enough re-tests to give you a reliable timeline. Assistants that search the web at answer time can pick up a corrected listing as soon as they read it. Answers drawn from what a model learned in training can lag much longer. So re-check on a schedule: the same questions, the same assistants, three runs each, a month after you fix the sources. Compare counts of wrong facts, not single answers. If a wrong fact survives, look for a source you missed. ### When it isn't a listing Sometimes the wrong fact comes from a news story, a court record or a forum post you can't edit. Don't post rebuttals on the sites that carry it. Publish the correct fact clearly on your own site and your claimed profiles, so the assistants have a consistent version to weigh against it. For the other half of the problem, firms that are described correctly but rarely named, read [why ChatGPT recommends other law firms](/law-firm-marketing/why-does-chatgpt-recommend-other-law-firms/). ## Why does ChatGPT recommend other law firms instead of mine? URL: https://connercrowe.com/law-firm-marketing/why-does-chatgpt-recommend-other-law-firms/ Published: 2026-10-27 | Updated: 2026-10-28 Answer: Because a client's question rarely includes your name, and on those questions AI answers lean on directories, reviews and lists instead of your website. The firms they name are the ones other sites already describe, consistently. Fix what AI says about your firm first, then get onto the sources the answers cite for your practice area. I'm a marketer, not a lawyer, and nothing here is legal advice. Your state bar's rules decide what a directory profile or a review request may say. This page covers how AI answers pick firms and what a firm can check. ### Two kinds of question, two different answers Someone who already knows your firm types its name: "Is [firm] any good?" Someone who doesn't describes a problem: "My employer didn't pay my overtime. What can I do, and who handles that near me?" AI assistants answer those differently. In my [AI answers field report](/benchmarks/ai-answers-field-report/), which pools graded answers about 9 businesses in different industries, not law firms specifically, the business was named 90.2% of the time when the question named it (751 of 833 answers). On problem questions that named nobody, it was named 2.6% of the time (16 of 623). 6 of the 9 businesses were never named on a problem question at all. In the pooled answers, AI found the business when asked and rarely offered it when it wasn't. ### What the answer is built from When an assistant answers "who should I hire," it reads other sites and summarizes them. In the field report, pooled across 9 businesses in different industries, not law firms specifically, the domains cited for the most businesses were facebook.com, google.com, linkedin.com, reddit.com, bbb.org, yelp.com, youtube.com. Your website explains your services well, and the assistants do use it to explain how something works. When the question turns to who to hire, they want someone else vouching for you. For a law firm that usually means: - **Legal directories.** Avvo, Justia, FindLaw, Martindale and Super Lawyers profiles, many of them built from state bar records whether you claimed them or not. - **Your state bar's lawyer directory.** The record many directories copy. - **Reviews.** Your Google profile and the review sites the answers quote. - **Lists and articles.** "Best [practice area] lawyers in [city]" pages, local news and legal Q&A sites. ### What the firms that get named tend to have I don't have a law-specific measurement. The figures below are pooled across 9 businesses in different industries, not law firms specifically, and they point to four things I'd check first: 1. **The same facts everywhere.** In the shorter snapshot checks, 4 of 4 businesses had facts that disagreed across their own site and listings, at least 6 each. An assistant that finds two addresses or two sets of practice areas hedges, or picks the wrong one. 2. **Reviews on the platforms the answers quote.** 4 of 4 snapshot businesses had fewer Google reviews than the median competitor the answers named. That's a pattern, not proof of cause, but it matches what the answers cite. 3. **Complete, claimed directory profiles.** An unclaimed profile carries whatever the directory compiled. The [law directory guide](/law-firm-marketing/do-avvo-justia-super-lawyers-findlaw-profiles-affect-ai-answers/) covers claiming each one. 4. **A page that answers the client's question.** If a client asks about unpaid overtime, the firm with a plain page on overtime claims in that state has something an assistant can quote. ### What doesn't explain it **Blocked crawlers, almost never.** In the field report data, pooled across 9 businesses in different industries, not law firms specifically, all 530 fetches I made with AI and search crawler user agents across 9 sites came back HTTP 200. Check your robots.txt once, then move on. **One test isn't a ranking.** Ask the same question three times and you'll get different firms. Count across runs before deciding where you stand. **Fake reviews break federal rules.** The FTC's [consumer review rule (16 CFR Part 465)](https://www.ecfr.gov/current/title-16/chapter-I/subchapter-D/part-465) (checked October 27, 2026) bans fake reviews and rewards tied to a positive review, and your bar's rules on testimonials apply on top. Ask real clients, and reply to what they write without disclosing anything about their matter. ### Check what AI says about your firm The free [What AI says about you](/what-ai-says-about-you/) template has ten questions to ask, a log for each answer and a ledger for your facts. For a firm, fill it in like this: - Questions 1 to 6 with your firm's name, so you can see what the assistants say when a client already knows you. - Question 9 with a problem your best clients describe in their own words, before they know they need a lawyer. - Questions 7, 8 and 10 with your city and your main practice area, and the firm you lose consults to. Ask each question three times on the assistants your clients use, then count. Fix any wrong fact at its source first: it reaches the client closest to calling, and it's a change you control. The [guide to wrong firm details](/law-firm-marketing/why-does-chatgpt-get-my-law-firms-details-wrong/) shows how to trace each one. If the firm's ads are part of how it grows, the [signed-case tracking guide](/law-firm-marketing/how-law-firms-track-google-ads-leads-to-signed-cases/) covers the other half: knowing which calls turn into cases. --- # Answers: Pre-Consent Tracking Hub: https://connercrowe.com/pre-consent-tracking/ ## Can a law firm use the Meta Pixel and Google tags on practice-area and intake pages? URL: https://connercrowe.com/pre-consent-tracking/can-a-law-firm-use-the-meta-pixel-on-practice-area-pages/ Published: 2026-10-27 | Updated: 2026-10-27 Answer: Many firms run them. What matters is what the tags send from practice-area and intake pages, where the address, title or form can name a visitor's matter before any consent choice. On those pages I run Google tags in basic consent mode, turn off Meta's automatic events and advanced matching, and report conversions without case details. I'm a marketer, not a lawyer, and nothing here is legal advice. Last verified October 27, 2026; changes are logged at the end of this page and in [the library's change log](/pre-consent-tracking/#what-changed). Whether your firm's tracking is a problem under your state's law or your professional duties is a question for your ethics counsel. This page covers what the tags send and how I set them up. ### Why practice-area pages are different Most website tracking risk comes down to what a request reveals. A request from `/contact/` reveals very little. A request from `/practice-areas/dui-defense/` or `/bankruptcy/chapter-7/` tells the ad platform which legal problem a visitor has, along with an IP address and a cookie that can tie the visit to a person. The tags send that by default. The Meta Pixel's page-view request carries the page address. GA4's carries the address and the page title. Google's advanced consent mode still sends a cookieless ping before consent, and Google's [consent mode overview](https://developers.google.com/tag-platform/security/concepts/consent-mode) lists what's in it, including the referrer and whether the visitor arrived from an ad click. Intake adds more. A consultation form asks for a name, a phone number and often a description of the matter. If Meta's automatic advanced matching is on, the pixel reads form fields and sends them hashed. If session replay runs on the page, the recording can include what was typed unless masking is set strictly. Law firms also have a professional rule that stores don't. [ABA Model Rule 1.18](https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_1_18_duties_of_prospective_client/) treats a person who consults a lawyer about a possible matter as a prospective client and limits use or disclosure of what's learned in that consultation (rule text checked October 2026). States adopt their own versions. I haven't found an ABA opinion on sending intake-page data to an ad platform, so that's a question for your counsel to apply. ### What I've seen so far I haven't found a reported website tracking suit with a law firm as the defendant. The published cases I found sit mostly in healthcare, tax and finance. On October 23, 2026 the Fifth Circuit [revived a Meta Pixel class action](https://www.privacyworld.blog/2026/10/fifth-circuit-revives-meta-pixel-privacy-class-action-expanding-litigation-risk-for-website-data/) against an optical retailer, treating browsing and eye-exam scheduling as plausibly health information. I haven't found a court that has said whether that reasoning reaches legal topics, but a page about a criminal charge or a debt does reveal what the visitor's problem is. ### How I set up a firm's tags 1. **Inventory first.** Every script on the site, with who added it and why. Law firm sites collect plugins, chat widgets and old agency pixels. Anything without an owner comes off. 2. **Basic consent mode, with consent defaulted to denied, on sensitive templates.** Nothing goes to Google from a practice-area or intake page until the visitor makes a choice. On the rest of the site the firm can choose advanced mode with its eyes open. 3. **Meta set to send less.** Automatic events off, automatic advanced matching off, and no custom events or parameters that name a practice area. If the firm doesn't run Meta ads, the pixel doesn't belong on the site. 4. **Practice areas out of the URLs that tags forward.** Where a page has to be measured, collapse the address to something generic before it reaches the vendor. Thank-you pages get a generic path with no form answers in the query string. 5. **No session replay on intake.** Replay tools are useful on landing pages. On a consultation form they're a recording of someone describing their case. 6. **Chat with a notice.** A chat widget on a firm's site should say before the first message that the conversation is recorded, who provides it, and that chatting doesn't create an attorney-client relationship. Ask counsel to approve the wording. 7. **Conversions without case details.** The firm still needs to know which ads produce signed cases. I send Google a qualified call or a booked consult: an identifier, a time and a value for the matter tier. No transcript and no description of the matter. That's covered in [tracking Google Ads leads to signed cases](/law-firm-marketing/how-law-firms-track-google-ads-leads-to-signed-cases/). Calls are their own question. Recording consent, Google's own call recording and what a marketing score measures are in [recording and scoring intake calls](/law-firm-marketing/can-a-law-firm-record-and-score-intake-calls/). ### Check it on your site [The five-minute browser test](/pre-consent-tracking/check-what-your-website-sends-before-consent/) shows what your site sends before a visitor consents. Run it on your homepage and your highest-traffic practice-area page. If Meta, Google or a session recorder shows up before any click, start with the practice-area and intake templates. ### Common questions #### Will basic consent mode hurt a firm's Google Ads results? Expect fewer reported conversions from the pages where it runs. Basic mode still feeds Google's general modeling, but website calls and form fills from visitors who never choose won't be observed. Calls from call assets and Local Services Ads don't depend on the site tag, and qualified calls or booked consults imported from intake don't depend on a browser cookie, so those survive. #### Does a law firm need a cookie banner at all? That depends on where your visitors are and what your tags do, so it's a question for counsel. If you run one, test that Reject all stops the tags. A banner that records a choice and changes nothing gives the firm no help. #### Can a firm keep remarketing? Google restricts personalized advertising for some legal categories, including [criminal defense](https://support.google.com/adspolicy/answer/16701956) and [divorce](https://support.google.com/adspolicy/answer/16700849) (both checked October 2026), so some firms can't use their own remarketing lists anyway. Check the current policy for your practice area, and ask counsel before building audiences from intake-page visitors. ### Update log - **October 27, 2026 (later):** Removed the worked-example card, tightened three sourcing lines, and noted that basic mode needs consent defaulted to denied. - **October 27, 2026:** Published. ## How do I check what my website sends before a visitor consents? URL: https://connercrowe.com/pre-consent-tracking/check-what-your-website-sends-before-consent/ Published: 2026-10-27 | Updated: 2026-10-27 Answer: Open your site in a private window with the browser's network panel open and don't touch the cookie banner. Filter for Meta, Google, Microsoft, TikTok and replay tools. Every data request that shows up went out before consent. Then repeat after Reject all, with Global Privacy Control on, and after typing a test word into your site search. I'm a marketer, not a lawyer, and nothing here is legal advice. Last verified October 27, 2026; changes are logged at the end of this page and in [the library's change log](/pre-consent-tracking/#what-changed). This test shows what your site sends and when. Whether any of it is a legal problem where you operate is a question for a licensed attorney in your state. ### Your tag manager won't show what your tags load Website tracking complaints describe network traffic: which tag fired, on which page, what it carried, and whether it went out before the banner was answered or kept going after "Reject all." Fisher Phillips [tracked six September 2026 rulings](https://www.fisherphillips.com/en/insights/insights/courts-pushed-back-on-website-tracking-lawsuits-in-september), four for the companies sued, several turning on what the complaint said the site actually sent. A tag manager lists what someone installed, and tags load other tags. When I logged what [the Microsoft Ads tag loaded on 32 sites](/blog/bing-ads-tag-loads-clarity-session-recording/), it was the only thing on the page requesting Microsoft's session recorder, Clarity, on 4 of them. A tag list wouldn't show that. The browser's network panel shows every request, so that's where the test runs. ### Set up 1. Open Chrome in an Incognito window, so there are no extensions and no cookies from earlier visits. 2. Open DevTools with F12 and go to the **Network** tab. Tick **Preserve log** and **Disable cache**. 3. Paste this into the filter box. Chrome treats text between slashes as a pattern: ``` /facebook|google-analytics|googletagmanager|googleadservices|doubleclick|pagead|\/collect|recaptcha|maps\.google|tiktok|linkedin|licdn|bat\.bing|clarity\.ms|hotjar|fullstory|mouseflow|luckyorange|crazyegg|intercom|livechat|drift|callrail|whatconverts|calltrackingmetrics|invoca|youtube|vimeo|pinterest|reddit|snap|nextdoor|ads-twitter/ ``` `/collect` catches Google requests routed through your own domain, including Google tag gateway paths. Add any vendor your site uses that isn't on the list. ### Test 1: before any choice Load your homepage, then the page a visitor would least want shared: a condition or treatment page, a practice-area page, or a booking form. Don't touch the banner. Every row that appears left the browser before the visitor chose anything. That can be by design: many US banners say tracking is on until the visitor opts out, and on those sites Test 2 and Test 3 are the ones that show whether the opt-out works. Sort the rows by the Type column. Script downloads (type `script`) fetch a tag's code, which tells the vendor your visitor's IP address and the site they're on. The requests that carry the page and identifiers show up as `ping`, `gif`, `xhr` or `fetch`. Click a few and read the request URL and payload: - **The page.** `dl=` on GA4 and Meta requests and `url=` on Google Ads requests is the page address. `dt=` is the page title. On `/services/dui-defense/` the address alone names the subject. - **Identifiers.** `_fbp` or `fbp=` (Meta's browser cookie), `cid=` (Google Analytics' client id), `gclid` (the ad click). - **Form data.** `ud[` on a Meta request is hashed customer data the site passes to Meta. `udff[` shows up in practice when automatic advanced matching picks up fields on its own. Meta doesn't document that name, so treat any `ud[` or `udff[` value as customer data. FTC technology staff wrote in July 2024 that [hashing doesn't make data anonymous](https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/07/no-hashing-still-doesnt-make-your-data-anonymous). - **Replay.** Any `clarity.ms`, `hotjar` or `fullstory` request means a session recorder is running. ### Test 2: after Reject all Open a fresh Incognito window, so cookies set in Test 1 don't show up as new. Load a page, click **Reject all** (or "Do not sell or share"), clear the network log, and open two more pages. Any new Meta, TikTok or session-recorder request after that click, or any Google `collect`, `pagead` or `doubleclick` request carrying anything other than `gcs=G100`, means the banner recorded the choice without acting on it. Then check **Application, Cookies** for `_fbp`, `_ttp`, `_gcl_au` or `_uetsid` appearing after the rejection. ### Test 3: Global Privacy Control Global Privacy Control is a browser signal that says "don't sell or share my data." Haynes Boone counted 12 states, California and Colorado among them, whose laws require businesses to honor opt-out signals like this ([July 2026](https://www.haynesboone.com/news/publications/george-and-dargan-for-reuters-global-privacy-control-what-businesses-need-to-know.pdf)). California's privacy agency fined Tractor Supply [$1.35 million](https://cppa.ca.gov/announcements/2025/20250930.html) on September 30, 2025, in a case that included opt-outs that didn't work. Firefox sends it in every Private window by default ([Mozilla](https://blog.mozilla.org/netpolicy/2023/11/21/global-privacy-control-empowers-individuals-to-limit-privacy-invasive-tracking/)). Firefox, Brave and DuckDuckGo all block some trackers on their own, so a quiet network panel there can prove nothing about your site. In Firefox, click the shield icon in the address bar and turn off Enhanced Tracking Protection for your site before testing. Firefox's filter box takes the same pattern with `regexp:` in front instead of the slashes, and its Ctrl+F doesn't search request contents, so use Chrome for Test 4. In the browser, open the console and type `navigator.globalPrivacyControl`. It should return `true` ([W3C explainer](https://w3c.github.io/gpc/explainer)). Reload your site and repeat Test 1. If the site honors the signal, ad tags stay quiet or run in a restricted mode. Meta's restricted mode shows as `dpo=LDU` on its requests, and Google's can show as `rdp=1` or `npa=1`. Google also [applies restricted data processing](https://support.google.com/google-ads/answer/9614122) on its own servers when it receives the signal in the states that require it (checked October 2026), so a missing `rdp=1` doesn't prove Google ignored it. ### Test 4: a test word in your site search Type a word nobody would search, like `canarytest417`, into your site's search box and run the search. In the network panel press Ctrl+F (Cmd+F on a Mac), which searches headers, payloads and responses, not just addresses. Search for `canarytest417`. Any hit on a third-party host means your search terms go to that vendor. Site search bars [became a target of these claims in 2024](https://ipandmedialaw.fkks.com/post/102jkiv/the-next-target-of-cipa-litigation-search-bars), followed by a wave of demand letters. The most common route is the results page address: if it reads `?s=canarytest417`, every tag on that page received the term through `dl=`. GA4's enhanced measurement also logs it as `search_term`. Then do the same with your own contact form, using fake details like `test+audit@example.com`. Watch for the email, phone or service choice in any third-party request, and look at the thank-you page address. If it carries the form answers as query parameters, every tag on that page received them. ### What the codes mean | Code | Where | Meaning | |---|---|---| | `gcs=G100` | Google requests | Ads and analytics storage denied | | `gcs=G111` | Google requests | Both granted. Before any click, the default for this visitor's region was granted | | `gcs=G1--` | Google requests | No consent state reached the tag (or `gcs` is missing entirely) | | `gcd=` | Google requests | The Consent Mode v2 state, including ad_user_data and ad_personalization, which `gcs` doesn't show | | `asc=D` / `asc=G` | Microsoft UET | Ad storage denied / granted | | `dpo=LDU` | Meta | Limited Data Use, Meta's restricted mode | | `rdp=1` / `npa=1` | Google | Restricted data processing / non-personalized ads | Google doesn't publish the `gcs` and `gcd` codes. They're read from the requests themselves. A script download from `googletagmanager.com` isn't data about the visitor's activity. A `collect` or `pagead` request to Google before any click means the site runs Consent Mode in advanced mode, or no Consent Mode. In basic mode Google receives no measurement until the visitor chooses. The [Consent Mode answer](/conversion-tracking/google-consent-mode-explained/) covers what those early requests contain, and Google's [consent debugging guide](https://developers.google.com/tag-platform/devguides/consent-debugging) shows the same checks in Tag Assistant. ### What this test can't tell you - **Whether anything is unlawful.** It shows behavior. The law is your attorney's call. - **Server-side destinations.** If your site sends data to its own server first, the browser only shows that hop. Ask whoever runs the server container for the list of destinations and a sample of what each one receives. - **What a vendor does afterward.** It shows what leaves your site, not what the vendor keeps. - **Every visitor's experience.** One browser, one location, one visit. Banners often behave differently by state or on a second visit, which is how [a consent bug I found](/results/consent-mode-denied-for-consenting-users/) hid from everyone who tested it on a first visit. Save what you find. Export each test as a HAR file (the download arrow in the Network panel) along with a dated screenshot. The HAR keeps the payloads, and a screenshot doesn't. Chrome's default HAR leaves out cookie headers, so screenshot Application, Cookies too. That's the record your attorney will want if a letter ever arrives. Run the tests again after any site, theme, plugin or tag change. ### Common questions #### How long does the whole test take? About five minutes for the homepage and one sensitive page. Add a minute for each extra page you care about. #### Do I need special software? No. Chrome's built-in DevTools does Tests 1, 2 and 4. Firefox in a Private window, with Enhanced Tracking Protection off for your site, covers Test 3. #### What should I fix first if things fire early? Start with the pages that name a condition, a legal matter or a booking. On a [law firm's practice-area and intake pages](/pre-consent-tracking/can-a-law-firm-use-the-meta-pixel-on-practice-area-pages/) and a [medical practice's booking flow](/medical-practice-marketing/how-to-track-booked-patients-without-sending-phi/), that usually means removing tags. ### Update log - **October 27, 2026 (later):** Added the opt-out banner note to Test 1, the Firefox tracking-protection step, a sourced count of Global Privacy Control states, and a wider filter. Tightened the Test 2 rule and the Meta form-data note. - **October 27, 2026:** Published. --- # Answers: Wasted Ad Spend Hub: https://connercrowe.com/wasted-ad-spend/ ## Best practices for conducting an audit of my digital advertising account URL: https://connercrowe.com/wasted-ad-spend/best-practices-audit-digital-advertising-account/ Published: 2026-05-25 | Updated: 2026-10-30 Answer: Start with conversion tracking, because broken tracking is the top finding in my 2026 audits: in my Q3 2026 data, 6 of 29 lead-gen accounts counted a non-lead action or click proxy as primary, and 2 of 11 home-services accounts were too broken to report a cost per lead. Then rank every finding by recoverable dollars. An ad audit is a measurement exercise. It answers one question: where is this account losing money, and how much. The output is a written list of findings ranked by dollars. Changes to the account come after, as a separate job. Everything below is how I run that measurement, in the order I run it, and what it costs when someone else runs it for you. If you're deciding whether to hire someone, the [Google Ads audit service and cost breakdown](/google-ads-audit-cost/) covers my ads audit, which is free at any spend level and back within 24 hours of read-only access, and what the rest of the market charges for the same memo. ### Principle 1: run audits on a fixed quarterly cadence The most common failure mode is auditing the account only when performance drops. By the time the numbers slip far enough to trigger an audit, the leak has usually been running for two or three months. A quarterly cadence catches the leak inside the quarter it starts. Four audits a year is enough on a mid-size account. Monthly is too often and produces audit fatigue without surfacing new findings. Annual is too slow and lets seasonal leaks compound. The quarterly slot also matches how most founders close their books, which makes the audit easier to schedule and easier to budget against. Pick a fixed day. The first business day of each quarter works. Put it on the calendar a year out so the audit is never a reactive decision when traffic dips. ### Principle 2: snapshot the account before changing anything Before the first finding is written down, capture the baseline. Export the campaigns view at ninety days, the search-terms report at sixty days, the conversions table at thirty days, and the account settings page. Save the four exports in a dated folder. The reason is simple. Three weeks after the audit, when a campaign is performing better or worse, the question that matters is what changed. Without the snapshot, the question is unanswerable. With it, the audit becomes a reference document the account gets measured against for the next quarter. A founder who skips the snapshot loses the ability to attribute future results to the audit findings. The audit then becomes folklore instead of evidence. ### Principle 3: scope the audit across all five layers A useful audit covers settings, data, structure, creative, and measurement. Skipping a layer is the second most common failure mode after skipping cadence. Settings means the account-level configuration. Conversion goals, attribution model, location targeting, network expansion settings, brand exclusions. Data means the integrity of the conversion column and the agreement between platform and analytics. Structure means how campaigns, ad groups, and keywords are organized. Creative means the ads themselves and the landing pages they point to. Measurement means whether the reporting answers the question the business is asking. Most agency audits cover structure and creative because those are the visible layers. Settings and measurement get skipped because they're tedious. The largest findings usually live in the layers that get skipped. The [free 25-page Setup Checklist PDF](/audit/) covers the two layers that get skipped, as ten settings and tracking red flags with the menu path to fix each one. ### Principle 4: document findings in plain English The audit itself is a written document, not a screenshot dump and not a spreadsheet. Each finding gets one paragraph that names the leak, the dollar exposure, the fix, and the expected lift. A founder reading the audit six months later should understand each finding without opening the platform. Plain English forces clarity. A finding that can't be written as a single paragraph is usually a finding the auditor doesn't fully understand. Writing the audit out also surfaces contradictions, which is where the bad findings get removed before they reach implementation. Length isn't the goal. A useful audit on a mid-size account may surface twelve to twenty observations, but the decision memo should lead with three to five structural fixes across six to twelve pages. Each fix gets the supporting evidence, recoverable-dollar estimate, and implementation sequence, with larger exhibits referenced by filename. Extra pages have to earn their place by making a decision easier. ### Principle 5: enforce a hard split between audit and intervention The single most violated principle. The audit pass and the intervention pass are two different jobs on two different days. During the audit, nothing in the account changes. Findings get written down. The account is left alone. The reason is contamination. If the auditor starts pausing keywords and rewriting headlines during the audit, the snapshot is invalid before it's complete. The auditor also loses the ability to rank findings by severity, because half the leaks are gone before the full picture is assembled. The clean sequence is audit, document, prioritize, then intervene. The intervention happens against the ranked list, in spend order, largest leak first. On a high-volume DTC retail account I audited, the Meta dashboard reported a huge click count at a few cents a click across the trailing ninety days. The temptation during discovery was to start culling campaigns because the click cost looked unreal. The discipline that mattered was waiting. Three days later the audit document showed the campaigns were optimizing for engagement and the click metric was counting post engagements rather than landing-page visits. The intervention that ranked first wasn't a campaign pause. It was a tracking and objective rebuild. Pausing during discovery would have buried the structural finding under noise. ### The order to run the pass in The five principles govern how the audit is run. This is the order the steps go in. Budget an afternoon for a self-audit, or four to six hours on a mid-size account when nothing is documented. **Step one, verify the conversion signal (45 to 60 minutes).** Every number downstream is computed from the conversion column. Open Goals, then Conversions, and answer four questions. Which actions are set as primary, because primary conversions are what Smart Bidding buys toward. Is every call counted as a win, because a conversion rate above 30 to 40 percent on a lead account usually means misdials and spam are being counted. Two home-services accounts in my benchmark reported 67% and 69% conversion rates in July 2026 because taps on their phone numbers counted as leads. Once most of those taps stopped counting, they reported 6.3% and 11.0% in Q3 2026 (N = 2), against 10.3% for the typical account across 9 clean accounts (N = 9). Are the counts real, meaning they reconcile with the store or CRM. Do browser and server events deduplicate against a shared event ID. Read the all-conversions column too, because it shows how much is being counted outside primary. In my Q3 2026 benchmark, all conversions typically ran 2.3x primary conversions across lead-gen accounts (N = 22), and above 10x in 5 of 22. 6 of 29 lead-gen accounts counted a non-lead action or a click proxy as primary (N = 29). Those rows are in my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/). Then compare three numbers over thirty days: the Google Ads conversions table, the GA4 key events report, and the Shopify or Stripe order count. Google Ads should land within 10% of Shopify orders attributed to paid search, and, on a multi-channel account, GA4 paid-channel conversions within 15% of the sum of Google Ads and Meta reported conversions. A gap outside those bands is worth explaining. Above twenty-five percent the cause is usually double-counting or a missing event, and [the Google Ads versus Shopify reconciliation](/conversion-tracking/google-ads-conversions-dont-match-shopify-orders/) holds the band for each pair. A negative gap, where Google Ads reports fewer conversions than the platform, usually means a broken tag or a consent-mode misconfiguration eating attribution. The [Tracking Stack reference](/frameworks/tracking-stack/) covers the deduplication contract the rest of the audit depends on. This isn't a rare edge case. Broken conversion tracking, meaning the wrong primary action, dead tags, or phone taps counted as leads, is the most common finding in my 2026 audits. In the twelve home-services accounts I benchmarked in July 2026, a third had conversion tracking that was inflated or missing entirely. In the Q3 2026 refresh it was 2 of 11 (N = 11), and one of those logged more conversions than it had clicks. Those owners had been making budget decisions on fiction. The full data sits in the [home services benchmarks](/home-services-paid-search-benchmarks/). If step one fails, stop and fix the signal. The rest of the audit reads through this lens. **Step two, audit structure (45 to 60 minutes).** Open the campaigns view at ninety days, sorted by spend descending, and ask whether the money is going where the strategy intends. Three patterns matter. Paid clicks on the client's own brand name, which inflate the apparent return of whatever campaign catches them. A single Performance Max campaign carrying more than forty percent of spend with no audience signals, which usually means the algorithm is harvesting brand traffic and reporting it as prospecting. Ad groups holding more than twenty keywords, which means match types are competing inside the same group. Any of those on a campaign above ten percent of total spend is a finding. My default is no bidding on a client's own brand. The brand name and its misspellings go in as phrase negatives across the account, Performance Max gets a brand exclusion list, and the organic listing takes those searches. The one exception is a competitor's ad holding the brand search above the organic result, and then I bid on the brand on purpose and move the brand negatives from the account level to a shared list applied to every non-brand campaign. A clean structure keeps brand out with those negatives, bounds Performance Max with audience signals, and holds five to fifteen tightly themed keywords per ad group. **Step three, mine the search terms (60 to 90 minutes).** This is the step that pays for the audit on most accounts. Pull sixty to ninety days at the account level, filter for spend above one hundred dollars, sort by cost descending, and read the top fifty terms the way a stranger would. The leaks fall into three buckets: competitor brand terms the account is paying for without intent, generic informational queries that convert at a fraction of commercial intent, and match-type drift where a broad keyword pulled in queries unrelated to the product. Any single term above five hundred dollars with zero conversions over sixty days is a finding. So is any cluster of similar terms above two thousand dollars converting below half the account average. Read it knowing what it leaves out. In Q3 2026 the search terms report showed 52% of account spend for the typical account, across the 11 accounts where I could pull every row (N = 11). Across the 25 clean-tracking accounts with search-term data, the typical one put 30.7% of the spend the report did show into terms with 2 or more clicks and zero conversions (N = 25). The keyword view can look clean while that spend sits one report deeper. **Step four, check audience signals and demographic skew (30 to 45 minutes).** Open the audience tab on the top three campaigns by spend. Two failure modes dominate: no audience signals on Performance Max at all, and signals attached but never reviewed, where a list built eighteen months ago still feeds a campaign whose buyer has shifted. Pull the demographic breakdown. Any single segment above thirty percent of spend converting below half the account average is a finding. **Step five, read landing pages against click cost (45 to 60 minutes).** Reports, Predefined Reports, Landing Page. Sort by clicks descending and read conversion rate, mobile speed score, and behavior in GA4. Any page above one hundred clicks converting below one percent is a finding. So is any page above five hundred clicks with a mobile speed score below six. Founders skip this step most often, because the fix lives outside Google Ads. The fix is also where the largest lift usually sits. On Shopify product pages I audit, moving the price above the fold, swapping a category-grid hero for a single-product hero, or rewriting the H1 to mirror the ad headline are the changes I test first, because they're cheap and the result shows inside a week or two. **Step six, the settings that quietly leak (15 minutes).** Search partners and Display expansion turned on without a decision. Location settings on presence-or-interest instead of presence. Auto-applied recommendations enabled. Bid-strategy targets that contradict the goal set in step one. And the change history, which tells you whether the account is being managed or just billed. Write every finding down with a dollar estimate beside it, rank by recoverable dollars, and fix from the top. The [free Google Ads Setup Checklist](/audit/) is a 25-page workbook covering the ten settings and tracking red flags behind steps one and six, with a wasted-spend worksheet, and it doesn't ask for your email. ### What a free analysis has to cover to be worth the word There are a dozen free "grader" tools online. They score the account against a generic template and produce a PDF with a number on the front. The number means little because the template is generic. What an API scan can't see: whether conversion actions are configured against the right pages, whether enhanced conversions are firing with a hashed email payload, whether the offline-conversion uploader runs on a cadence, whether Performance Max is buying the client's own brand searches inside the asset-group reports. It also can't see the business behind the account. A 4x ROAS on a furniture brand with a six-month repeat cycle is a different number than a 4x ROAS on a one-time-purchase supplement brand. The first is underbidding. The second is barely breaking even after refunds. A real free analysis covers four layers in order: campaign structure, conversion-tracking integrity, search-term and audience quality, and bidding-strategy fit. That last one gets skipped most. Target ROAS on a campaign with twelve conversions a month is statistically meaningless, and Maximize Conversions on a budget-capped campaign is a different campaign than the dashboard pretends it is. Three inputs make the analysis useful. Read-only account access, because screenshots hide the diagnostic context that lives one click deeper. The KPI the business steers on, because a 3x ROAS is excellent in one vertical and bankrupt in another. And three sentences of business context: what the product is, what the AOV runs, what the margin band looks like, and whether the business runs on first-purchase profit or lifetime value. ### Who runs audits, and what each tier costs Four categories of audit exist, and each catches a different kind of leak. **Free DIY workbooks.** A workbook makes the operator type the numbers in by hand, which is the moment most leaks become obvious. A tool that runs in the background while a founder ignores it does nothing. The [free 25-page Setup Checklist](/audit/) is the checklist I start every account audit from. Cost: zero. What it misses: live anomaly detection and account-specific quality-score history. **Automated SaaS auditors.** WordStream Advisor generates a free Performance Grader that scores an account on quality score, click-through rate, account activity, impression share, and wasted spend in about a minute. It's a temperature check, not an audit. Optmyzr is the heavyweight, and its rule engine catches drift a manual review misses. It's priced by ad spend, so check its current pricing page. Adalysis is stronger on responsive search ad analysis. I use paid tooling on accounts above fifty thousand a month in spend, where the line item earns its keep. Below that, the free platform reports plus a fifteen-minute weekly cadence do the same work for nothing. What none of them catch: whether the account architecture matches the business, whether the conversion setup is double-counting, and whether the wrong campaigns are being scaled. Those are the expensive leaks. **Agency audit reports.** Most performance agencies offer a free audit as the front of a sales funnel. A junior analyst spends thirty to ninety minutes in the account, fills out a templated deck, and the senior team uses the findings as a pitch hook. Quality ranges enormously. A specialist shop can flag five-figure leaks in an hour. A generalist runs the same eighteen-point checklist on every account regardless of fit. The honest read is that an agency audit is a sales tool, which makes the incentives obvious: it leans heavy on findings that argue for a retainer and light on findings a founder could fix in an afternoon. Read the deck, take the three highest-impact findings, and judge the retainer math on its own merits. **Senior-consultant manual reviews.** Three to six hours in the account, a written diagnosis against the specific business model, and a prioritized fix list with no upsell attached. Across the market these run one to five thousand dollars depending on account size. I write mine free, at any spend level, back within 24 hours of read-only access. The [audit pricing breakdown](/google-ads-audit-cost/) covers what the other tiers cost. On a law firm I rebuilt, the senior-review pass surfaced findings no grader flagged: broad-match keywords pulled from the rep's recommendations, conversion tracking firing on every form submit including spam, and generic ad copy fighting every other firm on the same headline. Restructuring around case type and rebuilding conversion tracking to count WhatConverts-qualified calls cut monthly ad spend in half and doubled qualified signups in 90 days. | Category | Example | Cost | What it catches | What it misses | |---|---|---|---|---| | Free DIY workbook | Setup Checklist PDF (this site) | Free | Structural and strategic leaks | Live anomaly detection | | Automated SaaS | WordStream, Optmyzr, Adalysis | Free, or priced by ad spend | Bid drift, match-type decay | Account architecture fit | | Agency audit | Templated discovery deck | Free with sales call | Obvious leaks, retainer-friendly fixes | Findings that lose the retainer | | Senior consultant memo | One-time written diagnosis | $1K to $5K one-time across the market (mine is free) | Structural calls, attribution gaps | Ongoing weekly cadence | Ongoing management is a separate purchase from the audit. Google's own Recommendations tab is free and updates daily, and it's the option with the worst incentives on the list, because the platform earns more when the account spends more. Scan it weekly, apply the two or three items a human would have flagged anyway, and decline the rest. In July 2026 I cleared that tab across twenty accounts by adding 97 keywords by hand and dismissing 1,476 recommendations, and every account's Optimization Score went up anyway, because the score rises on a dismissal the same as on an apply. The detail is in [why I dismissed 1,476 of Google's ad recommendations](/blog/google-optimization-score-recommendations/). Agency retainers run two to ten thousand a month under one hundred thousand in monthly spend and start to make math sense above seventy-five thousand. A fractional operator gives you senior judgment without agency overhead. Mine runs $2,500 to $8,500 a month, month-to-month, set by how many channels and add-ons the account needs. Below thirty thousand a month in spend, a written outside audit every six months can beat every retainer on the list. The gap between a dashboard and a diagnosis shows up fastest on Meta, where a cheap CPM hides what the campaign is aiming at. In my Q3 2026 Meta benchmark, 9 of 18 accounts, carrying 42% of spend, optimized for no conversion event at all (N = 18). Reach ad sets returned a 0.06% link CTR against 2.84% for conversion ad sets, and 92.8% of Marketplace spend came from those 9 accounts. A grader scores CPM and CTR. The memo has to say what the campaign is optimizing toward, and the rows are in my [Meta CPM by placement benchmark](/benchmarks/meta-ads-cpm-by-placement/). ### What the first thirty days with an outside operator looks like A clean outside audit runs on a four-week shape. **Week one is access and scope.** Read-only access on each platform: MCC manager-level read on Google Ads, a Business Manager partner add at analyst role on Meta, an agency link with read permission on Microsoft Advertising. Conversion access is the part most founders miss. If the operator can't see how conversions are configured, the audit catches symptoms and never reaches the source. Week one also includes the KPI call, where the founder names the metric the business is measured on at the leadership level, and an hour of business context: what the brand sells, who buys, what the AOV and margin band are, and what changed in the last ninety days. **Week two is the diagnosis pass, read-only by design.** No campaign changes. The diagnosis covers structure, match-type discipline, conversion configuration, audience layering, Performance Max guardrails, attribution fit, and the cross-check between platform-reported conversions and the source of truth in Shopify or the CRM. On a Shopify brand I work with, the week-two read showed three months of Meta spend producing a handful of purchases. Read against the Google Performance Max campaigns on the same brand, it was clear Meta was capturing a small fraction of the buying signal while the founder had been told to scale it for a year. The waste wasn't inside Meta. The waste was that Meta had budget at all. **Week three is the memo.** A written document, three to five findings, ranked by recoverable dollars per month, each with the report it came from, the dollar figure, the proposed fix, and a directional expected lift. The ranking matters more than the count. A founder who tries to fix fifteen things at once fixes none of them. The memo also names what was checked and found clean, because a founder who only sees problems doesn't know what was tested. **Week four is intervention or handoff.** The operator implements, the in-house team implements, or the memo goes to the existing agency with implementation notes. All three are valid endings. A memo specific enough to name the report, the figure, the change, and the expected outcome is a memo an agency can't argue around. Three red flags inside those thirty days. An operator who pauses campaigns during the diagnosis week is contaminating the baseline, and the memo becomes partly a story about their own changes. An operator who pushes a retainer before the memo lands is pricing the relationship ahead of the diagnosis. An operator whose findings all happen to require their ongoing involvement is selling the next engagement. ### Where this leaves you Cadence, snapshot, scope, plain English, change control. The principles travel across Google Ads, Meta, and Microsoft Advertising because they govern how an audit gets run, not what platform it runs on. The order is fixed: tracking, structure, search terms, audiences, landing pages, settings. Put the quarterly slot on the calendar a year out. Run the [free 25-page Setup Checklist](/audit/) against your own account first, because the diagnosis is portable and you own it at the end. The [pricing breakdown](/google-ads-audit-cost/) covers what the market charges for a review and why mine is free, and [the free ads audit](/audit-request/) is the right move when the ranked list runs longer than what an in-house team can absorb before the next quarterly review. I write it myself, back within 24 hours of read-only access. If you'd rather talk it through, [a strategy call](/contact/) is open too. The rest of the leak categories live in [the wasted ad spend library](/wasted-ad-spend/). ## Can I see what my competitors pay per click on Google Ads? URL: https://connercrowe.com/wasted-ad-spend/can-i-see-competitor-cpc-google-ads/ Published: 2026-10-21 | Updated: 2026-10-21 Answer: No. Google never shows a rival's CPC, bid or budget, and no tool can. Prices vary even inside one account: in a law firm account I manage, phrase match cost 23% to 68% more per click than broad on the same words. And 23 of 31 accounts I manage lost more impression share to budget than to Ad Rank. Most pages that rank for this question are published by companies selling a keyword tool, and they show a "CPC" column without saying whose CPC it is. It's an estimate for the keyword across every advertiser. ### What Google shows you about competitors [Auction insights](https://support.google.com/google-ads/answer/2579754) is the only report inside Google Ads that names the other advertisers in your auctions. It works for Search, Shopping and Performance Max, where it splits the Search and Shopping data by channel. For Search you get six numbers per competitor: impression share, overlap rate, position above rate, top of page rate, absolute top of page rate and outranking share. Shopping gets three of them. Together they tell you who shows up when you do, how often they sit above you, and how often you rank above them or show when they don't. None of them is a price. The report has no column for bids, CPC or budget, and Google doesn't publish those numbers for any advertiser but you. ### One advertiser, one keyword, several prices A keyword doesn't have one price, even for a single advertiser. Here's one law firm account I manage, over the 30 days to October 20, 2026, for keywords running in both broad and phrase match: | Keyword | Broad match CPC | Phrase match CPC | Phrase costs more by | |---|---|---|---| | property lawyer | $8.47 (116 clicks) | $12.18 (200 clicks) | 44% | | property attorney | $10.35 (111) | $12.72 (291) | 23% | | real estate attorney | $9.42 (25) | $13.09 (36) | 39% | | real estate lawyer near me | $8.31 (30) | $12.54 (56) | 51% | | real estate attorney near me | $8.74 (18) | $12.70 (56) | 45% | | property law lawyer | $8.41 (33) | $14.16 (64) | 68% | That's one firm, on the same words, in the same month. Part of the gap is that broad match enters different auctions, matching looser searches with less competition, which also brings in more irrelevant ones. Part is Smart Bidding pricing each auction on its predicted value. So when a tool prints one CPC next to a competitor's domain, it can't describe what that competitor paid. It describes the keyword in general, and only roughly. ### What do "top of page bid (low range)" and "(high range)" mean? They're the two bid columns in [Keyword Planner](https://support.google.com/google-ads/answer/3022575). The low range is roughly the 20th percentile of the top of page bids advertisers paid for that keyword, and the high range is roughly the 80th percentile. Both use the last 30 days and your own location and Search Network settings. Read them as a distribution. If a keyword shows $2.10 to $7.80, about a fifth of the bids that reached the top of the page were under $2.10, and about a fifth were over $7.80. Some tool guides call them a minimum and a maximum, which they aren't. Four things change how you use them: - **Your price can land outside the range.** The range is market-wide and ignores your Quality Score. - **Blank means thin data.** Keywords with little bid history show no range. - **Historical metrics ignore your bid and budget.** Keyword Planner's forecasts take them into account. The ranges don't. - **The keywords report's estimate is usually missing.** The "Est. top of page bid" column is [Google's estimate for your own keyword](https://support.google.com/google-ads/answer/2472742), and it isn't available on campaigns using automated bidding. When I pulled it for the law firm account above, which runs Maximize Conversion Value, every keyword came back blank. Using the high range as a bid cap is a practitioner habit, and Google doesn't recommend it. Set the cap from your break-even, below. ### How accurate are Semrush, SpyFu and Ahrefs CPC numbers? They're estimates of what the keyword costs in general, and most start from Google's own keyword data. SpyFu's documentation [calls its costs estimates](https://help.spyfu.com/en/articles/3040771-how-accurate-is-spyfu) built partly from Google's keyword tools. Ahrefs says it [refreshes CPC every few months](https://help.ahrefs.com/en/articles/1015791-how-accurate-is-cpc-in-ahrefs) and that Google Ads itself is the most accurate source. Similarweb [builds its average CPC](https://support.similarweb.com/hc/en-us/articles/6624361069201-CPC-Cost-per-click) from the same 20th and 80th percentile figures Keyword Planner shows. Semrush describes its method only as [machine-learning algorithms and third-party providers](https://www.semrush.com/kb/998-where-does-semrush-data-come-from). The table above shows the limit. One keyword in one account carried prices up to 68% apart depending on match type, so a single estimate per keyword can't be one advertiser's price. ### Why your competitor pays a different price than you would Google charges you the least you need to [clear the Ad Rank thresholds and beat the advertiser directly below you](https://support.google.com/google-ads/answer/6297), and never more than your max CPC bid, apart from the exceptions Google lists. Ad Rank combines your bid with the quality of your ad and landing page. A competitor with a better landing page can pay less than you for the spot above you. Copying their bid would tell you nothing, even if you could see it. ### Is a competitor the problem at all? Usually not the main one. Google Ads reports how much impression share each campaign loses to budget and how much to rank. In my [Q3 2026 benchmark](/benchmarks/google-ads-account-benchmarks/), 23 of 31 Search accounts I manage lost more impression share to budget than to Ad Rank (N = 31). On a family law account I run, the campaign lost 54% of its eligible impressions to budget in the 30 days to October 20, 2026, and 6% to rank. Budget was the limit. The law firm in the table is the other case. Its three campaigns each lost between 35% and 46% to budget and between 39% and 42% to rank in the same period, so it competes on both fronts. More budget alone would recover only part of the lost impressions, and so would better Ad Rank alone. Check this split before you worry about what anyone else pays. In the campaigns view, add the competitive metrics columns. ### The number that decides your bid Work out the most a click can cost before it loses money. For ecommerce, that's average order value times gross margin times conversion rate. A store with a $180 average order, a 40 percent margin and a 2.5 percent conversion rate breaks even at $1.80 a click ($180 x 0.40 x 0.025). For lead generation, use the profit from a closed job times your lead-to-sale rate times the click-to-lead rate. If a closed job earns $1,500, one lead in five closes and 6 percent of clicks become leads, the break-even is $18 a click ($1,500 x 0.20 x 0.06). Compare that figure with what you pay now and with the bid range. For reference, the median cost per click across the 33 Google Ads accounts in my Q3 2026 benchmark was $8.53. When break-even sits below the market, the levers are a better conversion rate, a better margin, a lower position, or longer search terms with less competition. ### What about competitors bidding on your brand name? I don't bid on my clients' own brand names by default. By default, my accounts carry the client's brand names as phrase negatives across the account, and the organic listing takes those clicks for free. The exception is narrow: a competitor's ad sits in position 1 above your organic result on your own name and is taking the click. Then a tightly built brand campaign is worth testing, with the brand terms kept out of every other campaign. The other causes of a rising CPC are on [why CPC climbs without sales](/wasted-ad-spend/common-reasons-high-cpc-without-sales/). If your campaign can't spend its budget at the bids you've set, start with [whether to raise bids](/wasted-ad-spend/should-i-raise-bids-when-the-budget-is-not-spending/). ## Do more clicks mean more sales on Google Ads? URL: https://connercrowe.com/wasted-ad-spend/do-more-clicks-mean-more-sales-google-ads/ Published: 2026-10-21 | Updated: 2026-10-22 Answer: Not reliably. More clicks often mean cheaper, lower-intent traffic. In a 14-day AI Max test on one service account, clicks rose 72% and CPC fell 42%, but qualified leads fell 43% and cost per qualified lead rose 76%. I judge an account by cost per booked appointment instead. A click is a visit you paid for. A sale is a person who wanted what you sell, found it, and acted. The two move together only when the extra clicks come from the same kind of search as the ones you already convert, and on a fixed budget they usually don't. When clicks and sales split, the pattern in the account usually points to where to look first. | What you see | Most likely cause | Check first | |---|---|---| | Clicks up, conversions flat, CPC down | The extra clicks came from looser searches | The search terms report, sorted by cost | | Clicks up, conversions up, booked work flat | The conversion counts the wrong thing | Which action is primary, and whether spam fills count | | Clicks steady, conversions at zero | The conversion tag broke | The action's status column, then a test form fill | | Clicks up on Maximize Clicks | The strategy buys clicks and never checks for leads | The bid strategy column | ### Why more clicks usually means cheaper clicks On a budget that doesn't change, the only way to buy more clicks is to pay less for each one. Cheaper clicks tend to come from looser searches: research questions, nearby services, people comparing prices with no plan to book this week. Fewer advertisers compete for those searches, which is why they cost less. Google is plain about what its click-focused strategy does. [Maximize clicks](https://support.google.com/google-ads/answer/6268626) is built to "get as many clicks as possible within your budget." Nothing in that goal asks whether the click books an appointment. ### What happened when I let the clicks rise I ran a 14-day head-to-head on one service account. One campaign ran AI Max with search term matching on. The other ran on my own keyword list, with the same budget, creative and landing pages. The full write-up is in [my AI Max test](/blog/google-ads-search-term-matching-ai-max/). | Metric | My keywords | AI Max | Change | |---|---|---|---| | Clicks | 642 | 1,107 | +72% | | CPC | $3.89 | $2.26 | -42% | | Form fills | 32 | 45 | +41% | | Qualified leads | 14 | 8 | -43% | | Cost per qualified lead | $178 | $313 | +76% | Every dashboard number got better while the one the business cared about got worse. Most of the extra volume came from educational searches the landing page pulled in, and the sales team spent more time sorting 45 fills to find 8 real prospects than it had spent on 14 from 32. That's one account over two weeks, so I treat it as a warning sign rather than a rate. ### The account where the clicks were the problem The extreme version was an inherited account I audited in early 2026. It ran Maximize Clicks and had a month of spend with zero conversions. Session recordings in Microsoft Clarity showed five-second visits with no scrolling and no clicks on the page. I switched it to Maximize Conversions. Within about two weeks spend and clicks had dropped sharply, and the first real lead came in soon after. That's one account (N = 1). Losing those clicks cost the account nothing, because none of them had produced a lead. How bot traffic shows up in that pattern is in [how ad fraud and click bots affect your budget](/wasted-ad-spend/ad-fraud-click-bots-budget-impact/). ### Fewer clicks, more booked work On a law firm account I rebuilt, I fixed the conversion signal first so it counted WhatConverts-qualified calls and validated forms instead of every submit. Then I rebuilt the campaign around case type and cut the broad match. Within 90 days spend was down 50%, qualified signups doubled and cost per qualified lead was down about half. I didn't isolate which change did the most, because four things moved in one window on one account. A fence contractor showed the other side. Its CPC had climbed to about two and a half times its level three months earlier, and click volume fell 75% on an unchanged budget. The phones had slowed, and the account said website leads were zero. Underneath, a button renamed from "Send" to "Submit" had silently killed the only website conversion, so the account couldn't see its own form leads. After I rebuilt the trigger and cleaned up the account, the 90-day view at the follow-up audit showed 582 clicks and 42 conversions. The [dead form case](/results/dead-form-conversion-fix/) has the detail. Click count told the owner nothing useful in either direction. ### How common is bidding for clicks? Across 30 Google Ads accounts I manage, in the 30 days to October 21, 2026, 8 of 51 Search campaigns ran on Maximize Clicks. Another 35 ran Maximize Conversions, 4 Maximize Conversion Value, 3 Manual CPC and 1 Target Impression Share. Click-hungry traffic also shows up in the search terms. In my [Q3 2026 benchmark](/benchmarks/google-ads-account-benchmarks/), a median 30.7% of visible search-term spend went to terms with 2 or more clicks and zero conversions (N = 25). That's only the spend Google shows, which was a median 52% of the account (N = 11). ### When more clicks do help More clicks pay when two things are true at once. The campaign is limited by budget, and its search terms are already clean. Then more money buys more of the same searches you already convert, at about the same rate. The first condition is common in my accounts: in Q3 2026, 23 of 31 Search accounts lost more impression share to budget than to Ad Rank (N = 31). The second is rarer. If your campaign sits in that group and the search terms report reads like your customers talking, a higher budget is a fair test. If the campaign can't spend what it has, the question is different, and [whether to raise bids](/wasted-ad-spend/should-i-raise-bids-when-the-budget-is-not-spending/) covers the arithmetic. Maximize Clicks still has a place on a new campaign with very little conversion history, with a CPC cap, while data builds. Once it has conversions to learn from, I move it to a strategy that bids on them. ### What to watch instead of clicks **Cost per qualified lead.** For a service business, a qualified lead is a booked appointment. My clients mark those in WhatConverts, which tracks each call and form with its source. Divide spend by booked appointments, by campaign. **Conversion rate on the action that matters.** Read leads per click on the primary conversion, and check its status column first. A rising click count with a flat lead count means the new clicks aren't buying. **The search terms report.** Sort by cost and read the top 50. If the terms that grew are questions, job titles or other services, the extra clicks came from searches your customers don't make. When CPC falls and you want to know if it's a win, [lowering CPC without losing conversions](/wasted-ad-spend/strategies-lower-cpc-without-sacrificing-conversions/) walks through which cuts hold up. ## How can I identify if my online ad campaigns are overspending? URL: https://connercrowe.com/wasted-ad-spend/how-to-identify-overspending-ad-campaigns/ Published: 2026-05-25 | Updated: 2026-10-30 Answer: A campaign is overspending when two warning signals appear together, like brand clicks and inflated conversion counts. Two accounts showed 67% and 69% conversion rates that fell to 6.3% and 11.0% once phone taps mostly stopped counting. ### Why most founders miss the signals The ad platforms are built to hide overspending. Optimization scores reward more campaigns, more match types, more spend. The recommendations tab tells you what to add, almost never what to cut. By the time a founder asks the question, the leak has compounded for three to six months. Diagnose this the way a senior operator does. Pull five specific reports. Look for two of the five signals to appear together. That's the threshold for an audit. One signal in isolation is noise. Two together is structural. ### Signal 1: search-term reports leaking irrelevant queries Open the search terms report inside Google Ads. Filter to the last ninety days. Sort by impressions descending. Read the top one hundred queries the way a stranger would. If more than a quarter of those queries describe a product you don't sell, an intent you can't serve, or a job seeker looking for employment at your company, broad match is teaching the algorithm what your business isn't. The fix isn't a longer negative-keyword list. The fix is structural: tighter match types, audience signals layered onto Performance Max, and a deliberate decision about where broad match is allowed to operate. A solo founder running a six-figure account often has zero search-term review cadence. That's the first place the money goes. On live accounts in Q3 2026, the search terms report showed only 52% of account spend for the typical account (N = 11). Across the 25 clean-tracking accounts with search-term data, the typical one put 30.7% of the spend the report did show into terms with 2 or more clicks and zero conversions (N = 25). Both figures come from my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/), read through the Google Ads API. That makes a quarter a weak pass line on its own. Zero-conversion spend and irrelevant queries aren't the same measure, but a 25% bar on zero-conversion spend would flag the typical account in that benchmark and more than half the sample. Use the share to rank which campaigns to read first, and count it as a signal when it sits well above the account's own last pull or pairs with a second signal below. ### Signal 2: branded search inflating your ROAS Branded search queries convert far better than non-branded search. If your reported ROAS lives in the 5x to 8x range and a meaningful share of your spend runs on campaigns that include brand terms, the math is misleading. You're paying Google for traffic that would have arrived organically. My default is to stay out of that auction. I add the client's brand name and its misspellings as phrase negatives across the account, and the organic listing takes those searches. After the negatives go in, watch cost per lead or cost per sale for a month. If it rises, the brand clicks were flattering the campaign, and the higher number is the one to manage against. The exception is narrow. Search your own name in an incognito window. If a competitor's ad holds the top position above your organic result and is taking the click, test a separate brand campaign, kept apart so it can't inflate non-brand numbers, and move the brand negatives from the account level to a shared list applied to every non-brand campaign. If the competitor sits below your listing, let them pay for clicks your organic result still wins. [High CPC without sales](/wasted-ad-spend/common-reasons-high-cpc-without-sales/) covers that test in detail. ### Signal 3: Display and partner-network spend Performance Max and Display campaigns rarely surface the network breakdown unless you specifically ask for it. Pull the placement report and the network report. If a non-trivial share of spend is going to mobile-app placements, parked-domain networks, or generic Display inventory, that money is doing almost no work for a Shopify store or a service business. Exclude the worst placements explicitly. Trust nothing the platform auto-suggests on Display targeting. A new Performance Max campaign left on default settings can serve on inventory with no commercial intent, and nothing on the campaign card tells you how much. ### Signal 4: conversion rates outside the benchmark band Read conversion rate against a benchmark with a stated sample, or against the account's own history. For home-services Google Ads, the typical click-to-lead rate in my [Q3 2026 home-services cost-per-lead benchmark](/home-services-paid-search-benchmarks/) was 10.3%, ranging from 6.3% to 13.2% across the accounts with clean tracking (N = 9). Ecommerce purchase rates run far lower than lead rates, so a store is better judged against its own trailing rate. The [vertical breakdown for furniture and decor brands](/for-home-brands/furniture/) covers the conversion benchmarks against a longer consideration window. A campaign-level conversion rate far below that reference on traffic that looks relevant points at a landing-page problem, a tracking problem, or a match-type problem. A rate far above it usually means the account is counting things that aren't sales or leads: branded overlap, phone taps counted as calls, or a de-duplication failure between the browser pixel and the conversion API. Two home-services accounts in that benchmark reported 67% and 69% conversion rates in the July 2026 edition. Once taps on their phone numbers mostly stopped counting as leads, they reported 6.3% and 11.0% in Q3 (N = 2). At the same click price, that correction alone multiplied the reported cost per lead by about eleven on one account and about six on the other. The higher cost per lead was the honest one. The [Tracking Stack reference](/frameworks/tracking-stack/) covers the de-duplication contract in detail. If your numbers don't line up with Shopify, that document is the first place to look. ### Signal 5: CPA climbing while margin holds flat Cost-per-acquisition climbs over time in most paid accounts. That isn't waste on its own. It becomes waste when CPA climbs faster than average order value, contribution margin holds flat, and the account manager tells you to trust the algorithm through the worsening trend. The fix isn't more spend. The fix is a redistribution of budget between campaigns and a fresh read on which audience segments are compounding. Customer Match lists fed from email and SMS belong in that read, because they show which segments keep buying. ### Three more signals that hide on the campaign card The five above surface in reports a founder opens. These three sit one column deeper and cost more than most bid changes. **Quality Scores below 5 on top-spend keywords.** The score itself isn't an auction input. It's Google's summary of expected CTR, ad relevance, and landing-page experience, and Ad Rank uses real-time versions of those same three. Don't read it as a CPC verdict, though. Across 30 Google Ads accounts I manage, 30 days to October 21, 2026, In the 7 of those accounts with enough clicks on both, keywords on QS 1 to 5 typically cost 0.98x what QS 7 to 10 keywords cost per click inside the same account, and more in only 3 of them (N = 7). A low score matters when the click it buys doesn't convert. Pull the keyword report, add the Quality Score, expected click-through rate, ad relevance, and landing page experience columns, and sort by cost descending. For any keyword in the top twenty by spend scoring below 5, compare its cost per conversion with the account's. The fix is ad copy written to the query and a landing page that names the query in the H1, and no bid change is needed. **Broad match running without audience signals.** Broad match with no first-party audience data attached is a blank check. The match type expands the query universe to whatever Google decides is related, and without a signal telling the algorithm who matters, it spends against whoever clicks. Layer Customer Match lists from email and SMS, recent purchasers, and high-intent site visitors onto every broad-match ad group and every Performance Max asset group. Set them as signals, not exclusions. Bid strategy and match type are one paired decision, not two separate ones. **Retargeting frequency left uncapped.** Pull the frequency column on retargeting campaigns. Serve the same person enough impressions in a week and the campaign crosses from reminder to nuisance, usually sooner than operators assume. Cap frequency where CTR starts falling. Rotate creative on a thirty-day cadence at minimum. Segment the pool by recency, because a visitor from two days ago needs a different message than one from sixty days ago. Most retargeting waste is the same ad served too many times, not the targeting. ### Four checks above the platform Every ad platform reports its own performance. Google attributes generously, Meta attributes more generously, TikTok attributes most generously of all. A single-channel dashboard is a sales document. The honest read of efficiency sits above the platforms, in Shopify or the order-management system, against contribution margin. **Platform-claimed revenue against actual revenue.** Add platform-reported revenue from Google Ads, Meta Ads Manager, and TikTok Ads Manager for the last thirty days. Compare the sum to total Shopify revenue for the same window. If the platforms together claim well above real revenue, attribution overlap is inflating every channel's ROAS, and each platform is taking credit for the same orders. On the one store in my [ecommerce tracking accuracy benchmark](/benchmarks/ecommerce-tracking-accuracy/) where both sides were readable, the ad platform's counted purchases equaled 58.5% of the store's entire 90-day order book, while the store attributed none of those orders to paid (N = 1). Blended ROAS, meaning total revenue divided by total ad spend, is the only number to manage against. **New-customer ROAS quarter over quarter.** Pull new-customer revenue from Shopify by month for six months and divide by ad spend by month. The trend matters more than the level. A declining new-customer ROAS while blended ROAS holds steady means returning buyers are masking acquisition inefficiency. The platforms are recycling existing customers through retargeting and Customer Match and reporting their revenue as ad-driven. Healthy acquisition holds new-customer ROAS flat or improving against a stable margin. **Click-to-conversion ratio.** Total clicks across all channels for ninety days, divided by total orders attributed to those channels in Shopify. Track it against its own trailing average. A ratio that keeps climbing points at one of three causes: the landing page doesn't match the ad, tracking is firing on engagement events the platform counts as clicks, or the targeting is pulling audiences with no purchase intent. **View-through inflation.** In Meta Ads Manager, switch the attribution window from the default seven-day click and one-day view to seven-day click only. If reported ROAS drops sharply, the campaign is taking credit for impressions that never produced a click. On a brand with a working organic and email program, view-through credit pulls organic revenue into paid reporting. The seven-day click window reads honestly. Manage against that one. ### Two checks that settle it on search Search campaigns give you a cleaner answer than social does, because the query is the intent. **Score search-query relevance.** Export the top two hundred queries by cost over ninety days. Tag each one relevant or irrelevant to what you sell. Sum the cost of the relevant queries and divide by total cost. Broad-heavy accounts score far lower than accounts on tight match types. A low score is a structural leak rather than a tuning issue, and the fix is to tighten match types before adding negatives, then re-score after two reporting cycles. **Compare blended CPA to an LTV-adjusted ceiling.** Blended CPA is total ad spend across all paid channels divided by new customers acquired in the same window. The ceiling is contribution margin per customer across the first twelve months. For a Shopify brand at a 60-dollar average order value, 45 percent contribution margin, and a 1.6x repeat rate, that ceiling lands near 43 dollars. For a service business at a 4,000-dollar average contract and 35 percent margin, it's closer to 1,400 dollars. Cross the ceiling and the account is buying customers at a loss even while individual campaigns look fine. One more note on the spend-to-revenue gap. Google will almost always report higher revenue than Shopify or the CRM, because it counts assisted conversions on a longer window. Google Ads conversions landing within 10 percent of the Shopify orders attributed to paid search over thirty days is attribution. Wider than that, Google is claiming revenue another channel produced or a setting is padding the value, and [the Google Ads versus Shopify reconciliation](/conversion-tracking/google-ads-conversions-dont-match-shopify-orders/) lists the causes in the order I rule them out. Trust the platform of record and recalculate ROAS from that number before touching a bid. ### The Meta version of the same read | Ad sets optimized for (my Q3 2026 Meta benchmark) | Pooled CPM | Link CTR | N | |---|---|---|---| | Reach and impressions | $2.88 | 0.06% | 6 | | Conversion events | $20.35 | 2.84% | 9 | On Meta, the first thing to check is what each ad set was told to buy. In my [Meta CPM benchmark](/benchmarks/meta-ads-cpm-by-placement/), 9 of 18 accounts, carrying 42% of the spend, optimized for no conversion event at all in Q3 2026 (N = 18). Their ad sets looked cheap. Reach and impressions ad sets bought impressions at a $2.88 pooled CPM and returned a 0.06% link CTR, against $20.35 and 2.84% on conversion-optimized ad sets (N = 6 and N = 9). A low CPM on an ad set with no conversion event is the platform finding the cheapest people to show the ad to, with no weight on who buys. Open Ads Manager, add the optimization event column, and sort by spend. Any ad set spending real money with no purchase or lead event behind it gets the same treatment as a Search campaign with no conversion tracking: no verdict on its efficiency until the event is set and firing. Then take the leads or orders it does produce into the CRM or Shopify, because the signals above tell you whether the dashboard reads honestly, and only the back office tells you whether the honest read is profitable. ### What to do once you have spotted two signals Two signals together is the threshold. From there, either work the [free 25-page Setup Checklist](/audit/) against the account yourself, or add me as a read-only user for my [free written audit](/audit-request/), back within 24 hours of access. Fix them in order, because the order is load-bearing. Fix the search-term leak and Quality Scores often correct themselves. Put the brand name in as a phrase negative and CPA reads honestly for the first time. Fix the attribution overlap and platform ROAS becomes a real number. Fix the tracking and the entire diagnostic moves from guesswork to math. A leak in one place almost always means a leak in three others. The whole library was built to walk through the rest of them. The [services overview](/services/) covers the structural fix at the same depth as my paid engagements, and [the wasted ad spend library](/wasted-ad-spend/) indexes every diagnostic. ## How do ad fraud and click bots affect my digital advertising budget? URL: https://connercrowe.com/wasted-ad-spend/ad-fraud-click-bots-budget-impact/ Published: 2026-05-25 | Updated: 2026-10-21 Answer: Junk traffic and bots waste the most budget on app inventory with suspiciously high click rates and on bid strategies that buy clicks, not conversions. In my Q3 2026 Meta data, Audience Network rewarded video posted a 13.85% link CTR against 1.26% in Feed, and an inherited Max Clicks account I audited spent a month with zero conversions. ### The number you have probably heard is wrong Most fraud statistics circulating in marketing trade press come from vendors who sell fraud-detection tools. Those reports routinely claim that twenty, thirty, even forty percent of ad clicks are fraudulent. The number is shaped by the business model of the company publishing the number. The real picture is more useful, and it splits by inventory. Google Search itself, the keyword-matched results page, is the cleanest inventory you can buy once the platform has filtered invalid traffic. Display, Search Partners, YouTube placements, and Performance Max app inventory are where the junk concentrates, and how much depends on the category. Meta's Audience Network has similar issues. In my Q3 2026 Meta benchmark, Audience Network rewarded video posted a 13.85% link CTR across 6 accounts, about 11 times the 1.26% in Facebook Feed (N = 6 and N = 18). Rewarded video plays inside mobile apps and games, where people watch an ad to earn something in the app. In the 3 accounts with about 91% of spend on Audience Network, Meta reported more link clicks than total clicks (N = 3). The rows are in my [Meta CPM by placement benchmark](/benchmarks/meta-ads-cpm-by-placement/). Facebook and Instagram feed placements themselves are reasonably clean. I'm not going to hand you a fraud percentage, because nobody outside the platforms can measure it cleanly. Your own invalid-activity credits, covered below, are the closest real number you have. So the honest answer to "is fraud eating my budget" depends entirely on where the budget is running. ### What Google already does for free Google removes invalid clicks before they bill the account. The platform's [invalid traffic filtering](https://support.google.com/google-ads/answer/2549113) runs in two stages. Real-time filtering blocks obvious bot patterns at click time. Offline filtering reviews delivered clicks against more sophisticated patterns and issues invalid-click credits within sixty to ninety days. Those credits show up in the billing summary as "invalid activity adjustments." Most advertisers never look at the line. Open the billing tab, filter to the last twelve months, and read the credit totals. If the credit on a Display-heavy account is less than one percent of spend, the filtering is either catching very little or very little is getting through. Both are worth investigating. Meta does similar filtering but is less transparent about the credit mechanism. Invalid traffic on Meta is more often dealt with by excluding the audience network and by tightening placements to feed and stories. ### Where fraud lives in your account Three placement categories carry almost all of the real fraud exposure on Google Ads. [Search Partners](https://support.google.com/google-ads/answer/1722047) is the network of non-Google sites that show Google search ads. Quality varies wildly. Toggle it off in any campaign that doesn't need the marginal reach. The setting lives under campaign networks. It's off almost everywhere I run Search: across 30 Google Ads accounts I manage, 30 days to October 21, 2026, Search Partners was on in 1 of 51 Search campaigns (1.1% of spend), and Display expansion was on in 0. Display Network placements on long-tail sites and apps are where I look first for junk impressions and clicks. The standard fix is placement exclusion lists, not vendor tools. Pull the placements report, sort by spend, and exclude any mobile app bundle ID that doesn't have an obvious commercial relationship to your offer. Performance Max app placements are the modern version of the same problem. PMax distributes spend into mobile games and utility apps by default. The placement report is buried inside the campaign insights section. The [wrong-audience diagnosis walkthrough](/wasted-ad-spend/warning-signs-ads-targeting-wrong-audience/) covers the placement-exclusion process for Performance Max in detail. On one ecommerce store I manage, the Performance Max placement report showed spend going to mobile-game and utility-app inventory, and I excluded those categories at the account level. On a Shopify account, placement quality is the more useful frame than fraud rate. App placements are rarely fraud. They're usually bad inventory dressed up as reach. ### How to spot fraud in your own data Five tells separate fraud from ordinary low-quality traffic. Click-through rates that are improbably high on a single placement, twenty, thirty, fifty percent CTR, are almost always automated. Real humans don't click ads at those rates. Session durations under two seconds at meaningful volume from a single source signal headless browsers or click farms. GA4's engagement-time metric makes this readable. Conversion rates of zero across hundreds of clicks from a placement that drives volume on no one else's account. If a placement converts for nobody, it's either fraud or terrible context. Either way, exclude it. Geographic patterns that don't match your buyer base. A US-only ecommerce brand seeing significant click volume from data center IP ranges in Vietnam or Brazil is paying for bot traffic. Device patterns skewed extremely toward older Android versions are a click-farm signature. Real users distribute across recent OS versions. The bid strategy decides how much of this you buy. On one inherited account I audited in early 2026, Max Clicks had run a month of spend with zero conversions. The Microsoft Clarity recordings showed 5-second visits with no mouse movement, no scroll and no click. I switched the bid strategy to Max Conversions, and within about two weeks spend and clicks dropped sharply and the first real lead came in. That's one account (N = 1). A strategy that buys clicks will buy the cheapest clicks available, and bots are cheap. ### The third-party tools worth paying for Three tools earn their fee for the right kind of account. None of them are necessary on a Search-only Google Ads account spending under twenty thousand a month. [ClickCease](https://www.clickcease.com/) works well for service businesses running heavy Google Search competition where repeat clicking from one source is a real risk. The tool blocks repeat clickers at the IP level via auto-applied IP exclusions. The value is highest where each click is expensive enough for repeat clicks to hurt. In my Q3 2026 benchmark, home-services accounts paid a median $13.27 per click (N = 11), against $8.53 across all 33 accounts (N = 33). [Legal advertisers](/for-service-brands/law-firms/) face the same math in practice-area auctions. ClickGUARD targets the same use case with deeper customization. The interface is more technical. The reports are more useful for accounts running multiple campaigns where the patterns need to be sliced finely. Lunio, formerly PPC Protect, is the option for accounts running significant Display, YouTube, or PMax spend. The platform handles invalid placement detection at scale and integrates with Google Ads to push exclusions automatically. For accounts with seven-figure annual paid budgets across networks, the math works. ### When the tools don't earn their fee A Shopify brand running Google Search and Shopping with under fifty thousand a month in spend will see almost no measurable ROI from fraud-detection tools. That inventory is already the cleanest Google sells, and paying a vendor a cut of spend to remove a fraction of what's left is usually a negative return. For [Shopify home brands](/for-home-brands/furniture/), the placement-settings discipline at this spend band looks different. The honest move for most accounts is the structural fix. Disable Search Partners. Exclude the mobile app placement categories on Display and Performance Max. Audit the invalid-activity credit line quarterly. Pull the geographic and device reports once a month. That stack covers most of the fraud exposure that matters, at zero added cost. ### What to ignore Treat any fraud statistic from a vendor selling a fraud-detection product with skepticism. The same applies to whitepapers, industry reports, and webinars sponsored by those vendors. The number is rarely a lie. The framing is almost always engineered to sell software. Treat the noise from competitors claiming "my competition is bombing me with bot clicks" with similar skepticism. Real competitive click fraud exists in a handful of verticals. Most of the time, the bad ROAS comes from the structural problems covered in the [wasted-ad-spend diagnosis library](/wasted-ad-spend/), not from a malicious actor. The [Google Ads Setup Checklist](/audit/) covers the placement and network settings that close most of the real fraud exposure in under an hour. The [Tracking Stack reference](/frameworks/tracking-stack/) covers the analytics layer that makes the fraud diagnosis honest in the first place. A noisy data layer makes ordinary low-quality traffic look like fraud, which sends accounts down the wrong rabbit hole. If Display, YouTube, or PMax spend dominates the account and the placement report reads as untouched, the [services overview](/services/) covers the exclusion lists I apply on the first pass. ## How do I diagnose a low conversion rate on my ads? URL: https://connercrowe.com/wasted-ad-spend/how-to-diagnose-low-conversion-rates-digital-ads/ Published: 2026-05-25 | Updated: 2026-10-30 Answer: Check the tracking first, then the landing page, the audience-offer fit, and mobile speed, in that order. The rate itself is often wrong: two home-services accounts I manage reported 67% and 69% conversion rates that fell to 6.3% and 11.0% once phone taps mostly stopped counting as leads. The typical home-services account with clean tracking converted 10.3% (N = 9). ### Start with benchmarks, not feelings Before touching a campaign, get honest about what a healthy conversion rate looks like for the account, from a benchmark with a stated sample or from the account's own history. For home-services Google Ads, the typical click-to-lead rate across the accounts with clean tracking in my [Q3 2026 home-services cost-per-lead benchmark](/home-services-paid-search-benchmarks/) was 10.3%, ranging from 6.3% to 13.2% (N = 9). A rate far above a range like that usually means the account is counting things that aren't leads. Two accounts in the July 2026 edition reported 67% and 69%, because a tap on the phone number counted as a call. Once most of those taps stopped counting, they reported 6.3% and 11.0% in Q3 (N = 1 each). For ecommerce, purchase rates run far lower than lead rates, so use the store's own trailing rate. Traffic that looks relevant in the search terms and still barely converts is a tracking problem or a landing-page problem. Bidding comes last. Most founders skip this step and start adjusting bids. Bids are the last lever to touch in a low-CR diagnosis, not the first. Pull the campaign-level conversion rate, the [landing-page conversion rate from GA4](https://support.google.com/analytics/answer/9216061), and the device split. Read the device split against the business type. Across the lead-gen accounts with clean tracking in my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/), mobile converted at a higher rate than desktop in 21 of 23 (N = 23). On all four purchase-tracked stores it went the other way (N = 4). So on a lead-gen account, mobile converting well below desktop is unusual, and it points at speed and layout before a single audience setting gets reviewed. A current reference point for lead-gen accounts: across the 23 lead-gen accounts in my Q3 2026 benchmark with lead tracking I trust, the typical account paid $119 a lead in Q3 2026, with the middle half between $75 and $210 (N = 23). The full table, with every N, is in the same [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/). ### Root cause 1: tracking is broken and you don't know it This is the root cause founders blame last and should blame first. Broken conversion tracking is the most common finding in the audits I ran in 2026: the wrong action set as primary, tags that stopped firing, phone taps counted as leads. A low rate on traffic that looks relevant can be a conversion that stopped recording, and a high rate can be a click proxy counted as a lead. The dead version is the one that hides best. On one home-services account, the only website conversion died when someone renamed the form button from Send to Submit, which broke the click-text trigger behind it. The account kept showing a cost per conversion built from call-button taps alone, and Smart Bidding kept steering budget toward people likely to tap a phone number. After the trigger was rebuilt on the form software's own success event, the 90-day view at the follow-up audit showed 582 clicks and 42 conversions. The receipt is in [the dead-form case study](/results/dead-form-conversion-fix/). Diagnostic order: 1. Open the conversion actions list and read the Status column before any performance report. An action marked Inactive has recorded nothing in 30 days. 2. Open [Google Tag Assistant](https://tagassistant.google.com) on the live site. Walk a real purchase or form submission. Confirm one fire of the conversion event, not zero, not three. 3. For a store, compare Google Ads conversions for the last 30 days against the Shopify orders GA4 attributes to paid search. If they sit more than ten percent apart, the tracking is the diagnosis. Stop tuning ads until that's fixed. The full band table is on [why Google Ads conversions don't match Shopify orders](/conversion-tracking/google-ads-conversions-dont-match-shopify-orders/). 4. Read the category of every action counted in the bidding metric. On two of three ecommerce accounts in my [tracking-accuracy benchmark](/benchmarks/ecommerce-tracking-accuracy/), add-to-cart or begin-checkout events sat inside the counted conversions next to purchases. One more failure mode to rule out before trusting any engagement or bounce number: a page firing two `page_view` events on load will misreport engagement upward and hide the problem entirely. The [Tracking Stack reference](/frameworks/tracking-stack/) covers the browser-pixel and conversion-API de-duplication contract in detail. If reported CR is suspiciously low across every campaign at once, the answer is in that document before it's in the ad copy. ### Root cause 2: landing-page mismatch with the ad Open the ad. Read the headline. Click the ad. Watch what loads on a cold mobile session. If the hero promise on the page doesn't echo the ad in the first 1.5 seconds of scroll, the bounce is already baked in. Common mismatches I find in audits: - Ad promises a specific product, the landing page is a category grid with 40 options. - Ad promises a price point, the landing page hides price below the fold. - Ad targets a problem ("squeaky stairs"), the landing page leads with a brand story. - Ad targets cold traffic, the landing page assumes the visitor already knows the company. Use GA4's paths exploration to see where ad sessions drop. Then run Microsoft Clarity or Hotjar on the top three ad-destination pages, and list it in your privacy policy. Watch ten session recordings each. The pattern shows up inside fifteen minutes. Cold visitors don't read past the hero unless the hero answers the question the ad implied. If you run Microsoft Ads, [your UET tag may already be loading Clarity](/blog/bing-ads-tag-loads-clarity-session-recording/). On a law firm rebuild, the original ad copy promised "Real Estate Lawyer Near You" against landing pages that walked through every practice area the firm covered. Cold paid traffic landed on a wall of unrelated case types and bounced. The rebuild wrote ad copy and landing pages around specific case types (purchase disputes, contract reviews, title issues), so the hero answered the search that earned the click. On [legal accounts](/for-service-brands/law-firms/), the same pattern punishes harder because intent windows are short. ### Root cause 3: audience-offer mismatch The campaign is reaching humans. The humans aren't the buyers. This shows up two ways. On Meta, the audience is too broad and the offer is too narrow. A premium price point shown to a $30k-household lookalike will produce clicks and zero purchases. The clicks look like engagement. The denominator inflates. The CR collapses. On Google, the intent is wrong. A Performance Max campaign left free to serve on your own brand name gets credit for buyers who would have arrived anyway, which flatters its rate while the non-brand queries it was bought for get dumped into broad match and fill the search-term report with irrelevant queries. My default is that no campaign bids on the client's own brand: the brand name and its misspellings go in as phrase-match negatives at the account level. The one exception is when a competitor's ad holds the brand search above the organic result. With brand out, the rate you're reading belongs to the traffic the campaign found. The fastest confirmation is a paid-versus-organic comparison on the same URL. Pull the GA4 Landing page report under Engagement and read engagement rate for paid sessions against engagement rate for organic sessions on the identical page. A small gap is normal, because paid traffic skews colder. A wide one, say organic engaging at 65 percent while paid engages at 45, means the audience clicking the ad isn't the audience the page was written for. The diagnosis is targeting or creative, and no layout change will fix it. When paid Shopping traffic bounces on product pages that organic visitors engage with, the audience is the problem, not the page. Check the feed and the negative-keyword library against the buyer profile before you touch the page. Then read the surface directly. Pull the search-term report on Google and the placement report on Meta and spend twenty minutes reading them like a stranger. If a big share of the surface area is wrong-fit, the audience is the diagnosis. ### Root cause 4: page speed and mobile render failures If mobile CR sits far below desktop CR on a lead-gen account, run [Google PageSpeed Insights](https://pagespeed.web.dev) on the top three landing pages. Largest Contentful Paint above 2.5 seconds on a 4G connection is dropping sessions before the page is visible. Cumulative Layout Shift above 0.1 is bouncing visitors who tap the wrong element while the page is still settling. Speed problems hide inside ad accounts because the platforms still report the click. The session counts. The conversion never had a chance. On Shopify product pages, a slow Largest Contentful Paint on a 4G connection loses mobile sessions before the hero loads. Compressing the hero image, deferring third-party review widgets, and disabling video-poster autoplay closes that gap inside one development cycle, and it's a fix to make before any audience tuning. [Furniture brand operators on Shopify](/for-home-brands/furniture/) read these conversion patterns against a longer consideration window, and heavy-image catalogs get hit hardest. ### The landing-page audit: eight red flags Root cause 2 explains why the page fails. These are the specific failures, in the order I check them on a paid-traffic page. **1. The hero doesn't match the ad.** Open the ad in one tab and the page in another. If the noun the ad promised, a product, a price, or a problem, is missing from the H1 and hero subhead, the message match is broken. The test takes thirty seconds on the top three ads by spend. Fix by editing the page H1 to mirror the ad headline word for word on cold traffic, then run a one-week before-and-after on the same audience. On a home-services client, the landing page was being treated like a full website page: long scroll, multiple service descriptions, a team bio below the fold. The ad promised one specific service and the page answered with the company's entire menu. The fix was the opposite shape, a single above-fold viewport with the pain point named, the solution stated, one social-proof line, and one CTA. **2. Largest Contentful Paint over four seconds on mobile.** Covered above. Common causes are an uncompressed hero image over 500KB, a third-party review widget loading synchronously in the head, and a video poster set to autoplay at full resolution. **3. A form longer than five fields on cold traffic.** Each required field you add costs completions on first-touch traffic, and forced account creation before Shopify checkout does the same to cart traffic. Watch ten Clarity or Hotjar recordings of visitors who reached the form and left, and note which field they abandoned on. If three of ten quit on the phone field, the phone field is the diagnosis. On lead-gen accounts I audit, demoting the phone field to optional on cold-traffic forms is the first form test I run. It's worth running when the team is willing to filter on the follow-up call instead of on the form. **4. Mobile layout breaks at the hero.** Open the page on a real phone, not the devtools emulator. Look for a CTA that needs a scroll to reach, a hero image that crops the headline, a sticky header covering the H1, or an exit popup firing on a scroll gesture. If mobile CR is under half of desktop CR and the demographic split looks identical, the diagnosis is layout. **5. An ambiguous primary CTA.** "Learn more" isn't a CTA on paid traffic. Neither is "Submit". The primary CTA tells the visitor exactly what happens next: "See pricing", "Book a 15-minute call", "Add to cart". Specificity moves conversion rate more than color or size. **6. Social proof buried below the fold.** A visitor from a Meta ad has no relationship with the brand and needs a trust signal before scrolling. Press logos in the header band, a star rating next to the H1, one customer quote before the first feature block. Screenshot the first viewport on mobile and desktop. If no third-party trust signal appears in either, the page is asking a stranger for money with nothing in hand. **7. Multiple CTAs competing for attention.** Pages that ask visitors to do three things end up with visitors doing zero. A newsletter signup next to a demo request next to add-to-cart splits the signal. One primary CTA above the fold and zero secondary CTAs. If a heatmap shows clicks scattered evenly across three buttons in the first viewport, the page has no primary outcome. **8. Friction before the buy decision.** Hidden price is the worst offender. A cold visitor who has to fill a form to see a number usually leaves instead. Forced account creation before checkout is the ecommerce equivalent, and a required phone field is the service-business one. The [home and decor playbook](/for-home-brands/furniture/) covers the price-display patterns that hold across home brands. Build a four-step funnel in GA4 exploration, landing-page view, CTA click, form start, form submit, or the ecommerce equivalent, and the step with the largest drop is the friction point. ### When conversion rate alone is too thin to read On lower-volume campaigns, conversion rate bounces around too much to diagnose from. Engagement rate fills the gap. Universal Analytics defined a bounce as a single-page session with no second hit. GA4 inverted it. [Engagement rate](https://support.google.com/analytics/answer/12253918) is the share of sessions lasting longer than ten seconds, firing a conversion event, or generating two or more page views, and bounce rate is now its inverse, hidden until a report adds it as a secondary column. The practical read is the same. A paid session that fails the engagement test was billed by the platform and wasted by the page. There's no official healthy line, so read paid engagement on a page against organic engagement on that URL and against your other paid landing pages. Read it on paid traffic only. Reports, Acquisition, [Traffic acquisition](https://support.google.com/analytics/answer/9756891), filtered to Session default channel group equals Paid Search or Paid Social. Pull at least 28 days so the sample holds, because under 14 days on a single landing page hides too much daily noise to call. Then slice by URL in the Landing page report and read the top ten pages by paid sessions. A high-volume page whose paid engagement sits well below its organic engagement and below your other paid pages is leaking spend. A page that holds up against both is doing its job, and the diagnostic moves downstream in the funnel. ### What to do once you have isolated the root cause | Order | Root cause | Why it sits there | |---|---|---| | 1 | Tracking | Every metric downstream depends on it, and a dead conversion makes every other read wrong | | 2 | Landing page | It is the cheapest fix | | 3 | Audience-offer fit | It is a campaign-architecture decision | | 4 | Speed | It is the longest fix and the one most likely to need an engineer | Diagnose in the order above. Tracking first, because every metric downstream depends on it and a dead conversion makes every other read wrong. Landing page second because it's the cheapest fix. Audience-offer third because that's a campaign-architecture decision. Speed fourth because it's the longest fix and the one most likely to need an engineer. Inside the page itself the order is the same shape. Message match first, technical second, friction third. Message match is a one-line edit with a one-week before-and-after. Render fixes usually move paid mobile engagement within seven days of landing. Friction fixes are page-level decisions a founder can make in a Shopify or Webflow editor inside a day. If two of the four root causes apply at once, the [free 25-page Setup Checklist](/audit/) covers the account settings and tracking checks underneath them. Three at once and [the free ads audit](/audit-request/) beats another solo iteration: written by me, back within 24 hours of read-only access, no call. Conversion rate is the easiest metric to misread and the hardest to fix without the right diagnostic order. A page that loses one buyer in five at the hero can't be rescued by ad-side optimization. The fix lives on the page, not in the campaign. [The wasted ad spend hub](/wasted-ad-spend/) walks the rest of the failure modes I see in account reviews. ## How do I know if my retargeting ads are wasting budget? URL: https://connercrowe.com/wasted-ad-spend/retargeting-ads-not-converting-warning-signs/ Published: 2026-05-25 | Updated: 2026-10-22 Answer: Check who the ads reach before anything else. In one Google remarketing campaign I audited in May 2026, only 14 of 1,247 clicks came from past site visitors, because Optimized Targeting was left on. Then check purchaser exclusions, frequency against your own baseline, overlap with prospecting, and CTR decay on the same creative. ### Retargeting fails differently than prospecting A retargeting campaign and a cold-traffic campaign break for different reasons. Cold campaigns mostly fail on audience-offer fit or landing-page mismatch. Retargeting campaigns mostly fail on audience hygiene, saturation, and attribution overlap. The diagnostic order is different, and the warning signs are different. Fixing a retargeting problem with cold-traffic playbooks wastes weeks. Most retargeting waste is structural. The campaign was set up correctly six months ago and drifted as the audience pool shrank, a default setting widened who it served, and the same buyers kept seeing the same creative. The signs below are the ones I check first on any account review. ### Signal 1: the campaign isn't retargeting anyone Start with who the ads reach, because every other signal assumes the audience is made of past visitors. In May 2026 I audited a Google Display campaign labeled retargeting. Of 1,247 clicks, 14 came from real past website visitors. The other 1,233 went to "expansion and optimized targeting," which is Google serving the ad to people it predicts will convert. One account, one campaign, and the cause was a single checkbox left on by default. The check takes two minutes. Open the Display ad group, go to the Audience segments view, and compare clicks on your remarketing lists against clicks under expansion and optimized targeting. If expansion carries most of the clicks, the campaign is a cold Display campaign with warm-audience creative on it. Turn Optimized Targeting off, confirm the remarketing list is the targeting rather than a signal, and judge the campaign again after it has run on the real audience. Once the tag is live, Google Ads creates a few default data segments on its own, such as All visitors and All converters, and with dynamic remarketing parameters on the tag, segments like Shopping cart abandoners ([Google Ads Help](https://support.google.com/google-ads/answer/6335506)). Category viewers and the past customers you want excluded are each a rule you write or a GA4 audience you link. If the campaign only targets All visitors, it's retargeting everyone who ever loaded the homepage. ### Signal 2: high frequency on small lists On Meta, pull the frequency column at the ad-set level for the last 7 days and look at the small audiences. If frequency keeps climbing week after week, the campaign is paying to show the same ad to people who already saw it several times and chose not to buy. There's no universal ceiling to judge against, so judge against your own history. For scale: across 18 Meta accounts I manage, median frequency over the 92 days of Q3 2026 was 3.2 impressions per person reached, with the middle half between 2.70 and 4.40 (N = 18). That's a full quarter across every ad set, prospecting included, so a small retargeting list will run hotter than that over a single week. The fix isn't a budget cut. Cutting budget keeps frequency high because the small audience absorbs whatever spend it gets. The fix is a creative refresh and a time-window reset so the 7-day window doesn't roll the same users back in on day 8. If you reach for a frequency cap by moving the ad set to a Reach goal, know what that goal buys. In the same Q3 2026 sample, reach-optimized ad sets drew a 0.06% link CTR against 2.84% on conversion-optimized ad sets (N = 5 and N = 9). A reach goal asks Meta for the cheapest people to show the ad to, with no weight on who clicks. If the audience is very small, retargeting is the wrong tool. That audience belongs in [an email flow](/blog/email-flows-room-scenes-shopify-home-brands/), not a paid retargeting set. ### Signal 3: retargeting audiences include past purchasers Open every retargeting ad set and check the exclusions list. If `purchasers_180d` or the equivalent customer-list audience isn't in the exclusions, the campaign is paying Meta or Google to show ads to people who already bought. I check this on every account, including ones that ran clean two years ago, because exclusions get dropped when ad sets are duplicated or rebuilt. The leak compounds because past purchasers are the most engaged segment in the pixel. They click. They generate CTR. The ad set looks healthy in the dashboard while the incremental purchase rate is near zero. Suppress every past-purchaser window (30, 90, 180 days at minimum) from every retargeting ad set, and build a separate post-purchase upsell campaign with its own creative and its own cap. ### Signal 4: conversion volume declining while spend holds steady Pull a 90-day chart of retargeting spend on the same axis as retargeting purchases. If spend is flat and purchases have trended down across the last 60 days, the audience is likely exhausted. The buyers who were going to convert already did. The remaining pool is the visitors who looked once and left, and they get more expensive to convert every week. Audience exhaustion looks identical to creative fatigue in a dashboard. To tell them apart, swap creative on a single ad set and watch what happens. If CTR recovers, it was creative fatigue. If CTR stays flat, the audience is the problem and no new ad will save it. Refill the top of the funnel before adding retargeting budget. ### Signal 5: heavy audience overlap with prospecting Run the Meta Audience Overlap tool between the prospecting lookalike and the retargeting custom audience. Heavy overlap means the two campaigns are bidding against each other in the same auction for the same users. Both campaigns report a conversion. The deduped account-level conversion count is lower than the platform thinks. This shows up as a gap between Meta reported purchases and Shopify orders over a 30-day window. The [Tracking Stack reference](/frameworks/tracking-stack/) walks through the deduplication contract that surfaces it in the data. The fix is to exclude the retargeting custom audience from the prospecting campaign, not the other way around, and to verify the gap closes inside two weeks. ### Signal 6: retargeting takes credit for buyers who were coming back anyway A retargeting campaign is credited for buyers who would have returned through email, organic search, or by typing the store's name into Google. The reported ROAS looks strong. The incremental ROAS is lower, and the platform report can't tell you by how much. The test that can is a geo holdout. On one home-furnishings brand, I paused every Meta and Google campaign in one state for 21 days and held the rest of the country constant. Revenue in that state fell 38%. The attribution report had said 92% of revenue was paid-influenced. That's one brand, and it measured all paid media rather than retargeting alone, but it shows the size of the gap attribution leaves open. The setup is in [the 21-day geo-holdout test](/blog/21-day-geo-holdout-incrementality-test/). For retargeting specifically, run the same holdout with only the retargeting campaigns excluded from the test geography. If purchases in that market barely move, retargeting has been claiming credit for demand the store already had. ### Signal 7: CTR decay on the same creative Pull the last 60 days at the ad level. If average CTR fell steadily across the same creative set, the ads are fatigued. CPMs hold steady. Clicks drop. Conversions follow clicks down. Fatigue is the easiest signal to fix and the easiest to ignore, because the dashboard still shows spend going out the door. Rotate the top creatives on a fixed schedule on any retargeting set with real spend. Keep an evergreen control ad running alongside the rotation so the fatigue curve is measurable instead of guessed at. ### How to triage when several signals fire at once Work them in this order: who the ads reach (Signal 1), exclusions (Signal 3), overlap (Signal 5), frequency (Signal 2), fatigue (Signal 7), borrowed credit (Signal 6), then audience exhaustion (Signal 4). The first three are settings changes that take an hour. Frequency and fatigue are creative-and-cap changes that take a week. Borrowed credit and exhaustion are structural decisions that take a month to validate. [The wasted ad spend hub](/wasted-ad-spend/) has the adjacent diagnostics. The audience-segment view, the exclusions list, and the audience-overlap report are all free and all inside the platform, and they get skipped because the dashboard already shows green. If three of the seven signals fire on the same retargeting set, run the sequence above before adding a dollar of budget. ## How do I lower cost-per-click without losing conversions? URL: https://connercrowe.com/wasted-ad-spend/strategies-lower-cpc-without-sacrificing-conversions/ Published: 2026-05-25 | Updated: 2026-10-30 Answer: Lower CPC by cutting the clicks that never convert before you touch bids. Add negatives from the search terms report, tighten broad match where it buys off-offer searches, and fix ad and landing-page relevance. Then judge every change on cost per qualified lead: in one 14-day test on one account, CPC fell 42% while cost per qualified lead rose 76%. ### First, how Google sets the price "How is the cost per click figured out?" A contractor asked me that on a first call this year, so here's the short answer before the levers. Every search runs an auction. Google ranks the ads by Ad Rank, which weighs your bid against how likely people are to click your ad, how well your landing page matches the search, your ad assets and the context of the search. You usually pay just enough to hold your spot above the next advertiser, not your full bid. That's why a tight match between keyword, ad and page wins the same position for less, and why a crowded local market costs more for everyone in it. ### Why lowering bids is the wrong first move The instinct on a high-CPC account is to pull the max-CPC slider down. That move almost always costs conversions. Lower bids cut ad rank, ad rank cuts impression share on the queries that convert, and the algorithm starts spending the remaining budget on cheaper, lower-intent inventory. The total click count looks fine. The conversion count drops. The correct play is to lower the price Google charges per click without telling the auction to bid less. That's what every lever below tries to do. Each one changes a non-bid variable that feeds into Ad Rank or into what the campaign chooses to bid on. Judge every one of them by cost per conversion, not by CPC on its own, because a lower CPC can come with a worse customer. I tested that on one account in a 14-day split: an AI Max campaign against a manual control, same budget, same creative, same landing pages. CPC fell from $3.89 to $2.26, 42% lower. Form fills rose 41%. Cost per qualified lead rose from $178 to $313, 76% higher, and almost all of the extra volume came from educational queries the landing page pulled in. That's one account in one vertical over two weeks, and the full table is in [what the AI Max rows in the search terms report mean](/blog/google-ads-search-term-matching-ai-max/). It's the reason I never call a CPC drop a win until the qualified-lead column agrees. Check your vertical before you call a click expensive. In Q3 2026 the typical cost per click across 33 Google Ads accounts I manage was $8.53 (N = 33), but home services typically paid $13.27 (N = 11) and the other lead-gen group $10.47 (N = 10). Judge your CPC against your own vertical's row in my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/), not against the overall figure. How CPC changed year to year on 12 accounts I have managed every summer since 2023 is in my [Google Ads CPC trend, 2023 to 2026](/benchmarks/google-ads-cpc-trend/). ### Lever 0: negate the searches that never convert This one comes before the other seven because it costs nothing and moves spend the same day. Across the accounts I manage, the typical account put 30.7% of its visible search-term spend into terms with 2 or more clicks and zero conversions in Q3 2026 (N = 25). Sort the search terms report by cost, read the rows with no conversions, and add the ones that can't become customers to a shared negative list attached to every Search campaign. ### Lever 1: improve Quality Score on top-spend keywords The score itself isn't an auction input: it's Google's 1-to-10 summary of expected CTR, ad relevance, and landing-page experience, and Ad Rank reads real-time versions of those same three at every auction. In Google's model, improving the three lets you hold the same position for less per click. Google doesn't publish the size of the effect, and my own data says not to promise one. In the accounts I manage, I compared CPC on keywords scored 1 to 5 with keywords scored 7 to 10 inside the same account. Few accounts had enough clicks in both bands to compare, and in those, low Quality Score didn't reliably mean a higher CPC. A good share of keyword spend also runs on keywords with no score reported at all. So I work Quality Score for what the components tell me: whether the ad and the page match the search. That shows up in conversion rate more reliably than in CPC. Implementation. Open the keyword view in Google Ads, add the Quality Score, expected CTR, ad relevance, and landing-page experience columns, sort by spend descending. Pick the top twenty rows. Rewrite each ad to put the keyword in headline 1 and the first description line. Point the click to a landing page that contains the keyword in the H1 and the opening paragraph. Watch the components over the next ten to fourteen days, and measure the result as cost per conversion on those keywords. If CPC moves too, count it as a bonus. ### Lever 2: tighten match types where broad buys the wrong searches Broad match in 2026 reaches well past the keyword you typed. The algorithm expands to queries that share theme with the seed, and the expansion often lands on searches the offer doesn't serve. The cost of that is wasted spend per conversion, and tightening the match type doesn't reliably make the click cheaper. On one law firm account I manage, over the 30 days to October 20, 2026, phrase match cost 23% to 68% more per click than broad on the same six keywords. The [full keyword table](/wasted-ad-spend/can-i-see-competitor-cpc-google-ads/) is on my answer about competitor CPC. So I tighten a match type to stop paying for searches that can't convert, and I expect CPC to hold or rise while cost per conversion falls. Implementation. Pull the search-terms report for the last sixty days. Any broad seed with more than a quarter of its impressions on off-offer queries gets demoted to phrase. Any phrase keyword with persistent off-offer queries gets demoted to exact. Add the off-offer queries themselves as campaign-level negatives. The campaign loses volume on the noisy searches and keeps the converting volume. Read the result on cost per conversion two to four weeks later, not on the CPC column the next morning. ### Lever 3: layer audience signals on the campaign Audience signals are a directional steer for the algorithm. On Performance Max and broad-match Search, layering a converter list, a high-intent custom segment, or a remarketing list tells the bidder which auctions are worth paying for. The signals point the bidder at people more likely to convert, so more of the budget lands on clicks worth paying for. Implementation. Build three audience signals per campaign: a converter list from the past 540 days, a custom segment built from competitor URLs and high-intent search terms, and a remarketing list covering thirty-day site visitors. Layer them as signals on Performance Max or as observation audiences on Search. Review the audience report monthly and prune segments with low conversion rate. ### Lever 4: day-parting and device modifiers New campaigns run twenty-four-hour, all-device schedules by default, and the account's own conversion data rarely backs that up. A B2B account that converts between 9am and 6pm Monday through Friday is paying full CPC on evening and weekend clicks that convert at a fraction of the rate. Device is where I see people cut the wrong side. In my Q3 2026 benchmark, mobile converted at a higher rate than desktop in 21 of 23 lead-gen accounts with clean tracking (N = 23), and mobile clicks cost more than desktop clicks in 16 of 21 (N = 21). Down-bidding mobile on a lead-gen account to save on CPC would usually cut the device that converts best. All 4 purchase-tracked stores went the other way: desktop clicks cost more and desktop converted better (N = 4). The rows are in my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/). Read your own account's device split before you touch a modifier. Implementation. Open the time segment and device segment in the campaign report. Pull ninety days of data. Down-bid windows and devices that convert clearly below the campaign average, with a negative bid adjustment rather than an exclusion, so the algorithm keeps the data feed for learning. On Smart Bidding campaigns most bid adjustments are ignored (device -100% still works), so use ad schedules and the conversion data itself instead. ### Lever 5: fix the bid strategy when the bid strategy is the thing overpaying The first four levers change inputs the auction reads. This one changes who is doing the bidding. [Smart Bidding](https://support.google.com/google-ads/answer/7066642) sets a CPC for every auction from signals the platform thinks predict a conversion. When those inputs are thin, wrong, or pointed at the wrong event, the algorithm still has to bid, and the bid it picks can sit far above what a manual review would have approved. The platform never reports that as overpayment. It reports it as cost. Five conditions tell you the strategy itself is the leak. Each shows a different symptom, and the fix is rarely a tweak. It's a downgrade to a simpler strategy with a tighter cap until the data justifies a smarter one. **Target CPA running on thin data.** The practitioner floor is thirty conversions in the last thirty days inside the campaign before tCPA learns reliably. Run it on a campaign producing eight, twelve, or twenty conversions a month and the algorithm pays whatever CPC the auction demands. The symptom is a CPC that climbs week over week while conversion volume stays flat. Count the campaign-level conversion column for the last thirty days. Under thirty, tCPA is bidding blind. Switch to Maximize Conversions, and if CPCs run away, set a Maximum CPC bid limit near the historical average CPC. The cap protects the budget while data accumulates. **Maximize Conversions chasing a soft event.** The strategy optimizes against the count of whichever conversion action you selected. If that action is an add-to-cart, a form view, or a newsletter signup, the algorithm chases the easiest version of the event and pays inflated CPCs for clicks that produce it without producing revenue. The symptom is rising conversion volume on a soft event paired with flat or declining purchase volume. Check which actions sit in the Primary column. Demote soft events to Secondary so they report without influencing bidding, and promote only purchase or qualified-lead events to Primary. Give it two weeks before you read the result. This is common: in my Q3 2026 benchmark, 6 of 29 lead-gen accounts counted a non-lead action or a click proxy as a primary conversion (N = 29). On a law firm rebuild, every form submit was marked Primary in the conversion library, including spam, bots, and partial-field submissions. The algorithm was optimizing toward whichever campaigns produced the most submits, regardless of lead quality. I demoted the raw form-submit event to Secondary and promoted only WhatConverts-qualified phone calls and validated form submits. The number I judge that change on is qualified leads per dollar, not CPC. For [law-firm Google Ads](/for-service-brands/law-firms/), checking those qualified calls against the firm's intake log every month is what keeps that signal honest. **Target ROAS on inflated or unstable conversion values.** [Target ROAS](https://support.google.com/google-ads/answer/6268637) works only when the value side is stable and accurate. Values that include tax and shipping inflate the ROAS the algorithm thinks it's hitting, and values that swing month over month, common under fifty conversions, give it no stable target. The symptom is a CPC that spikes after every value adjustment or any month with an unusual order mix. Set values to the merchandise subtotal rather than the order total, and require fifty conversions a month before running tROAS. Below that, Target CPA on the merchandise value or Maximize Conversion Value with a bid cap performs better. **Manual CPC with no recent calibration.** Manual CPC is the safest strategy on a low-volume account and overpays when the bids haven't been recalibrated to current auction prices. A bid set fourteen months ago against a less crowded auction keeps winning expensive clicks long after the impression-share-lost-to-rank column says it sits above the clearing price. (Enhanced CPC, the old automated top-up on manual bids, is gone from Search and Display campaigns, which now run as plain Manual CPC.) The symptom is a flat top-of-page CPC well above your vertical's benchmark. Run a bid simulator review on the top twenty keywords by spend and bring manual bids in line with the simulator's curve. **Smart Bidding pointed at an event that fires too early.** This one hides inside an otherwise correct strategy. If the conversion tag fires on a thank-you page that loads whether or not payment cleared, the algorithm optimizes toward a population that includes failed checkouts, and CPCs rise because it thinks it's winning conversions that never closed. The symptom is a Google Ads conversion count running above the store's real order count, outside the 10 percent band in [the Google Ads versus Shopify reconciliation](/conversion-tracking/google-ads-conversions-dont-match-shopify-orders/). The fix sits in the tracking layer. The [Tracking Stack reference](/frameworks/tracking-stack/) covers the server-side conversion contract that prevents the drift, and CPC corrects on the next cycle once the tag fires only on completed orders. Implementation. Pick the strategy by conversion volume rather than preference, and move up the ladder only after the threshold is met inside the campaign itself, not the account. | Monthly conversions | Recommended strategy | Bid cap rule | Notes | |---|---|---|---| | Under 15 | Manual CPC or Maximize Clicks with a CPC cap | Bid simulator review | Recalibrate quarterly | | 15 to 30 | Maximize Conversions | Optional max CPC near historical avg | Cap protects budget while learning | | 30 to 50 | Target CPA | No cap, set tCPA at goal | Monitor 60-day signal | | 50+ with stable AOV | Target ROAS | No cap, set tROAS at goal | Watch value calibration | | Multi-campaign portfolio | Portfolio Strategy | Set at portfolio level | Use if campaigns share audience | When you move from Maximize Conversions to tCPA at roughly thirty conversions, set the target at the trailing thirty-day average CPA, not below it. Hold the setting fourteen days before any further adjustment, and lower the target by no more than ten percent per change. Aggressive tCPA cuts make the algorithm drop volume rather than chase efficiency. The error pattern to avoid is launching a new campaign on Target CPA because it worked on the mature campaign next to it. The new campaign has zero conversions, the algorithm has no signal, and the CPC defaults to whatever wins the auction. ### Lever 6: improve landing-page experience Landing-page experience is one of the three quality signals behind Quality Score and Ad Rank, and it's the one most often ignored. A slow page, a page that doesn't name the product or service in the first viewport, or a page that fails on mobile drags that signal down on every keyword that points to it, and in Google's model holding position costs more as a result, though my own within-account data says not to count on the CPC side. The bigger win is usually the conversion rate. Implementation. Test each landing page on a mobile connection throttled to 4G. Time to interactive over four seconds is a conversion killer. Move the primary offer into the first viewport. Match the H1 to the keyword that drives the click. Compress hero images, defer non-critical scripts, and remove third-party tags that haven't earned their load cost. ### Lever 7: pivot to long-tail keywords Head terms carry the highest competition and the highest CPC. Long-tail terms, four words and up, carry a fraction of the volume and usually a lower CPC, often with a higher conversion rate because the searcher intent is more specific. The pivot is a structural change to the keyword list, not a bid change. Implementation. Pull the search-terms report for the last ninety days and the keyword planner for adjacent long-tail terms. Build new ad groups around three to five long-tail clusters per product or service line, each cluster sharing a single landing page and a single ad set. Demote the head term to a lower-bid placeholder or pause it entirely if the long-tail cluster covers the same intent at a better price. ### How to sequence the seven levers Run the levers in the order above. Quality Score components and match-type discipline come first because they fix the inputs the later levers depend on. Don't expect them to move CPC the most; in my within-account comparison, low-score clicks didn't reliably cost more. Layer in audience signals and modifiers next. Shift the bid strategy only after the conversion data has stabilized, and run the five-condition bidding review against the campaigns spending the top eighty percent of monthly budget rather than the whole account. Pull the long-tail pivot last, because it works best on top of an account that already has Quality Score and match types under control. One case sits outside the sequence. When a campaign can't spend its budget at all and the loss has moved from the budget column to the rank column, the levers above are still the right work, and the bid raise that would clear the first-page floor may be worth skipping entirely. That decision is arithmetic rather than judgement, and [I ran it on my own account and left roughly $50 a day unspent](/blog/budget-stopped-being-the-constraint/). Two of the seven CPC levers misallocated on the same account, or two of the five bidding conditions firing on a single campaign, is where the [free 25-page Setup Checklist](/audit/) earns its time as the next read: it covers the ten settings and tracking red flags underneath these levers. With four firing at once, [the free ads audit](/audit-request/) is faster than tuning solo. I write it myself and send it back within 24 hours of read-only access. The [wasted-ad-spend library](/wasted-ad-spend/) covers the structural reasons CPC climbed in the first place, which is the input to picking the right levers in the right order. ## How do I make ad copy more relevant and raise click-through rate? URL: https://connercrowe.com/wasted-ad-spend/ad-copy-relevance-poor-click-through-rates/ Published: 2026-05-25 | Updated: 2026-10-21 Answer: Put the keyword in headline 1, a specific hook in headline 2 and the query's wording in description 1. Relevance lifts CTR more reliably than it cuts CPC, so judge by cost per conversion. Across 30 Google Ads accounts I manage, 30 days to October 21, 2026, low-Quality-Score clicks cost more in only 3 of the 7 I could compare. ### What ad-copy relevance moves, and what it doesn't Quality Score isn't an input to the auction. It's the 1-to-10 summary Google shows you of three things Ad Rank does weigh in real time: expected CTR, ad relevance, and landing-page experience. Two of the three live inside the ad copy itself. Better copy can earn a higher position at the same bid, so it's the cheapest quality fix in the account. Whether that shows up as a lower CPC is a separate question, and my own accounts say don't count on it. When I compared low-score and high-score keywords inside the same account, low Quality Score didn't reliably mean a higher CPC, and few accounts had enough clicks in both bands to compare at all. Most of the Search campaigns I run bid on Maximize Conversions or Maximize Conversion Value. When Google sets the bid each auction, a quality gain can go into position and volume instead of a cheaper click. The score is also missing on a lot of spend. In my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/), a median 38% of keyword spend ran on keywords with no Quality Score reported, and a median 4% sat on keywords scored 7 to 10 (N = 31). So judge a copy rewrite on CTR first and cost per conversion second. CPC is the third read, and the least reliable one. On Meta, check the ad set's goal before blaming the copy for a low CTR. Across 18 Meta accounts I manage in Q3 2026, ad sets optimized for reach drew a 0.06% link CTR, against 2.84% on ad sets optimized for a conversion event (N = 5 and N = 9). A reach goal asks Meta for the cheapest people to show the ad to, with no weight on who clicks. No headline fixes that. The full table is in my [Meta CPM benchmarks](/benchmarks/meta-ads-cpm-by-placement/). Copy fails in two ways, and they need different fixes. Copy nobody clicks is a relevance problem. Copy the wrong people click is an attraction problem, and it costs more because those clicks turn into low-value orders that read as wins inside the platform. Diagnose which one you have before rewriting a headline. ### Signal 1: average order value drops on a specific ad Click-through rate and conversion rate can both look healthy while the basket size collapses. A creative that sells the cheapest SKU in the catalog will print conversions and starve revenue. The leading indicator is AOV segmented by ad, not by campaign. Pull the report from Shopify or GA4 with ad name as the dimension and average order value as the metric. Look for a creative running clearly below the account-wide AOV across at least fifty orders. Below that volume the number is noise. Above it, the copy is selling the entry-level product instead of the catalog. The fix lives in the headline and the primary image. Lead with a mid-tier or premium SKU price point in the visual, name the upgrade benefit in the headline, and drop the "starting at" language that anchors buyers to the cheapest variant. ### Signal 2: new-customer share shrinks on an ad set A profitable ad set can lose its job over time as it stops finding new buyers and starts harvesting people who would have bought anyway. Meta and Google both report new-versus-returning splits at the campaign or ad-set level. Read the trend, not the absolute number. Set a baseline by averaging new-customer share across the trailing thirty days at the account level. An ad set whose share sits well below that baseline for a full fourteen-day window is attracting the wrong audience. The copy is speaking to existing customers when the budget is supposed to find fresh ones. Rewrite the hook to address a first-time buyer concern: sizing, fit, shipping speed, return policy, brand provenance. Cut the loyalty language ("welcome back," "you'll love this again") and the in-group references that only an existing customer parses. Then watch the share rebuild over the next fourteen days. ### Signal 3: ninety-day cohort LTV trails the account average Some ads acquire customers who never come back. The damage hides for ninety days because it lives in the second-order purchase, not the first. The signal is cohort lifetime value pinned to the ad that originated each cohort. Build the report with three columns: first-touch ad, ninety-day repeat-purchase rate, ninety-day cumulative revenue per customer. Read it only on ads with a hundred or more acquired customers, and look for the ones whose ninety-day LTV trails the account cohort average. The copy fix is filtering at the top of the funnel. Discount-led creative ("50% off," "lowest price ever") attracts buyers who came for the discount, not the brand, and they exit after the first purchase. Replace the discount hook with a quality, durability, or use-case hook, then compare the next cohort's full-price share and repeat rate against the one before. ### Signal 4: the search-terms report leans informational On Google Ads the search-terms report is a transcript of who the copy is reaching. Commercial intent reads like "buy walnut dining table" or "walnut dining table for sale." Informational intent reads like "how to clean a walnut dining table," "what is solid walnut," "walnut dining table reviews." Tag the trailing thirty days of search terms by intent. If informational queries make up a visible share of an ad group built for commercial intent, the copy is broad enough to pull in research-stage traffic that won't convert at full price. Tighten match types, add the informational stems as negative keywords ("how," "what is," "vs," "reviews," "guide," "tips"), and rewrite the description to gate the click. Lead with a price point, a stock status, or a buyer commitment cue. Researchers read past the gate and self-select out before the click costs money. For [home and furniture brands](/for-home-brands/furniture/), the copy patterns repeat with one extra gate: shipping lead time inside the description. ### Signal 5: price-objection tickets cluster on one campaign The support inbox is the last reliable lens on who the ads attracted. Tag every ticket with a price-objection field: "too expensive," "do you have a cheaper version," "can you match a competitor's price." Then map the tagged tickets back to the first-touch ad or campaign on the customer record. Look for a campaign whose price-objection rate clearly runs above the account-wide rate, and only once there are a hundred or more tickets to read. Lower volumes don't separate signal from random complaint. Above that, the creative mispromised the price band. The copy fix is the on-creative price cue. Show a representative price in the image, the caption, or the headline. Replace aspirational lifestyle photography that suggests a lower price tier with product-on-white shots at the actual price range. The wrong-budget buyer self-selects out at the impression instead of at the support ticket. ### Reading the five signals together A single signal can mislead. AOV can drop because a seasonal SKU is cheap. New-customer share can fall because the budget moved to retention. LTV can lag because a cohort is young. Two or more signals firing on the same creative is the confirmation. AOV down plus LTV down on one ad is a discount-led hook acquiring transactional buyers. Informational search terms plus price-objection tickets on one campaign is copy that promised a research resource and a budget product at the same time. When four fire at once, that's when to [request the free ads audit](/audit-request/). I write it myself and send it back within 24 hours of read-only access, no call. The sequence is signal first, hypothesis second, copy rewrite third, measurement window fourth. Skip any step and the rewrite reverts to taste. On a wellness brand I audited, ninety days of Meta spend ran at a sub-one-percent click-through rate, on enough impressions to read honestly. The audience looked right, so the copy was the suspect. The hooks were aspirational, and the buyer was looking for permission to spend. The rewrite path was price-anchored creative against the lifestyle set, tested on a controlled hold-out before any cross-campaign rollout. Once the diagnosis is in, the rewrite runs on seven levers. Work them in order. ### Lever 1: the keyword verbatim in headline 1 Headline 1 is the asset Google weighs most heavily for ad relevance. The keyword belongs there in the same form a searcher typed it. A query for "walnut dining table" matches an H1 reading "Walnut Dining Table" far more strongly than one reading "Handcrafted Dining Furniture." The second headline reads better as marketing copy and scores worse on the relevance axis. The implementation rule is one ad group per search theme with the theme keyword pinned to headline 1 position 1 inside the responsive search ad. Pinning is the only way to guarantee placement. Unpinned headlines rotate, and a strong relevance asset can land in position 3 where it carries less weight. Check every ad group for a headline 1 that doesn't contain the top-spend keyword. ### Lever 2: headline 2 as the differentiation hook Headline 1 confirms the searcher found the right product. Headline 2 answers the next question, which is why they should click your ad over the five others on the page. The hook lives in headline 2: the price band, the free shipping threshold, the warranty length, the inventory status, the geographic coverage. Generic differentiation kills CTR. "Quality Service" and "Best Selection" carry no information and read as filler. Specific differentiation earns the click. "Ships in 48 Hours" beats "Fast Shipping" because the timeframe is verifiable. Headline 2 is also where the price cue from Signal 5 belongs on search. Pin it to position 2 once a winner emerges from rotation testing. ### Lever 3: a description that mirrors search intent The first line of description 1 is the second-most-read piece of copy after headline 1. It should restate the searcher's intent in the language of the search itself, then offer the resolution. A query for "outdoor pendant lights waterproof" reads a description that opens with "Waterproof outdoor pendant lights rated for wet locations" and stops reading one that opens with "Illuminate your outdoor space with our latest lighting collection." The rule is mechanical. Take the top three queries by impression for the ad group, find the noun phrase common to all three, and use that phrase inside the first eight words of description 1. ### Lever 4: dynamic keyword insertion with guardrails Dynamic keyword insertion drops the searcher's query into the ad headline at serve time. On a tightly themed ad group, the feature tends to help CTR and saves the work of writing a separate ad per long-tail variant. On a loosely themed ad group, it inserts ungrammatical or irrelevant queries and damages both CTR and brand perception. The guardrail is theme tightness. Use DKI only when every keyword in the ad group describes the same product, the same intent, and the same buyer stage. An ad group that mixes "buy walnut dining table" with "what is solid walnut" shouldn't run DKI, because the buying query and the research query don't share a sensible headline. That's Signal 4 showing up as a structural problem rather than a writing one. Set a default value that reads cleanly when the insertion fails, and check the search-terms report monthly for inserted phrases that misfire. ### Lever 5: responsive search ad asset diversity A responsive search ad accepts up to fifteen headlines and four descriptions. The platform mixes them at auction time and learns which combinations earn clicks. An ad that ships with three or four headlines forces the algorithm to combine the same assets repeatedly and starves the system of learning data. The implementation rule is fifteen headlines per RSA covering five categories: keyword headlines (three or four), benefit headlines (three), proof headlines (two or three) such as review counts or warranty length, urgency headlines (two) such as inventory or shipping cutoff, and brand headlines (one or two). Three or four descriptions follow the same logic. ### Lever 6: sitelinks and callouts kept on theme Sitelinks and callouts extend the ad's real estate, and they help CTR only when the assets match the ad-group theme. Account-level sitelinks pointing at "About Us" and "Shipping Policy" pad the ad with low-relevance links. Build sitelinks at the ad-group level for the top-spend ad groups. Each one should describe a sub-category of the ad-group theme. An ad group for "leather sofa" runs sitelinks for "Brown Leather Sofas," "Sectional Leather Sofas," "Top-Grain Leather Care," and "Sofa Delivery Timeline." Callouts add non-clickable proof points: "Free White-Glove Delivery," "30-Year Frame Warranty," "Showroom Open Saturdays." Both asset types feed the relevance signal. ### Lever 7: auditing RSA asset ratings weekly Google rates each RSA asset as Best, Good, Low, Learning, or Pending. The ratings update as the algorithm collects impressions. Low assets underperform the alternatives and should be replaced. Open the ad view, select an RSA, and click Asset Details. Sort by rating. Any headline or description marked Low after a thousand impressions is a candidate for removal. Replace it with a new variation in the same category and re-check in two weeks. Three Low ratings on a single ad means the whole ad needs a rewrite against the lever stack above rather than incremental swaps. Asset ratings are the closest thing Google gives to a per-asset quality read. They tell you which copy is earning its clicks. The five signals above tell you whether the click is worth having, and cost per conversion tells you whether the rewrite paid. The [wasted-ad-spend library](/wasted-ad-spend/) covers the bidding, targeting, and landing-page layers that compound on top of copy that finally addresses the right buyer. ## How should I structure an ad account to reduce waste? URL: https://connercrowe.com/wasted-ad-spend/ad-account-structure-best-practices-reduce-waste/ Published: 2026-05-25 | Updated: 2026-10-30 Answer: Structure by search intent and keep your brand name out as phrase negatives. Lock the settings: in the 30 days to October 21, 2026, 50 of 51 Search campaigns across 30 accounts I manage targeted people in or regularly in the area. In my Q3 2026 benchmark, the typical account had 38% of keyword spend with no Quality Score. ### Why structure decides how much you waste Account structure is the chassis every other optimization rides on. A clean structure makes waste visible inside an hour of audit time. A messy structure hides waste under blended averages for months, sometimes years. The platforms don't care which structure you pick, because their revenue rises either way. The fix is upstream of bidding, creative, and landing page work. These six decisions cover the structural choices that most often separate accounts running clean from accounts leaking budget. Read the [Wasted Ad Spend hub](/wasted-ad-spend/) for how the diagnostic signals connect. ### Decision 1: segment campaigns by query intent, not by product A common default is one campaign per product line or one campaign per service. That bundling collapses four intent tiers into a single ROAS number. Branded queries (someone searching the company name), non-branded category queries (someone searching the category with no brand in mind), generic top-funnel queries (someone searching a problem), and competitor queries (someone searching a rival) behave nothing alike. Branded converts far better than generic. Competitor traffic usually costs more per click than branded and rarely converts at the same rate. The right pattern is one campaign per intent tier you pay for. Non-branded category sits in its own campaign with the bulk of the budget and conversion-based bidding. Generic and competitor run as separate test campaigns with strict CPA caps. Brand doesn't get a campaign by default. My rule is to stay out of the auction for the client's own name: the brand name and its misspellings go in as phrase negatives across the account, and the organic listing takes those searches. Pulling them out shows how much of the blended result was brand demand quietly covering the losses on competitor and generic. The one exception is a competitor's ad holding your brand search above your organic result. Then a brand campaign is worth testing, with a capped budget and exact match, kept separate so it can't inflate non-brand numbers. Move the brand negatives from the account level to a shared list applied to every non-brand campaign, or they'll block the brand campaign too. [High CPC without sales](/wasted-ad-spend/common-reasons-high-cpc-without-sales/) covers that test. Network settings belong in the same decision, because they decide which auctions each campaign enters. Across 30 Google Ads accounts I manage, 30 days to October 21, 2026, Search Partners was on in 1 of 51 Search campaigns, carrying 1.1% of spend, and Display expansion was on in none. A Search campaign that also buys partner and Display inventory reports one number for three kinds of traffic. ### Decision 2: one match type per ad group on tighter accounts Mixed match types inside a single ad group create internal auctions. The broad keyword cannibalizes the exact match keyword, the system optimizes against whatever bid is highest, and the search terms report becomes hard to read. On accounts under twenty thousand a month, the surface area is small enough that this matters every week. The right pattern on smaller accounts is one match type per ad group. Exact match ad group, phrase match ad group, broad match ad group, each with the same keyword theme but different intent thresholds. Negative keywords flow up the funnel: exact match terms become negatives in the phrase ad group, phrase match terms become negatives in the broad ad group. The search terms report now reads cleanly per match type, and the cannibalization stops. Larger accounts above fifty thousand a month can sometimes run mixed match types inside Smart Bidding ad groups, but only with a weekly search terms review. Skip the review and the broad keyword eats the budget inside thirty days. Keyword sprawl hides a second cost. In my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/), the typical account had 38% of its keyword spend on keywords with no Quality Score reported (N = 31). Counted separately, the typical account had 4% on QS 7 to 10 (N = 31). Unscored keywords give you no read on ad relevance or landing-page fit, so an ad group full of them can't tell you which half to rewrite. Fewer keywords per theme, each with enough traffic to earn a score, makes the structure readable. ### Decision 3: split Performance Max asset groups by audience tier A Performance Max campaign with a single asset group treats every product and every audience the same. The campaign reports one ROAS, the algorithm picks its own winners, and there's no way to see whether the spend went against best-sellers shown to existing customers or against new SKUs shown to cold prospects. That's exactly how Performance Max budget pools end up over-indexing on retargeting that was already going to convert. The right pattern is asset groups split by audience tier. One asset group for prospecting with cold-only audience signals and no customer-match lists attached. A second for retargeting with site-visitor and cart-abandoner signals. A third, sometimes, for existing-customer expansion with purchaser lists. Each asset group gets its own creative set, its own product feed segment if the catalog supports it, and its own performance read. On a Shopify account running Performance Max, this split can show prospecting costing far more per sale than the blended ROAS suggested, while retargeting ran above target. The fix is reallocating budget. The campaign stays. [Furniture brands running PMax](/for-home-brands/furniture/) hit this same asset-group split with one extra tier for catalog-heavy SKUs. ### Decision 4: separate geographies when service areas matter Service businesses with distinct service areas often run a single campaign targeting all of them. Cost per click in a dense urban market can run several times the cost per click in the surrounding suburbs. A blended campaign hides that. Worse, when one market has lower conversion intent than another, the campaign optimizes against the cheaper clicks regardless of close rate. The right pattern is one campaign per service area when close rates or unit economics differ. A roofing company serving a metro and three suburbs runs four campaigns, each with its own budget scaled to demand and capacity. A multi-location retail brand runs one campaign per distribution radius. The threshold to skip the split is when the markets have nearly identical CPC, conversion rate, and lifetime value. Whichever split you choose, check the location option on every campaign. Google's default reaches people searching about a place as well as people in it. Across 30 Google Ads accounts I manage, 30 days to October 21, 2026, 50 of 51 Search campaigns use Presence targeting, which reaches only people in or regularly in the targeted area. A service business that can't work outside its area has no reason to pay for the rest. For Shopify brands shipping nationally, geographic separation matters less except where shipping cost or seasonal demand swings hard between regions. On [legal accounts with practice areas across markets](/for-service-brands/law-firms/), the segmentation tightens further because intake teams can't handle blended jurisdictions. ### Decision 5: scope budgets at the campaign level, not the shared pool Shared budgets across campaigns sound efficient. They almost never are. A shared budget lets the platform shift spend toward whichever campaign it scores highest, and platform scoring favors volume and click-through rate over revenue quality. A high-CTR campaign, such as a brand campaign run under the competitor exception, pulls spend away from the non-branded campaign that produces incremental revenue. The right pattern is campaign-level budgets that match the strategic priority for each intent tier. Non-branded gets the bulk allocation. An exception brand campaign gets a capped budget sized to the brand searches the competitor is contesting, with overflow blocked. Test campaigns (generic, competitor, new geography) get explicit small budgets that fail fast if performance doesn't clear the bar. Shared budgets belong only on portfolio bid strategies where two campaigns target the same intent and the same audience, which is rare in practice. A campaign-level budget also lets each campaign carry the bid strategy its intent needs. Across the 51 Search campaigns in the accounts I manage, 30 days to October 21, 2026, 35 run Maximize Conversions, 8 Maximize Clicks, 4 Maximize Conversion Value, 3 Manual CPC and 1 Target Impression Share. A pooled budget makes campaigns on different strategies compete for one pot. To see what the current structure is leaking, my [free Google Ads audit](/audit-request/) ranks the top three leaks by spend impact, in writing, within 24 hours. ### Decision 6: a naming convention that surfaces structure in every report The deepest cause of structural drift is unreadable account history. Six months into an account, the original builder has often left, and the campaign names look like Campaign 1, Campaign 1 (copy), Final Campaign, and Test 3. Audits take days instead of hours because nobody can tell what each campaign was supposed to do. The right pattern is a naming convention that encodes intent, geography, match type, and date in every campaign and ad group name. A workable format: Platform | Intent | Geography | MatchType | LaunchDate. So Google | NonBranded | Metro | Phrase | 2026-05 reads cleanly in any column. Pivot tables and Looker Studio reports group automatically. New media buyers onboard in an afternoon instead of a week. The convention itself costs nothing. The absence of it costs every future audit hour. ### Where to start Apply decision one first. The intent split surfaces where the real money is going inside the first week and makes the next five decisions easier to scope. Decision four matters only for service businesses with real geographic spread. Decision six pays back every audit cycle for the life of the account. Founders who want a senior pair of eyes on the structure can [request my free written audit](/audit-request/): 5 to 7 pages, back within 24 hours of read-only access, no call needed. If you'd rather work it yourself, grab the [free 25-page Setup Checklist](/audit/) first. The same six-decision sweep is the first thing I check in every account I audit. ## Is Performance Max wasting my budget? URL: https://connercrowe.com/wasted-ad-spend/is-performance-max-wasting-my-budget/ Published: 2026-07-02 | Updated: 2026-10-21 Answer: Sometimes, and blended ROAS won't show it. In two home-brand accounts I manage, Performance Max took 82% of spend at 12x while Standard Shopping returned 61x on 18%, blending to 21x. Split the two, keep PMax off your own brand name, and confirm its conversions are purchases before judging either. ### Blended ROAS is the number that hides the answer Most founders judge Performance Max on the account's blended return. That number averages two different jobs into one grade. Here's a real example from my own book. Across two home-brand Shopify accounts over ninety days to late June 2026, the blended return came out to 21x. Nobody looking at that number asks whether anything is being wasted. Split it by campaign type and the story changes. Standard Shopping took 18 percent of the spend and produced 52 percent of the revenue at 61x. Performance Max took 82 percent of the spend and produced 48 percent of the revenue at 12x. Per brand, PMax returned 17x on one and 8x on the other. Shopping was mostly catching buyers already close to a purchase. PMax was doing the more expensive work of finding new ones. The blended 21x hid a 5x gap between the two jobs. The full breakdown is in [what a home brand's 21x ROAS hides](/blog/what-a-home-brands-21x-roas-actually-hides/). Neither number means PMax was failing. Break-even ROAS is one divided by contribution margin, so 12x clears break-even at any contribution margin above about 8.3 percent. The point is that you can't know until you separate the jobs. ### Three signals the waste is real **The conversion signal feeding it is wrong.** Performance Max is an amplifier. It optimizes toward whatever conversion data you give it, so broken purchase values, duplicate events, or cart events counted as conversions send it shopping for the wrong customers at full price. When I audited [Emma Grace Home](https://emmagracehome.com/), its Google Ads was [optimizing toward add-to-carts](/results/unassigned-revenue-dual-ga4-fix/) instead of buyers. Broken conversion tracking is the most common finding in the audits I ran in 2026, and it's invisible from inside the campaign screen. On two of three ecommerce accounts in my [tracking-accuracy benchmark](/benchmarks/ecommerce-tracking-accuracy/), add-to-cart and begin-checkout events made up 19.8% and 25.8% of the reported conversion value over 90 days. Nothing in the interface labels it. The column still reads "Conv. value." I wrote up the checks in [the conversion tracking library](/conversion-tracking/). **It takes credit for demand you already owned.** Out of the box, PMax can serve on searches for your own brand name and on your remarketing pools. Those buyers were already looking for you, so the conversions are cheap and the reported ROAS climbs without PMax creating anything. My default on every account is that no campaign bids on the client's own brand. Add the brand name and its misspellings as phrase-match negatives at the account level, which Performance Max respects, or apply a brand exclusion list on the PMax campaign. The one exception I make is when a competitor's ad holds your brand search above your organic result, and then I move the brand negatives from the account level to a shared list applied to every non-brand campaign. The mechanics of PMax and Shopping trading credit are in [why Performance Max gets credit for Shopping conversions](/blog/why-performance-max-gets-credit-for-shopping-conversions/). **One campaign carries every product and audience.** A single PMax campaign with one asset group across a whole catalog gives the algorithm nothing to differentiate. Margins, price points, and intent levels all get averaged, and the budget flows to whatever converts easiest rather than what earns most. I run [a six-campaign PMax matrix](/blog/why-i-run-six-performance-max-campaigns-instead-of-one/) on catalog accounts for this reason. ### How to run the split on your own account 1. Segment the last ninety days by campaign type. Note spend share and revenue share for Performance Max, Standard Shopping, and Search separately. 2. Pull the PMax search terms and search categories insight. If your own brand name or its misspellings show up with real spend, the brand negatives or brand exclusion aren't in place. Fix that before you judge the campaign, because those conversions would have happened anyway. 3. Open the conversion-action list and read the category of every action in the bidding metric. If add-to-cart or begin-checkout is counted next to purchase, the ROAS you're judging includes cart value. 4. Compare each campaign type against its own job. Harvest campaigns should run several multiples above your blended target. Prospecting should clear your break-even ROAS with room to spare. 5. Check the learning phase before judging anything. A new or heavily edited PMax campaign spends one to two weeks recalibrating, and cost per acquisition can run well above target during that window. ### What a healthy PMax returns In the two accounts above, PMax ran between 8x and 17x depending on the brand, while the Shopping layer ran far higher. If your PMax sits below break-even ROAS for thirty days or more with clean tracking, brand excluded, and a stable campaign, that's waste, and the fix is structural rather than a bid change. Even a PMax that clears break-even on paper is reporting attributed revenue, which isn't the same as revenue it caused. On one home-furnishings brand, a 21-day geo holdout of all paid media showed 38% real lift against the 92% of revenue the attribution report called paid-influenced. That's one brand, and the test is in [the 21-day geo-holdout write-up](/blog/21-day-geo-holdout-incrementality-test/). Run one before you scale PMax on its reported ROAS. ### The fix order Tracking first, structure second, creative third. Verify the purchase event, its values, and deduplication before touching the campaign, and take cart events out of the bidding metric. Then keep PMax off your brand and separate harvest from prospecting so each job gets judged and budgeted on its own numbers. Only then work the asset groups. Fixing them in the reverse order optimizes a campaign against data you can't trust. ## Is there a free Microsoft Ads (Bing Ads) audit? URL: https://connercrowe.com/wasted-ad-spend/free-microsoft-ads-bing-audit/ Published: 2026-10-21 | Updated: 2026-10-22 Answer: Yes. Microsoft offers a free consultation and an optimization score, and third-party tools score accounts automatically. I write a free Microsoft Ads audit at any spend level, back within 24 hours of read-only access. I start with the UET tag: on two stores I've worked on, it recorded nothing usable, one of them for six months. Most Microsoft Advertising accounts started life as a Google Ads import. That's why they get audited badly: the reviewer runs a Google checklist, finds the same campaigns, and stops. The leaks that matter on Bing sit in Microsoft-only settings. ### What Microsoft gives you for free Microsoft runs a [free consultation](https://about.ads.microsoft.com/en/resources/learn/free-consultation/united-states) with its own team, by phone or chat, from setup through optimization. It's worth a call if you're new to the platform. It isn't an independent read, because the team's job is to grow your use of Microsoft's products. The second free option is the [optimization score](https://learn.microsoft.com/en-us/advertising/msa-help/hlp_ba_conc_optscore) on the Recommendations page, a 0 to 100 percent rating for active Search, Shopping and Audience campaigns. Microsoft says you reach 100 percent by applying or dismissing every recommendation, so the score measures how much of Microsoft's suggested work you've applied or dismissed. It doesn't measure whether the account makes money. While you're on that page, check the auto-apply setting behind the lightning-bolt icon. Microsoft can [apply recommendations automatically](https://github.com/MicrosoftDocs/Advertising/blob/main/advertising/msa-help/hlp_BA_CONC_AutoApplyRec.md) after a seven-day review window, including new keywords, broad match keywords and new ads. Accounts were [opted in by default](https://about.ads.microsoft.com/en/blog/post/february-2022/optimize-your-ad-campaigns-with-auto-apply-recommendations) when the feature launched in 2022. Untick every type you haven't chosen on purpose. Third-party tools fill the rest of the free tier. They connect to the account, compare settings against a checklist, and return a score. That's useful for hygiene. None of them knows what a sale is worth to you, so they can't tell you whether the conversions Microsoft reports are real. ### Is Google Import overwriting your Microsoft edits? Check this second, after the tag, because it undoes every other fix. An imported campaign stays [permanently linked](https://learn.microsoft.com/en-us/advertising/msa-help/hlp_ba_conc_importwhatinfo) to its Google source. On a scheduled import, Microsoft's [update options](https://learn.microsoft.com/en-us/advertising/campaign-management-service/googleimportoption?view=bingads-13) are all on by default. They cover bids, budgets, campaign and keyword status, keyword URLs, ad distribution and conversion goals. Lower a bid in Microsoft Ads on Monday and the Tuesday import can put the Google bid back. Shared negative keyword lists are worse. Import merges the keywords from both platforms, and Microsoft's own documentation says the Microsoft list associations get overwritten. A negative list you attached to one campaign in Bing can come off without warning. A few things don't sync the way people assume. Match type copies over on the first import and never updates after that, so phrase match stays phrase match. Broad match negatives arrive as phrase match negatives. Bid strategies Microsoft doesn't support become Enhanced CPC, or manual CPC for Audience campaigns. Google's "Search Network plus search partners" setting maps to every Microsoft search network, syndicated partners included. Pick one platform as the source of truth for each item. If you manage bids and negatives inside Microsoft Ads, turn those update options off or stop the scheduled import. ### Where does the budget go outside Bing search? [Ad distribution](https://learn.microsoft.com/en-us/advertising/msa-help/hlp_ba_conc_aboutaddistributionhidden) is set per ad group, with two choices: the entire Microsoft Advertising Network, or Microsoft sites and select traffic. Both include Audience ads. Microsoft removed the clean opt-out for Search campaigns in early 2023, so the narrower option plus exclusions is the only way to limit that traffic. To see where the money went, run the website URL (publisher) report and sort by spend against conversions. Partner sites that take spend and return nothing go on the [exclusion list](https://learn.microsoft.com/en-us/advertising/msa-help/hlp_ba_proc_addexclusions), which holds up to 2,500 URLs on a campaign, and more in a shared exclusion list. You can't exclude Bing, AOL or Yahoo search itself, and the "select traffic" bucket reports in aggregate only. Ads in Copilot follow the same pattern. Eligible campaigns are [opted in automatically](https://learn.microsoft.com/en-us/advertising/msa-help/hlp_ba_conc_adsforcopilot) and advertisers can't opt out. Search ads need a logo extension to qualify. Your negative keywords still apply, and there's no Copilot-only report yet, so negative keywords are your only control. ### Is UET counting real conversions? Microsoft tracks conversions through its UET tag, and an imported account often doesn't have working goals. Import only brings over Google Ads tag-based and destination-URL goals. GA4 goals don't come across, and data-driven attribution becomes last click. The failure I remember best was on an inherited Shopify store. Three Microsoft Ads conversion tags had been [firing the placeholder text YOUR_VALUE_HERE](/blog/bing-pixel-your-value-here/) for six months, so the Bing campaigns were optimizing toward noise. No score flagged it, because the tags were firing. They were sending a template string where the conversion ID belonged. A second store, on WooCommerce, failed the other way. In June 2026 its Microsoft account had spent $255.50 on 853 clicks so far that month with zero conversions recorded, and Microsoft's overview flagged a goal with inactive tags. I had created those tags through the Google Tag Manager API, and the API had stored one setting, document-write support, as the text "false" rather than the value false. Any non-empty text counts as true, so the tags loaded in a mode that silently drops the script once the page has finished loading. Rebuilding them without that setting brought the Bing beacon back: it returned a success code on the live site. Both failures had a tag that existed and looked configured. The check that catches them is watching the purchase page in the browser's network panel for the request to bat.bing.com, then confirming a conversion shows up in Microsoft within a day. While that panel is open, filter for `clarity.ms/tag/uet/` too. A UET tag can load Clarity session recording on its own, and in my spot check of 32 sites running the tag, [it was the only thing loading Clarity on 4](/blog/bing-ads-tag-loads-clarity-session-recording/). Two more checks belong here. [Enhanced conversions](https://learn.microsoft.com/en-us/advertising/msa-help/hlp_ba_conc_uet_enhancedconversions) can be turned on per goal and recover matches that browser tracking loses. If you serve the EEA, the UK or Switzerland, [consent mode](https://learn.microsoft.com/en-us/advertising/msa-help/hlp_ba_conc_uet_consentfaq) has been enforced since May 5, 2025, and without it Microsoft stops recording UET conversions there. Then reconcile. My rule of thumb: Microsoft-reported conversions should sit within ten percent of the sales or qualified leads your own system attributes to Bing, and a gap past 25 percent points to a tracking problem, which no bid change fixes. In between, check attribution windows and conversion counting before anything else. The [Tracking Stack](/frameworks/tracking-stack/) covers the reconciliation in full. ### LinkedIn profile targeting [LinkedIn profile targeting](https://learn.microsoft.com/en-us/advertising/msa-help/hlp_ba_conc_linkedintargeting) lets you raise or lower bids by company, industry and job function, and since mid-2026 by job seniority on Search and Audience campaigns ([Microsoft, August 2026](https://about.ads.microsoft.com/en/blog/post/august-2026/ai-max-for-search-and-other-product-news-for-august-2026)), in a limited set of markets that includes the US. It works as bid-only: it adjusts what you pay for those people and doesn't narrow who sees the ad, unless you add exclusions. For B2B, legal and high-ticket services, a bid increase on the job functions that buy is, in my view, the strongest reason to run Bing at all. For most ecommerce brands it doesn't change much. ### When a tool score is enough A tool score is enough when the account was built natively in Microsoft Ads, conversions reconcile against your own records, and you only want a settings check. Get a human read when the account came from Google Import, when Microsoft's conversion count and your sales count disagree, or when Bing spend is big enough that partner and Copilot traffic matter. Microsoft's head of search [cited Comscore data](https://www.theverge.com/analysis/717167/bing-market-share-growth-google-ai) putting Bing near 29 percent of US desktop searches in mid-2025. Across all devices worldwide the share is far smaller, so Bing earns the most where buyers search from a work computer. ### What my free Microsoft Ads audit covers The terms match my Google Ads audit. It's free at any spend level, written by me, 5 to 7 pages, and back within 24 hours. Add me as a user with the read-only Viewer role, then remove the user from the same page once you have the audit. No call is required and no retainer comes attached. It covers the import sync, ad distribution and partner spend, conversion goals checked against your own sales or lead records, the auto-apply settings, what the UET tag loads on your live site (Clarity included), and one fix you can ship without me. If you also run Google Ads, I read both accounts together, because most Bing problems start on the Google side. Submit the account through [the free audit request form](/audit-request/). ### Common questions #### Is the Microsoft Ads audit really free? Yes, at any spend level, with no retainer attached. #### How long does it take? I send the written audit within 24 hours of read-only access. #### Do I need a call? No. The audit is written, 5 to 7 pages, and you can act on it without talking to me. #### What access do you need? The read-only Viewer role on the Microsoft Advertising account. Remove the user from the same page once you have the audit. ## My campaign is not spending its budget. Should I raise bids? URL: https://connercrowe.com/wasted-ad-spend/should-i-raise-bids-when-the-budget-is-not-spending/ Published: 2026-09-06 | Updated: 2026-10-22 Answer: Only if the arithmetic clears. In Q3 2026, 23 of 31 Search accounts I manage lost more to budget than rank. If rank is the limit, multiply the first-page bid by clicks per lead. My own account lost 75.1% to rank at a 1.33% lead rate, which priced a lead at $1,200 to $1,700, so I didn't raise. ### Why the underspend is almost never a budget question A campaign that ends the day under its cap is telling you it couldn't buy the inventory. Two reasons account for most of it: either it ran out of money before the day ran out, or it lost the auctions it entered. A campaign can also underspend because search volume is thin, targeting is narrow, a Smart Bidding target is set too tight, or an ad or keyword isn't eligible, and none of those show in the impression-share split. This answer covers the first two. Google reports those separately in the campaigns view, as Search Lost IS (budget) and Search Lost IS (rank), and the two point at opposite fixes. Lost to budget means the auction wanted to spend more than you gave it. Lost to rank means your bid and your ad quality couldn't clear the price. A campaign sitting at zero percent lost to budget isn't constrained by its budget at all, and raising the number in the budget field changes nothing. A campaign that can't spend is the less common case in the accounts I manage. In my Q3 2026 benchmark, 23 of 31 Search accounts lost more impression share to budget than to Ad Rank (N = 31), and the median account lost 37.9% to budget and 26.2% to rank. If yours ends the day under its cap, it's probably in the minority where rank is the limit, so read the split rather than assume it. The budget-capped majority shows up on the clock instead. Of 20 lead-gen accounts reporting spend by hour, the 12 that lost 35% or more of impression share to budget put a median 60% of their spend into hours 7 through 11, against 43.5% for the 8 that lost less (N = 12 and N = 8). The rows are in my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/). One thing to know before you read the split. Impression share and the two lost-IS columns add up to 100%, so a campaign losing most of its impressions to budget has a hard ceiling on what its rank column can report. Fix the budget loss and the rank number climbs on its own without the auction having changed. The [answer on which metrics to monitor weekly](/wasted-ad-spend/key-metrics-monitor-detect-poor-ad-performance/) covers where those columns live and what a normal week looks like. ### When rank is the constraint, the bid raise becomes a math problem Once the rank column has taken over, raising bids is a live option rather than an obvious one, and there's a short calculation that settles it. It needs four inputs you already have. The first-page bid estimate on your top keywords, from the keywords view. Google defines it as the bid you'd likely need for your ad to show anywhere on the first page. It computes that from the keyword's own Quality Score and from what competitors are bidding, so on an account with weak landing pages it sits well above the market CPC. That gap is the price of your own account. Your measured lead rate, from the same account over the same window. Leads divided by clicks. Use what the account has done, not what you hope it'll do after a fix you haven't shipped. Your close rate, meaning the share of leads that turn into paying work. This is the input people skip, and skipping it's what makes a cheap-looking cost per lead lie. What a sale is worth to you. Then: divide one by the lead rate to get clicks per lead, multiply by the bid it takes to clear the floor, divide by the close rate, and compare the result to what a sale earns. Cost per booked job is cost per lead divided by close rate, and that division is where most bid-raise decisions turn over. If the answer lands above what the sale is worth, the raise buys volume you can't pay for. Leave the money unspent and work the rate instead. Read the bid-based figure as a ceiling. You pay a cent over the next bidder rather than your own cap, so the arithmetic gives you the worst case. ### The math on my own account I ran exactly this on my own lead-generation account in August 2026. Between 21 and 25 August it spent $51.04 across 8 clicks against a $64.52 daily budget, at 0.0% impression share lost to budget and 75.1% lost to rank. My keyword caps were $6.50 against first-page estimates of $13 to $32. The account had measured one lead across 75 clicks that month, a 1.33% lead rate. At 75 clicks per lead, floor-clearing bids of $16 to $23 put cost per lead between $1,200 and $1,700 as a ceiling. Against an offer with a $5,000 floor, that needs a close rate somewhere between 24% and 34% to return the offer price before any of the work gets done. One lead isn't a close rate, so I had no denominator for the fourth step and no basis for the raise. The caps stayed and roughly $50 a day goes unspent. The full sequence, including the keyword cull and the bid cap that created the condition in the first place, is in [why I leave about $50 a day unspent in my own Google Ads account](/blog/budget-stopped-being-the-constraint/). ### The three cases where you should raise the bid Raising is right more often than the above makes it sound. Three conditions each justify it on their own. The lost-to-budget column is the one moving. More budget or a higher bid both buy real incremental volume here, because the auction has already shown it wants to spend. Your close rate is measured rather than guessed, and the floor-clearing cost per sale still clears what a sale earns. Two of those three inputs are yours rather than Google's, which is the reason this decision can't be made from the ads interface alone. You're buying data rather than leads, deliberately, with a number attached and a stop date. That's a legitimate reason to overpay, and it stops being legitimate the moment it has no ceiling. ### What raising the bid won't fix A bid raise moves your position in the auction. It doesn't move the price of that position, and on most underspending accounts the price is the problem. Google treats Quality Score as a diagnostic rather than an auction input, but the components underneath it describe the same thing the auction reads through Ad Rank, and the first-page estimate moves with them. So if the landing-page-experience component reads below average across your top keywords while ad relevance reads above average, the ads are being read as relevant and the page is what makes the floor expensive. Every extra dollar of bid can pay that difference again. The page is the cheaper fix. The non-bid levers that move the price rather than the position are the [seven non-bid levers on cost per conversion](/wasted-ad-spend/strategies-lower-cpc-without-sacrificing-conversions/). Work those first. If an ad group is frozen at zero impressions while the campaign is under budget, [the pause rules](/wasted-ad-spend/when-to-pause-underperforming-keywords-ad-groups/) call that a rebuild rather than a pause, and the arithmetic above is how you decide whether the rebuild is worth funding. ### Where to start this afternoon Add both impression-share columns and read the loss split. If rank owns it, price the floor from the keywords view, price a lead at the rate the account has measured, and then find your close rate, which will probably mean asking your sales records rather than your ad account. The [free 25-page Setup Checklist](/audit/) covers the ten account settings I check before any of this, which is the layer underneath the decision rather than the decision itself. ## Should I build SEO pages for misspellings of my brand name? URL: https://connercrowe.com/wasted-ad-spend/seo-pages-for-misspelled-brand-name/ Published: 2026-08-03 | Updated: 2026-10-21 Answer: No. A page per misspelling is doorway abuse under Google's spam policies, so use alternateName schema. In paid, add your brand and misspellings as phrase negatives. Bid only if a competitor's ad outranks your organic result. Brands built on a deliberate misspelling all inherit the same problem. Flickr wanted flicker.com, couldn't buy it from the owner, and dropped the e. Lyft reads as a respelling of lift. Every name in that class has a correctly-spelled string sitting out there that people type instead, and the founder's instinct is to write content for each variant so the site catches the traffic. That instinct is right that the traffic exists and wrong about how to catch it. ### The asymmetry that decides it Paid search lets you suppress a query. Organic doesn't. My default in paid is to stay out of the auction for your own name entirely. The brand name and every misspelling go in as phrase negatives across the account, so no campaign pays for a searcher the organic listing would have caught free. The negatives are the load-bearing part, because [exact match no longer means exact](/wasted-ad-spend/typical-mistakes-keyword-matching-wasted-spend/) and close variants pull misspellings into non-brand keywords on their own. Brand clicks also flatter the results they land in. On accounts I take over, brand searches often carry a large share of the recorded conversions, and some of those buyers would likely have come through the organic listing anyway, as [high CPC without sales](/wasted-ad-spend/common-reasons-high-cpc-without-sales/) covers. The one exception is a competitor's ad holding the misspelled search above your organic result and taking the click. Then a brand campaign on that variant is worth testing: exact match, a bid ceiling, kept separate so it can't inflate non-brand numbers. Move the brand negatives from the account level to a shared list applied to every non-brand campaign, so they don't block it. Add the colliding correctly-spelled term as a phrase negative in it, so you don't pay for the searcher who wanted someone else. If the competitor sits below your listing, leave them to pay for clicks your organic result still wins. A page on your site has no query-level equivalent. You can `noindex` it, or canonicalize it into another page, but those are on-off switches for the whole URL. There's no setting that keeps the page and declines the subset of searchers you didn't want. If the page ranks for the correctly-spelled generic, it's served to everyone who types that string, including the majority who wanted a different company's product. So the site's job is narrower than a content program. It tells search engines the variants are you, and the organic result catches the searches. Paid stays out unless a competitor forces it in. ### Check who owns the correctly-spelled string first This is the step that changes the answer, and it comes before any recommendation. Search the correctly-spelled version of your name and see what comes back. If an established company already owns that string in a different category, chasing it organically means competing for traffic that doesn't want you, and bidding it bare means paying for the same. In that case the unqualified term belongs in your account as a permanent negative, and if a competitor ever forces you to bid on your own name, the only versions worth bidding are the spelling that's uniquely yours and the qualified long tail, where a category word deselects the other company's audience. Related: [common reasons for high CPC without sales](/wasted-ad-spend/common-reasons-high-cpc-without-sales/) covers the reverse case, where someone else is bidding on a brand string that's unambiguously yours and the auction price is the first symptom. ### What to do on the site The legitimate on-site fix is schema plumbing, not a content program. It's roughly two hours of work. 1. **Add `alternateName` to your Organization schema.** Google lists it as a recommended property in its [Organization structured data documentation](https://developers.google.com/search/docs/appearance/structured-data/organization), described as another common name your organization goes by. Claim only the variants uniquely yours, which usually means the spoken-then-typed split of your name and any spelling your own legacy material already used. 2. **Fix your own inconsistencies first.** Old policy pages and footers routinely carry a spelling the company abandoned. Those pages are naming a nonexistent entity and they work against consolidating you into one. 3. **Answer the spelling question once, visibly,** in a single line on a page that already exists. 4. **Write the rationale into a comment next to the schema,** or a future editor will helpfully complete the list and undo the point of it. Two things to avoid. Don't build one page per spelling. Google's spam policies call this [doorway abuse](https://developers.google.com/search/docs/essentials/spam-policies), defined as creating pages to rank for specific, similar search queries, with substantially similar pages named as an example. Thin variant pages sometimes do rank on a misspelling nobody else contests, which is exactly why the tactic tempts people, and that's a separate question from whether the pattern is one you want assessed across your whole site. Don't reach for `FAQPage` schema either. Google stopped showing FAQ rich results on 7 May 2026, removed the Search Console reporting and Rich Results Test support in June, and dropped the API support in August, so the markup now earns nothing in Search. That leaves only the semantic argument for adding it, and a name-variant question fails that too: `FAQPage` sets the page's `mainEntity` to the questions it lists, which misdescribes a page that's about your product. Leaving existing FAQ markup in place does no harm, so this is an argument against adding it, not a cleanup job. ### `alternateName` is an identity claim, not a keyword slot The field says "this entity is also known as this." Claiming a string you don't own invites entity confusion at best and a trademark argument at worst. Restraint is the discipline here. What teaches search engines your name variants is mostly off-site anyway. Retailer listings, directories, review profiles, and social handles spelling you consistently outweigh anything you can put in your own markup. ### Then stop guessing Two reports show which spellings humans type. In Search Console, filter the Performance report's queries by a fragment of your name and read the variants that already reach you organically. In Google Ads, before the phrase negatives go in, the search terms report shows the variants your non-brand campaigns were paying for. Every variant list written before reading those two, including the one you're about to write, is inference. Add what you find to the negative list, add the variants that are uniquely yours to `alternateName`, then re-read both after thirty days and let them correct you. If you can't tell how much of your paid traffic is people typing your own name, that split is one of the checks in the [25-page Setup Checklist](/audit/), or [request a free ads audit](/audit-request/) and I’ll read the account myself. ## What are common mistakes that cause digital ad spending to go to waste? URL: https://connercrowe.com/wasted-ad-spend/common-mistakes-causing-wasted-digital-ad-spend/ Published: 2026-05-25 | Updated: 2026-10-22 Answer: The most common mistake in my 2026 audits is broken conversion tracking: 6 of 29 lead-gen accounts I manage counted a non-lead action or click proxy as primary in Q3. The others are broad match without negatives, accepting recommendations blind, paying for your own brand name, untested landing pages, missing negative audiences, and skipping offline conversion imports. ### Why these mistakes survive in six-figure accounts Most of these mistakes look like sensible defaults when a founder or a junior media buyer sets the account up. The platforms reward each one with a higher optimization score, which is why they persist. Read them in order. Fix tracking first, because every later fix reads from the conversion data. ### Broad match running without a negative-keyword discipline Broad match is the default in any new Google Ads search campaign and the only match type Performance Max uses under the hood. It learns from your conversion data, your landing pages, and the queries that show intent. With no negative-keyword list and no review cadence, it learns the wrong lessons fast. The fix lives in the search terms report. Filter to the last ninety days, sort by impressions descending, and read the top one hundred queries. Push everything irrelevant into a shared negative-keyword list at the account level. Add a recurring calendar block, weekly for an account spending over fifteen thousand a month, monthly for everything else. The list is never finished. ### Conversion tracking that under-counts the actual outcome On Shopify, the default Google Ads tag and the default Meta pixel both miss orders. Browser-side pixels lose a share of conversions to ad blockers, iOS privacy settings, and Safari ITP, and how big that share is varies by store and audience. Server-side conversion tracking, properly de-duplicated with the browser pixel, closes most of that gap. The result of under-counting is that the algorithm optimizes against incomplete data. Campaigns that look weak are often the strongest. The [Tracking Stack reference](/frameworks/tracking-stack/) covers the exact de-duplication contract between the pixel and the server event. If reported conversions sit more than ten percent below Shopify orders over thirty days, the tracking is the first thing to fix. Every downstream optimization depends on it. The most common finding in my 2026 audits is tracking that feeds Google the wrong signal: the wrong primary action, a dead tag, or phone taps counted as leads. In Q3 2026, 6 of 29 lead-gen accounts in my [Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/) counted a non-lead action or a click proxy as a primary conversion (N = 29). On one contractor account, someone renamed a form button from Send to Submit and [silently killed the only website conversion](/results/dead-form-conversion-fix/), so every form fill counted for nothing. Counting every phone call does the same damage from the other side. Picture ten calls in a week. Four are wrong numbers. Two are hangups inside thirty seconds. Four are real prospects. Google optimizes against ten when only four were buyers. The fix is lead-tracking software that imports only the qualified calls into the conversion column. ### Accepting the recommendations tab without reading it The optimization score is a sales tool, not a performance tool. The recommendations tab will tell you to raise budgets, add broad match keywords, opt into Display expansion, and turn on auto-applied recommendations. Each one of those increases Google's revenue. Most of them don't increase yours. The fix is a standing rule: nothing from the recommendations tab gets applied without a documented reason. Dismissing counts too. [I dismissed 1,476 recommendations](/blog/google-optimization-score-recommendations/) across twenty accounts, and dismissing raises the Optimization Score the same as accepting does. Turn off auto-applied recommendations under settings. Read each suggestion. Apply the ones that match the strategy. Dismiss the rest. A founder who applies every recommendation for six months ends up running a campaign structure that Google designed, not one that fits the business. ### Paying for your own brand name A broad or phrase keyword in a non-brand campaign matches searches for the business's own name, and the click the organic listing would have delivered free now costs money. Brand terms convert far better than category terms, so the campaign's ROAS or cost per lead looks healthy while the non-brand spend quietly loses money. My default is to stay out of that auction. The brand name and its misspellings go in as phrase negatives across the account, Performance Max included, and the organic listing takes those searches. The reported numbers on every remaining campaign then describe non-brand demand, and budget decisions become honest. The one exception is a competitor's ad holding your brand search above your organic result. Then a separate brand campaign is worth testing, with a bid ceiling, kept out of every non-brand number, and the brand negatives move from the account level to a shared list applied to every non-brand campaign. [High CPC without sales](/wasted-ad-spend/common-reasons-high-cpc-without-sales/) covers that test. ### Never testing the landing page Ad accounts get audited constantly. Landing pages almost never do. A campaign with a three-percent click-through rate and a 0.6 percent conversion rate isn't a campaign problem. It's a landing page problem. Every additional dollar spent against that page is waste. The fix is a quarterly test rotation on the top three highest-spend landing pages. Hero copy, primary call to action, social proof placement, and form length are the four variables that move conversion rate most for service businesses. Product page templates, above-the-fold imagery, and shipping policy visibility are the equivalents for Shopify. Without a testing cadence, the conversion rate is whatever the page happened to do on launch day. The page patterns repeat across the [home and furniture vertical](/for-home-brands/furniture/), and for legal practices the lead-form variables that move the close rate live on the [law-firm side of the practice](/for-service-brands/law-firms/). ### No negative audiences layered onto prospecting A prospecting campaign with no exclusions spends part of its budget showing ads to existing customers, recent purchasers, and current site visitors. That spend lifts reported ROAS because those audiences convert from other sources anyway. It doesn't buy incremental revenue. The fix is a negative-audience layer on every prospecting campaign. Exclude purchasers from the last sixty to one hundred eighty days depending on repurchase cycle. Exclude email subscribers. Exclude high-intent site visitors who already saw retargeting. The prospecting number will look worse on paper for two weeks. The incremental revenue from the budget will be higher within four. ### Skipping offline conversion imports on lead-gen accounts Service businesses and high-consideration brands have a lead-form click, then a sales process that decides which leads become revenue. If only the form submission gets reported back to Google, the algorithm optimizes for cheap leads, not closed deals. The campaigns that produce the highest-quality leads often look the most expensive in the interface. The fix is offline conversion imports. Push closed-won values back into Google Ads at the keyword and campaign level on a weekly cadence. The bidding algorithm then optimizes against revenue, not lead volume. For a lead-gen account spending over ten thousand a month with no offline imports running, this is the next change I make once tracking is clean. ### The Shopify version of these mistakes On a Shopify store, wasted spend tends to show up in four places. **Two purchase tags for one order.** The Google & YouTube app and a Tag Manager purchase tag both fire at checkout, so every order counts twice. The bid model then chases a conversion rate the store never had. Keep one primary purchase action per platform and demote the other to secondary. **Performance Max buying the brand.** Without brand exclusions, Performance Max serves on searches for the store's own name and on shoppers who were already coming back. ROAS looks better than the campaign is. Add the brand name and its misspellings as account-level phrase negatives, which Performance Max respects, let the organic listing take those searches, and judge Performance Max on new-customer revenue. **Ads for products that can't pay for themselves.** The feed advertises every product published to the Google channel, including low-margin items and stock that's about to run out. Split the feed by margin tier and leave out anything that can't reach the target ROAS. **Meta optimizing for browsing.** A campaign set to add-to-cart or landing-page views finds people who browse. Once the pixel and the conversion API send purchases reliably, optimize for Purchase. ### The account-layer settings that keep causing all seven Campaign settings control what one campaign does. Platform configuration controls what every campaign in the account believes is true. A wrong location toggle wastes one campaign's budget. A wrong conversion definition teaches every campaign to optimize toward the wrong outcome for as long as the configuration stands. Seven settings sit above the campaign layer and cause the mistakes above to keep returning after they're fixed. **The conversion library.** Open Tools, then Conversions, in Google Ads, or Events Manager in Meta, and list every active action with its primary or secondary status. Two leaks live here. Duplicate conversions firing for the same event, usually because the tag was placed once through Google Tag Manager and again through a Shopify-native pixel. And primary conversions that include email signups, scroll depth, or video views next to purchases, which teaches the model that a signup and an order are worth the same. Deduplicate first, then demote every non-revenue event to secondary. **The attribution model.** Google defaults new conversion actions to data-driven attribution. Older accounts may still run last-click. Meta defaults to seven-day click and one-day view. The leak appears when the model disagrees with how the buyer buys. A considered-purchase brand with a thirty-day research window undercounts paid social on last-click. A direct-response brand converting in-session overcounts Display on data-driven. Data-driven no longer needs a minimum conversion volume to switch on, but a low-volume account gives it little to learn from, so compare its credit against last-click before you trust it. **The conversion window.** Click-through and view-through windows are editable inside each conversion action. The longer the window, the more credit the platform claims for conversions that closed for other reasons. Pull the path-length report in GA4. If ninety percent of conversions close within seven days of the click, a thirty-day window is inflating reported conversions and training the bid model to chase credit. Shorten the window to the real close pattern, then watch reported conversions fall and cost per real conversion correct itself. **Account-level negative lists.** The shared library in Google Ads holds negative keyword lists that apply to every linked campaign. An account with campaign-level negatives and nothing at the account level launches every new campaign with zero protection. Build one master list per account: the business's own brand name and misspellings as phrase negatives, competitor brands you have decided not to bid on, employment terms like jobs and salary, free-intent terms on a paid offering, and geography outside the service area. Apply it to every campaign by default. Google Ads also has an account-level negative keyword setting, which covers Performance Max as well as Search. Set the Meta equivalent through placement exclusions and brand-safety filters. **MCC structure.** Shared resources, conversion imports, and analytics links flow through the manager account. A flat structure with no MCC means every account configures conversions, audiences, and linked products independently, which produces drift between accounts inside the same business. One MCC holding the Google Ads account, the Merchant Center account, and the GA4 property is the floor for a single brand. A parent MCC with sub-accounts per brand keeps one brand's conversion definitions out of another's. **The GA4 link.** Two errors recur. A property connected with auto-tagging disabled, which strips the gclid and breaks the session join inside GA4. And GA4 conversions imported into Google Ads alongside native conversions for the same event, which double-counts. Confirm auto-tagging is on in GA4 property settings, and confirm only one source per event is marked primary in the conversion library. **Customer-match refresh cadence.** Customer lists decay every week they sit static. New customers join the CRM, old customers churn, email addresses change. A list uploaded twelve months ago still appears in the targeting panel while its match rate and conversion rate have both halved. Sort the audience manager by upload date. Anything past ninety days needs a refresh. Anything past one hundred eighty days needs replacement, with the new list replacing the stale one rather than stacking beside it. None of these produce a red flag in the campaigns view. That's what makes them expensive. Run them on the first of every quarter, and run the customer-match refresh on the first of every month. ### The fix order | Step | Task | Time | |---|---|---| | One | Tracking integrity | Four to eight hours | | Two | Search-term negatives at the account level | Three to five hours | | Three | Consolidate campaign sprawl | Six to ten hours | | Four | Layer audience signals | Two to four hours | | Five | Rebuild the highest-spend landing page | Eight to twenty hours | | Six | Realign bid strategy | One to two hours | Every fix below works in isolation for about a week. Then it breaks. Negatives get re-added by broad match. Audience signals get overwritten by the next platform recommendation. A landing-page rebuild fails to move CPA because the conversion API was double-counting the whole time. The sequence below is the order I work an account on intake. Each step protects the work of the next one. **Step one, tracking integrity. Four to eight hours.** Walk the de-duplication contract in the [Tracking Stack reference](/frameworks/tracking-stack/) end to end. Verify the browser pixel and the conversion API fire the same event ID on every purchase, and that enhanced conversions receive the customer email (the Google tag hashes it before it's sent). Check the diagnostics tab in Google Ads and Events Manager in Meta for event-match quality below seven. Look for three failures in particular: the conversion API double-counting, the pixel firing twice on the thank-you page, or mismatched attribution windows between platforms. Verify by comparing Google Ads conversions against the Shopify orders GA4 attributes to paid search over thirty days. The gap should sit inside ten percent, the band set in [the Google Ads versus Shopify reconciliation](/conversion-tracking/google-ads-conversions-dont-match-shopify-orders/). If it doesn't, don't move to step two. **Step two, search-term negatives at the account level. Three to five hours.** Read the top two hundred queries by spend over ninety days and push everything irrelevant into three account-level lists: occupational queries, informational queries that won't convert, and competitor terms you have decided against. For Performance Max, add them yourself: account-level negative keywords apply to it, and it now takes campaign-level negative keywords in the interface too. Re-pull the report two weeks later and check the irrelevant share has dropped. **Step three, consolidate campaign sprawl. Six to ten hours.** Every campaign the account doesn't need fragments conversion data and starves the algorithm of signal. Keep the splits that separate intent (non-brand and competitor) or service areas with different close rates, as set out in [the account-structure guide](/wasted-ad-spend/ad-account-structure-best-practices-reduce-waste/). Fold the rest: duplicates, copies, and campaigns split by product with the same intent. One Performance Max campaign per margin tier is usually enough. Pause every campaign that hasn't produced ten conversions in thirty days, and every ad group under a Quality Score of five that nobody has touched in ninety days. Brand terms go in as phrase negatives. A brand campaign exists only under the competitor exception, with a hard daily cap. Display and Demand Gen get paused unless there's a written reason they exist. **Step four, layer audience signals. Two to four hours.** With clean tracking, clean negatives, and a clean structure, the algorithm finally has something to learn from. Build Customer Match lists from the email and SMS file, segmented by purchase recency at zero to ninety days, ninety-one to three-sixty-five, and lapsed. Add cart abandoners from the last sixty days and high-AOV buyers where the data supports it. Upload email plus phone. Both platforms hash a plain-text file in their uploaders, so don't hash it by hand unless you're sending through the API. A low match rate usually means stale or sparse contact data. **Step five, rebuild the highest-spend landing page. Eight to twenty hours.** Match the headline to the ad's strongest copy variant, cut form fields to the minimum the offer needs, compress images, and defer any third-party script that isn't load-bearing. Confirm the mobile PageSpeed score clears seventy, then watch conversion rate across fourteen days of stable traffic. Judge the rebuild against the page's own trailing rate, not an industry average. **Step six, realign bid strategy. One to two hours.** Set target ROAS to the contribution-margin number rather than the platform-suggested number. Set target CPA on lead-gen to a third of customer lifetime value. Fourteen days later, cost per conversion should sit inside ten percent of target. If it doesn't, something upstream is still wrong. ### Where to start Fix mistake two first. Tracking is the foundation everything else reads from. Mistake one is where the visible money goes: in Q3 2026, a median 30.7% of the search-term spend Google showed went to terms with 2 or more clicks and zero conversions (N = 25). Mistake four reveals how much of the reported result was people who already knew the business. Before step one, run the [free 25-page Setup Checklist](/audit/) against the account to find which mistakes are live in the current build. If any step takes more than double its time estimate, the account has a deeper issue the standard sequence won't fix on its own. The accounts that stay clean treat the search-terms report and the tracking audit as recurring calendar items rather than one-time cleanups. Closing two of these seven changes the read on an account inside a month. When the list runs longer than the calendar allows, [send the account for my free written audit](/audit-request/), and [the wasted ad spend hub](/wasted-ad-spend/) indexes the rest of the diagnostics. ## What are common reasons for high cost-per-click without sales? URL: https://connercrowe.com/wasted-ad-spend/common-reasons-high-cpc-without-sales/ Published: 2026-05-25 | Updated: 2026-10-30 Answer: Six reasons explain high CPC without sales: auction crowding, paying for your own brand searches, weak ad and landing-page match on top-spend keywords, bid-strategy mismatch on thin data, audience drift after broad-match expansion, and landing-page friction. The brand fix is phrase negatives on your own name, so the organic listing takes those searches. ### Why CPC and sales drift apart CPC and conversion rate are governed by different systems. CPC is set by the auction. Sales are set by the offer, the landing page, and the match between the searcher's intent and what loads on the page. When CPC climbs and sales stall, one of those two systems has shifted, and the platform rarely tells you which one. The [Google Ads recommendations page](https://support.google.com/google-ads/answer/3448398) can suggest bidding, keyword, and ad changes, but those suggestions still need to be read against your actual conversion and margin data. Diagnose the gap in a fixed order. Auction first, account structure second, landing page third. Skip the order and the fixes contradict each other. Before you call your CPC high, check what you're comparing it to. [WordStream and LocaliQ's 2026 benchmarks](https://www.wordstream.com/blog/2026-google-ads-benchmarks) put the cross-industry search average at $5.42 for April 2025 to March 2026, in a figure that blends Google and Microsoft Ads. My own Q3 2026 benchmark of 33 live Google Ads accounts came in higher, at $8.53 for the typical account (N = 33), with home services typically at $13.27 (N = 11) and the other lead-gen accounts at $10.47 (N = 10). On 12 accounts I have managed every summer since 2023, CPC rose unevenly (a typical 19%, then 11%, then 76%, N = 12). The detail is in my [Google Ads CPC trend, 2023 to 2026](/benchmarks/google-ads-cpc-trend/). ### Reason 1: more advertisers in your auctions Pull the [auction insights report](https://support.google.com/google-ads/answer/2579754) in Google Ads. Filter to the last ninety days against the prior ninety. Look at impression share lost to rank and the list of competing domains. A new entrant bidding aggressively on a keyword you used to own can lift your CPC sharply without you changing a single setting. The fix isn't matching their bid. The fix is a structural read on which keywords are worth defending at the new clearing price and which ones move to a different match type or a different campaign. Some keywords need to be ceded to the new bidder while you reallocate to terms with thinner competition. The bid simulator inside the keyword view shows the volume curve at each CPC. Use it before raising any bid. Auction insights shows who you're up against, not what they pay. Google never shows a rival's CPC, and [what you can see about competitor CPC](/wasted-ad-spend/can-i-see-competitor-cpc-google-ads/) also explains the top of page bid ranges in Keyword Planner. ### Reason 2: paying for your own brand name Most accounts pay for searches of their own name without meaning to. It's spend on sales you'd mostly have had anyway. A broad or phrase keyword in a non-brand campaign matches "your brand plus service", and the click that organic search would have delivered free now costs money. Those clicks also convert well, so they make the campaign's conversion rate and cost per acquisition look better than the non-brand traffic deserves. When a competitor starts bidding on your name, the price of those brand clicks climbs, and the average CPC climbs with it. My default is to stay out of that auction. By default, my accounts carry the client's brand names as phrase negatives across the account, and the organic listing takes those searches. When I rolled that rule across my accounts in July 2026, the accounts that needed the most care were the ones whose recorded conversions leaned hardest on brand searches. On accounts I take over, brand searches often carry a large share of the recorded conversions, and some of those buyers would likely have come through the organic listing anyway. Start with the search terms report. Filter for your brand name and its misspellings, add them as phrase negatives, and watch the campaign's cost per lead over the next month. If it rises, the brand clicks were flattering it. #### When a competitor bids on your name Type your own brand name into an incognito search. If a competitor's ad sits above your organic listing, they're bidding on your terms. Google's [trademark policy](https://support.google.com/adspolicy/answer/6118?hl=en) permits bidding on a competitor's brand keyword in most regions. It doesn't permit using a registered trademark inside the ad headline or body copy. If their ad copy carries your mark, use the [trademark troubleshooter](https://support.google.com/adspolicy/troubleshooter/14327740). The complaint may require registration details from the relevant trademark office. It doesn't stop them bidding on the keyword. It stops them displaying the trademark, which usually makes their ad less relevant and less clicked. The exception to my default is narrow: a competitor's ad holds position 1 above your organic result on your own name and is taking the click. Then a brand campaign is worth testing, kept separate so it can't inflate non-brand numbers, and the brand negatives move from the account level to a shared list applied to every non-brand campaign. Exact-match coverage of the brand, a bid ceiling, and an ad whose headline names the brand and the offer. On a service-business account where this happened, branded CPC had climbed from forty cents to nearly two dollars inside a month. A dedicated brand campaign with the brand in headline 1 and a landing page that mirrored it reached a Quality Score of 10 within two weeks, and CPC fell back to the thirty-to-fifty-cent range. If the competitor sits below your organic listing, or their ad is weak, leave them to pay for clicks your listing still wins. #### The monthly check Auction insights is the only place inside Google Ads that lists the advertisers in your auctions, and most founders open it once when they suspect a problem, then forget it exists. The advertisers shift every quarter, and a new entrant arriving in week one is invisible until the CPC damage lands in the campaign report eight weeks later. Pull it on the first business day of the month, filter to the prior 30 days, and note new entrants over a 10 percent overlap rate on every campaign. A spreadsheet with the date and the domain is enough to show which competitors are seasonal, which are permanent, and which abandon the auction after a single month. ### Reason 3: low Quality Score on top-spend keywords Quality Score is the lever most under-used by solo founders. The score itself isn't an input to the auction. It's Google's 1-to-10 summary of expected CTR, ad relevance, and landing-page experience, and [Ad Rank](https://support.google.com/google-ads/answer/1722122) uses real-time versions of those same three. So a low score is a symptom: the weaknesses behind it pull Ad Rank down, and holding the same position then costs more per click. Google doesn't publish how much, and I won't invent a multiplier for it. Open the keyword view. Add the Quality Score, expected CTR, ad relevance, and landing-page experience columns. Sort by spend descending. Any keyword in the top twenty spend rows with a Quality Score under 5 is where to look first. The three component columns tell you which one is below average. The fix is mechanical. Rewrite the ad to use the keyword in the headline. Send the click to a landing page that contains the keyword in the H1 and the first paragraph. Watch the components over two weeks. Judge it on cost per conversion. Across 30 accounts I manage, low-score clicks cost more in only 3 of the 7 accounts with enough clicks in both Quality Score bands, so don't count on CPC following. ### Reason 4: bid strategy mismatch Target CPA and target ROAS need a minimum conversion volume to learn. Google's own [Smart Bidding documentation](https://support.google.com/google-ads/answer/7066642) covers the data thresholds and the learning period, and the longstanding practitioner floor is at least thirty conversions in the last thirty days per campaign before tCPA learns reliably. Most solo-founder accounts run tCPA on campaigns with eight or twelve conversions a month, and the algorithm responds by paying any CPC it has to in order to chase a target it can't reliably hit. The fix is to demote the bid strategy when the data is thin. Maximize Clicks with a CPC ceiling is the correct starting point on a new campaign or a low-volume one. Move to tCPA only after thirty conversions accumulate inside the campaign itself, not the account. Portfolio bid strategies that pool conversions across campaigns can hold the data threshold, but only if the campaigns share an audience profile. ### Reason 5: audience-offer drift after broad-match expansion Broad match in 2026 reaches well past the keyword you typed. The algorithm expands to queries that share theme with your seed keywords, and the expansion frequently lands on intent that doesn't match your offer. Search-impression-share rises, CTR rises, CPC rises, and conversion rate collapses. The campaign looks busier and earns less. Pull the search-terms report for any broad-match campaign. Filter to the last sixty days. If more than a quarter of the high-impression queries describe a product variant you don't sell or an intent you can't serve, the broad-match expansion has pulled the campaign away from the offer. The fix is a deliberate negative-keyword pass plus a downgrade of the worst seed terms to phrase match. Audience signals layered onto the campaign also reanchor the algorithm. ### Reason 6: landing-page friction killing the conversion A click that lands on a slow page or a generic homepage doesn't convert. The CPC was paid in full. The sale was lost on the storefront, not on Google. This is the failure mode that ad managers blame on the algorithm and that the algorithm blames on the offer. Test the landing page on a mobile connection throttled to 4G. Time to interactive over four seconds is a conversion killer. A product page that loads fast and names the product in the H1 will out-convert a homepage that makes the visitor go find the product. Match each ad group to a landing page that names the product or the service in the first viewport. The [Tracking Stack reference](/frameworks/tracking-stack/) covers the conversion-API contract that confirms the sale fired correctly. A broken pixel reads as a landing-page problem in the data, and the fix is in the tracking layer instead of the storefront. ### What to do once you have spotted the cause Two of these reasons together is the threshold for a full audit. Work the [free 25-page Setup Checklist](/audit/) against the account, or give me read-only access for my [free written audit](/audit-request/). If a competitor's ad now holds your brand search above your organic result, test the exception campaign above first. The [services overview](/services/) covers the structural fix at the same depth as my paid engagements. Sequence the work by how long each fix takes to land. The auction-insights pass and the brand phrase negatives are jobs an in-house operator ships inside an afternoon. The Quality Score climb and the trademark complaint take a few weeks. The reasons compound. Brand clicks slipping into non-brand campaigns hide the problem inside an averaged report. On the non-branded side the chain runs like this: a weak match between query, ad and page pushes cost per conversion up, which makes the tCPA target harder to hit, which makes the algorithm chase volume on broader match, which lands clicks on a generic page. Fix the match between query, ad and page and the chain unwinds. The [wasted-ad-spend library](/wasted-ad-spend/) walks through the rest in order. ## What do 'AI Max expanded matches' and 'landing page matches' mean in my search terms report? URL: https://connercrowe.com/wasted-ad-spend/what-are-ai-max-expanded-matches-and-landing-page-matches/ Published: 2026-10-21 | Updated: 2026-10-22 Answer: They're two AI Max total rows in the keywords report. Expanded matches is traffic from broad versions AI Max built from your keywords. Landing page matches is traffic matched from your site, with no keyword. On one service account I tested for 14 days, AI Max cut CPC 42% and raised cost per qualified lead 76%. Most people find these rows by pasting a line from an export into Google, something like `Total: AI Max expanded matches,,,,,,usd,0,0`. Google adds these rows, and they count traffic outside your keyword list. If those rows show spend, Google is buying searches you never chose, and the dashboard will call it a win. ### What the two rows count Google Ads Help defines both in [About reporting in AI Max for Search campaigns](https://support.google.com/google-ads/answer/16470459). They sit at the bottom of the keywords report as two aggregate rows for AI Max performance. - **Total: AI Max expanded matches** is traffic from broad match keywords AI Max created, based on the keywords you added. - **Total: AI Max landing page matches** is traffic from searches that matched because of your landing pages or assets, outside your keywords. So the first row is AI Max stretching your list. The second is AI Max reading your site and deciding what to buy without a keyword at all. In the Google Ads API the same split shows up as `AI_MAX_BROAD_MATCH` and `AI_MAX_KEYWORDLESS`. A row of zeros needs no action. It usually means AI Max is off, search term matching is off for the ad group, the campaign is paused, or nothing matched in the date range. In the exports I've read, the rows appear even when AI Max matched nothing. ### Why you saw them "in the search terms report" The total rows belong to the keywords report. The search terms report carries the same traffic a different way: each AI Max query gets Match type "AI Max", and you can add a Source column by hand to see what triggered it. Google adds a caveat on the same help page. Filtering the search terms report to Match type = AI Max leaves out the "Other search terms" bucket, so the footer totals can be larger than anything you can list term by term. ### What a "Total:" row is A total row adds up a group of rows. In the CSV export it leaves the campaign and keyword columns blank, which is why the label is followed by a run of commas before the clicks, cost and conversions. ### Why a nonzero row isn't free volume Adalysis found that the expanded-matches total [can include searches your exact and phrase keywords were already winning](https://adalysis.com/blog/ai-max-search-term-reporting/), so I don't read it as extra traffic. I read both rows as a list of traffic nobody picked, which makes it the first place I audit. Here's what that traffic did on one service account. I ran a 14-day split: AI Max with search term matching on in one campaign, a manual phrase and exact control in the other, same budget, creative and landing pages. | Metric | Control | AI Max | Change | |---|---|---|---| | CPC | $3.89 | $2.26 | -42% | | Form fills | 32 | 45 | +41% | | Qualified leads (CRM-marked) | 14 | 8 | -43% | | Cost per qualified lead | $178 | $313 | +76% | The dashboard looked like a win. Almost all the extra volume came from educational queries the landing page pulled in, people researching rather than buying. That's one account in one vertical over two weeks, so it isn't a verdict on AI Max everywhere. The full write-up is in [I ran AI Max for two weeks against a manual campaign](/blog/google-ads-search-term-matching-ai-max/). ### What I do with these rows **I judge the traffic by cost per qualified lead.** In that test, a qualified lead was one the sales team marked qualified after a discovery call. In the accounts I run day to day, it's a booked appointment, tracked in WhatConverts. CPC and form fills both looked better on the dashboard and pointed the wrong way. If your tracking can't separate a booked appointment from a form fill, fix that before you read anything AI Max reports. **I pull the terms first.** Add the Source column, filter Match type to AI Max, open the "Search terms and ad combinations" view (formerly "Search terms and landing pages from AI Max"), sort by cost and read the zero-conversion rows. Across the accounts I manage, a median 30.7% of visible search-term spend went to terms with 2 or more clicks and zero conversions in Q3 2026 (N = 25), and the report itself showed only a median 52% of spend (N = 11). Both rows are in my [Q3 2026 benchmarks](/benchmarks/google-ads-account-benchmarks/). **I negate on a shared list.** I wait two weeks before I negate anything, because acting on three days of AI Max terms is guessing. After that, research queries that don't book go on a shared negative list attached to every Search campaign, so the same term can't leak back in through another campaign. You can also add a term straight from the report with "Add as negative keyword", or a bad page with "Add as negative URL". **I use brand exclusions.** You add brand lists in the Brands section of the AI Max settings. [Google's AI Max setup page](https://support.google.com/google-ads/answer/15909989) makes brand exclusions a campaign-level control. My default for a client's own name is phrase negatives on a shared list, and I only bid on it when a competitor holds that search above the organic listing. A brand exclusion keeps AI Max from quietly rebuilding the brand campaign I chose not to run. **I consider turning off final URL expansion.** It's on by default, and it lets Google send clicks to whatever page on your domain it predicts will do better. You turn it off in the Asset optimization panel of the AI Max settings, and turning off text customization switches it off too. If you want to keep it, URL exclusions block specific pages at the campaign level. On a lead-gen account, a blog post or resources page is exactly the kind of page that pulls in researchers, so those are the first pages I'd exclude. If the landing page row keeps filling with the wrong intent, the bigger switch is search term matching. It has to be on at the campaign level, and you can turn it off ad group by ad group. Google says search term matching, final URL expansion and text customization need conversion-based Smart Bidding to work well, so a weak conversion signal makes all three worse. The match-type mistakes that feed the same leak are on [keyword matching mistakes that waste spend](/wasted-ad-spend/typical-mistakes-keyword-matching-wasted-spend/), and the CPC trap is on [lowering CPC without losing conversions](/wasted-ad-spend/strategies-lower-cpc-without-sacrificing-conversions/). ## What is a good return on ad spend for an ecommerce business? URL: https://connercrowe.com/wasted-ad-spend/good-roas-benchmark-ecommerce-businesses/ Published: 2026-05-25 | Updated: 2026-10-30 Answer: A good ROAS is one that clears breakeven, and breakeven is 1 divided by contribution margin, so a 30 percent margin needs 3.3x. Then look under the blend: two home brands I run report 21x, but Shopping earns 61x on 18% of spend and Performance Max 12x on 82%. One 21-day geo holdout measured 38% real lift against 92% reported. ### "Good ROAS" is the wrong question A 5x ROAS on a 20 percent contribution margin product is a money-losing campaign. A 2x ROAS on a 60 percent margin product is profitable. Asking what a good ROAS looks like without naming a margin is asking a question the number can't answer. ROAS is revenue divided by ad spend. Profit is what survives cost of goods, shipping, payment processing, ad spend, and everything else that scales with an order. The dashboard knows none of that. Start with contribution margin per order, then back into the ROAS the business needs to clear. Skipping that step is why an account looks healthy in Ads Manager and dies on the P&L. ### The breakeven math **Breakeven ROAS = 1 / contribution margin** A 30 percent margin breaks even at 3.3x. A 50 percent margin at 2x. A 20 percent margin needs 5x to stop losing money on incremental orders. Read against gross margin instead, 40 percent breaks even at 2.5x and 55 percent at 1.82x. Then layer in profit. A healthy target sits well above breakeven on cold traffic, and the formula tells you why: a brand at 70 percent margin breaks even at 1.43x, so 3x is comfortable, while a brand at 35 percent margin breaks even at 2.86x and needs far more than 3x before cold traffic pays. Founders who do this math for the first time often find their internal "good ROAS" target was set below breakeven. Breakeven also sets the floor you enforce in the platform. Google Ads target ROAS bidding should sit at least 20 percent above breakeven to absorb tracking variance and returns. On Meta, the equivalent is a minimum-ROAS rule on the campaign. Watch for campaign-level ROAS sitting near breakeven across a 30-day window, or below it for two weeks running. The same math gives a target CPA: gross profit per order minus a planned contribution to fixed costs. On a Shopify home brand at 55 percent gross margin and a $160 average order value, that lands between $50 and $70. Reported CPA above target on a rolling 30-day window is the concern line. Two days is noise. Two weeks is structural, and the fix is redistribution toward the campaigns already converting under target, not a bid adjustment. ### Why category averages don't help much I don't publish a ROAS table by category, because the margin inside a category varies more than the average between categories. Home and furniture: big baskets hide thin per-order economics, and freight eats margin the dashboard never sees, as the [furniture playbook](/for-home-brands/furniture/) covers. Apparel: returns compress effective margin, so platform ROAS has to clear higher than the math first implies. Consumer electronics: thin margins mean a ROAS that looks strong is often a losing account dressed up by branded-search overlap. A furniture brand at 55 percent margin on an $1,800 AOV also operates differently than one at 30 percent margin on a $400 AOV. Run the brand's own breakeven first. On one ecommerce store I manage, I set the Performance Max target with headroom above the brand's breakeven, because Google's target ROAS undershoots more than it overshoots on its own reporting. The weekly read swings hard: 3x, then 4x, then 2x, then 6x in consecutive weeks. Over a quarter the blended number lands near the target. The ninety-day blended is the read that matches the P&L. ### Blended ROAS vs platform-reported ROAS Every platform claims credit for the same conversion when journeys overlap, so the sum of platform numbers usually exceeds actual revenue. Blended ROAS is total revenue divided by total ad spend across every platform for the period, and it's the only number that ties to the bank account. A 5x Meta and a 4x Google on top of a 2.5x blended is a flashing light. Those platforms are each claiming revenue the other one also claimed, and the bank balance exposes it a month or two later. Set blended as the north star and run a monthly blended-vs-platform reconciliation. The [tracking stack reference](/frameworks/tracking-stack/) covers the de-duplication contract that makes it trustworthy. The gap can be wide. On the one store in my [ecommerce tracking accuracy benchmark](/benchmarks/ecommerce-tracking-accuracy/) where both sides were readable, the ad platform's counted purchases equaled 58.5% of the store's entire 90-day order book, while the store attributed none of those orders to paid (N = 1). The blend is also an average that hides its parts. Two home brands I run report a combined 21x on Google Ads, but [split by campaign type](/blog/what-a-home-brands-21x-roas-actually-hides/), Shopping earned 61x on 18% of the spend and Performance Max 12x on the other 82%. A blended number can be honest about the bank account and still say nothing about which campaign is buying new customers. Even blended revenue overstates what the ads caused. On one [21-day geo holdout](/blog/21-day-geo-holdout-incrementality-test/) I ran for a home-furnishings brand, pausing paid media in one state cut revenue there by 38%, while attribution had called 92% of revenue paid-influenced (N = 1). Run a holdout before treating any ROAS, platform or blended, as incremental. ### Branded-search overlap inflation Clicks on searches for the store's own name usually carry the highest ROAS in any account that buys them, and much of it isn't incremental, because the organic listing would have delivered those buyers free. My default is to stay out of that auction. I add the store's brand name and its misspellings as phrase negatives across the account, Performance Max included, and let the organic result take those searches. The one exception is a competitor's ad holding the brand search above the organic result. Then a separate brand campaign is worth testing, kept out of every non-brand number, and the brand negatives move from the account level to a shared list applied to every non-brand campaign. Strip brand clicks out of a healthy-looking Google Ads account and the ROAS left over is often far lower, and that's the number that matters for growth. Brand traffic flatters lead-gen accounts too. Meta retargeting works the same way. A 10x ROAS against existing site visitors recaptures demand the brand already paid to create. ### New-customer ROAS is the metric that scales Total ROAS includes returning buyers who would have purchased through email, organic, or direct. New-customer ROAS strips them out and predicts whether the account can grow. Whether a given new-customer ROAS is healthy depends on how much a first buyer is worth over the following year, so set the floor from your own LTV. A platform reporting 4x total ROAS against 0.8x new-customer ROAS is recapturing existing demand, and the growth ceiling is in sight. Chasing 4x new-customer ROAS on cold traffic shrinks that ceiling from the other direction, because targeting tightens until only existing-intent buyers convert. Loosen the target, accept first-purchase contribution near breakeven, and let LTV pay back acquisition cost across month two through month twelve. The report is GA4 with a `new_customer` parameter on the purchase event, segmented by paid channel. Setup takes an afternoon. On one ecommerce store I manage, platform-reported ROAS looked healthy at the 4x target. Once the GA4 split was running, the new-customer cut sat closer to 2x. Total ROAS was the dashboard read. New-customer ROAS was the growth read. ### Contribution margin after ad spend The most honest number in the account. Revenue, minus cost of goods, minus ad spend, divided by revenue. Negative means the campaign is paying customers to take inventory. Below zero is a stop-loss, and below zero on cold-traffic orders for more than 30 days with no LTV plan to recover the gap is the definition of bad ROI regardless of what Ads Manager says. Anything negative on first purchase needs a clear second-order or subscription pickup to be defensible. Pull the campaigns printing negative numbers, redistribute into the campaigns with the healthiest margin after ad spend, then think about scaling. It's common to find a few campaigns silently losing money inside an account reporting a healthy blended ROAS. ### Reading social spend against the same targets Meta shows one number, Shopify another, TikTok a third that disagrees with both. Each measures a different slice of the same purchase, with different windows and different signal-loss adjustments stitched on after [iOS 14.5](https://www.facebook.com/business/help/331612538028890). Six reads make social spend legible against the math above. **CPM.** Pull CPM by ad set on a rolling 14-day window and read it against its own history, not a published average, because CPM depends heavily on the optimization event. In my [Meta CPM benchmark](/benchmarks/meta-ads-cpm-by-placement/), conversion-optimized ad sets paid a $20.35 pooled CPM against $2.88 for reach and impressions ad sets (N = 9 and N = 6). A sharp CPM rise on cold prospecting without a matching lift in ROAS calls for a creative refresh. More budget won't fix it. **CPC and click-to-purchase rate.** TikTok tends to run cheaper on impressions and expensive on commercially qualified clicks. Compare the paid-social click-to-purchase rate against your store's own conversion rate for those pages. If paid clicks convert at a fraction of the store average, the offer-to-page match is broken, the page is slow, or the audience is wrong. **CPA against contribution margin.** A $48 CPA on a $90 contribution margin is healthy. The same CPA on a $35 contribution margin is bleeding. **Attribution window.** Meta defaults to [7-day-click plus 1-day-view](https://www.facebook.com/business/help/458681590974355), written 7DC/1DV. Read 7DC/1DV for directional health, 7-day-click when comparing against Shopify, 1-day-click when stress-testing. If 7DC/1DV looks strong but 1-day-click ROAS collapses below 1, the campaign is harvesting credit for purchases it didn't cause. **Frequency.** On cold prospecting, watch for frequency climbing while CTR falls and CPA inflates in step. That's fatigue. Warm and retargeting audiences tolerate a higher frequency before the same thing happens. **CAPI deduplication.** The one most founders skip, and it decides whether everything above is signal or noise. The browser pixel loses a meaningful share of conversion events post-iOS 14.5. The [Conversions API](https://developers.facebook.com/docs/marketing-api/conversions-api) fills the gap server-side, but every browser event must match a server event by `event_id` or Meta double-counts the purchase. Check deduplication in Events Manager. If browser and server events aren't being deduplicated, reported ROAS is wrong. The [tracking stack reference](/frameworks/tracking-stack/) walks the `event_id` handshake, the `fbp`/`fbc` passthrough, and the match keys that lift Event Match Quality above 7. A Meta-reported 4.2x against a Shopify-reported 1.9x is normal in a poorly tracked account. Meta claims purchases inside a 1-day-view window and Shopify doesn't, so strip the view-through column to compare like-for-like. Shopify credits the last touch, often a session Meta legitimately introduced. And a dedup failure counts the same purchase twice. The fix lives in the tracking stack, not the bid strategy. ### The indicators that move before ROAS does **Conversion rate against a vertical benchmark.** Read against a benchmark, conversion rate tells you whether the leak sits upstream or downstream of the click. Use a benchmark with a stated sample, or your own trailing rate. For home-services Google Ads, the typical click-to-lead rate across the accounts with clean tracking in my [Q3 2026 home-services cost-per-lead benchmark](/home-services-paid-search-benchmarks/) was 10.3% (N = 9). Ecommerce purchase rates run far lower than lead rates, and furniture lower still because of the consideration cycle. Worry when relevant traffic, judged by the search-term report, converts at a fraction of that reference. Relevant traffic that won't convert is a page problem. Irrelevant traffic is a match-type problem. **Impression share lost to budget.** The share of auctions a campaign was eligible for but didn't enter because the daily budget was spent. A large share lost to budget on a campaign already hitting target CPA or target ROAS is a campaign asking for more money. A small share on a campaign missing its targets means budget isn't the constraint. The structure is. **CTR-to-CR ratio.** A high CTR with a low CR means the ad promised something the page didn't deliver. Watch for a strong CTR paired with a conversion rate well under your own trailing rate. Match the ad headline to the H1 of the page, then match the H1 to the query that triggered the ad. Check what the conversion column counts before reading any rate off it. In my [ecommerce tracking accuracy benchmark](/benchmarks/ecommerce-tracking-accuracy/), two of three ecommerce accounts counted add-to-cart or begin-checkout events inside the metric their bidding optimized toward, and 19.8% to 25.8% of their reported conversion value was cart value rather than revenue. The column still read "Conv. value". A conversion rate or ROAS built on that column runs high, and nothing in the interface says so. ### Trend and cohort: the reads that need time Single-month ROAS bounces. The signal is the rolling 90-day trend. Worry when blended ROAS slides across a rolling 90-day window while spend holds roughly constant. That gradient means acquisition cost is rising faster than the brand can absorb. Plot a 13-week rolling chart of blended ROAS next to spend and pull it monthly. Read the fourth quarter against the full year, not against the volume it carries. On one ecommerce account I manage, purchase-only ROAS in [Q4 came in under the full-year average](/blog/q4-roas-trough-not-peak/) both years I could measure: 14.1% under in 2024 and 10.5% under in 2025 (N = 1). The busiest quarter was the weakest return both times, so a Q4 target copied from the annual number sets the account up to overspend. The second read is LTV-adjusted ROI. A 2x first-purchase ROAS that turns into 4x by month six is a great account. A 3x that stays at 3x for 12 months is an account with no second order, which means no business. The warning sign is 90-day or 180-day LTV-adjusted ROI flat or below first-purchase ROI. The report is a Klaviyo or Shopify cohort view of revenue per acquired customer at day 30, 60, 90, and 180, plotted against the channel they came in on. Paid channels producing flat cohorts get cut first, even when first-purchase ROAS beats channels with rising cohorts. A payback period longer than 90 days on a first-time buyer is the same failure read from the cash side. Any single read on this page in the red is a yellow flag. Two or more at once is an account funding itself on working capital. Founders who hit four should [request the free ads audit](/audit-request/) instead of spending another week tuning alone. I write it myself and send it back within 24 hours of read-only access. ### What to target by growth stage Smaller brands usually have to accept thinner headroom over breakeven to buy growth, and lower platform ROAS on cold traffic. As revenue grows, the target should rise above breakeven, with new-customer ROAS set from LTV and total ROAS allowed to climb as repeat purchase compounds. At the top end, guard hard against branded-search bundling and attribution inflation, because every point of inflated ROAS turns into real overspend at that size. A good ROAS is the one that funds the business after every variable cost. Benchmark ranges set the floor for a category. Margin and LTV set it for a specific business. If what the platforms report sits well above your blended number, the [free 25-page Setup Checklist](/audit/) covers the conversion settings that cause most of that inflation, and the [wasted ad spend library](/wasted-ad-spend/) covers the rest of the metrics that get misread in account reviews. ## What warning signs show my ads are targeting the wrong audience? URL: https://connercrowe.com/wasted-ad-spend/warning-signs-ads-targeting-wrong-audience/ Published: 2026-05-25 | Updated: 2026-10-30 Answer: The clearest sign is spend reaching people your settings never meant to target: on one Display campaign labeled retargeting, only 14 of 1,247 clicks came from past visitors, because Optimized Targeting was on. Also check demographic and age splits against buyers, spend outside the service area, Advantage+ or Performance Max drift, all-language targeting and informational queries. ### Why audience misfit looks like a performance problem A campaign with wrong-audience targeting doesn't announce itself. The cost-per-click can sit inside the benchmark range. The click-through rate can look healthy. Impressions can grow month over month. What gives it away is the demographic shape of who is clicking, and that data lives in reports most founders never open. Audience misfit is different from generic irrelevant traffic. Irrelevant traffic is off-topic queries and broken match types. Audience misfit is the right query coming from the wrong human. Same search, wrong income band. Same interest signal, wrong life stage. The platform can't tell the difference until you teach it. Both do two kinds of damage. The obvious damage is the spend on clicks that will never convert. The compounding damage is what the algorithm learns from those clicks. Every conversion-optimized bidding strategy treats every click as a signal. Feed it ten thousand clicks from the wrong audience and it builds a model of your customer anchored to people who will never buy. That's why a fix in month one often shows up in ROAS months later. The platform takes time to forget what you taught it. ### Seven signs the audience is wrong **1. A demographic split that doesn't match the buyer profile.** Open the demographics report in Google Ads or the audience breakdown in Meta Ads Manager. Pull gender, age, and household-income segments for ninety days and compare the spend distribution against the buyer profile from Shopify or the CRM. Look for a wide gap. If the buyer base is seventy percent women and half the spend lands on men, the campaign is teaching the algorithm that the wrong audience converts. The fix is exclusion at the campaign level, not a bid adjustment. Bid adjustments slow the leak. Exclusions stop it. **2. Income-proxy data skewed below the median buyer.** Google reports household income in deciles where the data exists. Meta uses zip-code and behavior proxies. For a brand selling considered purchases above four hundred dollars, watch how much spend lands in the bottom income brackets. A heavy share there means the offer is reaching people for whom the price is a non-starter. This matters most for furniture, jewelry, luxury skincare, and B2B services on annual contracts. One limit: in the US, housing, employment and credit ads can't target by income, age or gender, and health advertisers face Google's and Meta's personalized-ads restrictions, so check the policy for your category before you lean on these levers. For a furniture brand selling four-thousand-dollar dining tables, the income-proxy cutoff matters more than the interest segment does. **3. Interest-segment drift after Performance Max or Advantage+ expansion.** Both start with a seed and expand outward. After ninety days, the audience the algorithm is buying impressions for can look nothing like the seed. Pull the audience-insights report inside Performance Max and read the top twenty interest segments by spend share. Unrelated segments near the top are the warning sign. The fix is tighter audience signals at the asset-group level, and sometimes splitting back to standard Shopping or standard Search where audience control is explicit. **4. Geographic spend outside the defined service area.** Pull the location report in Google Ads or the regions breakdown in Meta and sort by spend descending. For a service business with a defined radius, spend outside that radius is wrong-audience spend by definition. Any region taking a visible share of spend without being a market the business can serve is a finding. [Law firms hit this differently](/for-service-brands/law-firms/): jurisdiction lines decide the geographic build more than radius settings do. **5. Age skew misaligned with the offer.** If the offer is a wedding-registry brand and the top spend bracket is fifty-five to sixty-four, the ad is reaching the buyer's parents. If the offer is retirement planning and the top bracket is eighteen to twenty-four, the algorithm is buying impressions from an audience that won't convert for decades. The tell is the top age brackets taking a bigger share of spend than of conversions. **6. Language targeting set to all languages.** This hides in campaign settings and defaults to all languages in many account templates. Impressions get served to users whose browser language doesn't match the ad copy. Click-through collapses while spend keeps flowing, because impressions still serve. For most US brands serving a US audience the answer is English only. In multilingual markets the rule is one ad set per language, with copy translated by a human rather than by the platform. **7. Informational-intent queries on commercial campaigns.** A commercial campaign should serve on queries with buying intent: "buy walnut dining table", "best mid-century desk", "Shopify SEO agency pricing". Queries like "what is mid-century modern" belong on content pages. Tag the top fifty queries as commercial, informational, or navigational. If informational queries take a large share of paid impressions on a campaign built for purchases, the match types are too loose for the intent. [Furniture brands](/for-home-brands/furniture/) carry a sharper split between research and buy queries than most verticals. ### How wrong-audience traffic behaves after the click The signs above describe who is clicking. These patterns describe what they do next, and they're the fastest confirmation available. **Page depth of one.** In GA4, filtered to Google Ads source, read pages per session on the engagement report. A median of 1.0 or 1.1 across a campaign that has run thirty days means the traffic arrives and leaves without clicking through to a second page. Read it against organic traffic to those pages: when paid depth sits well below organic, the paid traffic is behaving like accidental clicks rather than buyers. The fix is usually the match type, not the page. Phrase and broad on a tight budget pull queries the bidder thinks are close enough, and close enough is a long way from ready to buy. **High bounce and very short dwell time.** A paid session of a few seconds means the visitor didn't engage at all. Before changing targeting, cross-reference Core Web Vitals. If Largest Contentful Paint is slow on mobile, a share of what looks like irrelevant traffic is relevant traffic that gave up before the page rendered. That's a performance problem. Fix the page before touching the targeting. **Commercial-intent queries converting under 0.3 percent.** The search-term report is the most honest document in a Google Ads account. If a query like "[product name] price" or "[service] near me" pulls twenty or more clicks and barely converts, the query isn't the problem. The landing experience is breaking the promise the ad made. Read the ad headline and the page H1 side by side. **Mobile-heavy clicks on a desktop-converting offer.** If the offer needs a long form, a configurator, or a high-consideration decision, and mobile takes most of the clicks while desktop takes most of the conversions, the campaign is serving into the wrong context. Mobile clicks are cheaper, the bidder optimizes for click volume by default, and the platform never asks whether the offer can close on a phone. On manual bidding, set a negative device modifier on mobile. Smart Bidding ignores device modifiers other than -100%, so there the fix is splitting desktop and mobile into separate campaigns with separate pages. Check the split before you assume it, though. In my Q3 2026 benchmark, mobile converted at a higher rate than desktop in 21 of 23 lead-gen accounts with clean tracking (N = 23), and all 4 purchase-tracked stores ran the other way (N = 4). The rows are in my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/). **Search-term-to-landing-page mismatch.** Sample fifty top-spending search terms, open the page each one landed on, and read the H1. If several of the pairings have a topical gap, the campaign is buying clicks the page can't close. Final URL expansion on Performance Max and dynamic search ads make this worse by default, because the platform picks the page. Turn off final URL expansion and map every ad group to one page with one promise. If a query needs a different page, it needs a different ad group. **Bot and fraud signatures.** Three patterns flag non-human traffic. Placements with improbably high click-through rates and zero conversions. Thousands of impressions from a single domain that doesn't resolve when you visit it. App placements whose sellers aren't authorized in the app's own app-ads.txt file, which the IAB Tech Lab validators will show. On the search side, an unusually high invalid-click credit in the billing report points the same direction. Google refunds the invalid clicks it catches. The ones it misses stay in the conversion model and skew bidding. ### The six default settings that cause most of it A targeting setting is different from a targeting choice. A choice is who the campaign was designed to reach. A setting is the toggle the platform applied at creation, before anyone read the help article explaining what it does. Defaults exist to grow platform revenue. These six are the ones I check on every account, because founders rarely know they're on. **Location set to presence or interest.** The Google default serves impressions to anyone who searched something about the target area from somewhere else. A user in Florida searching "best dentist Spokane" matches a Spokane dental campaign. Switch to Presence for every local service campaign and every ecommerce campaign that ships to specific countries. It lives under Settings, then Location options. It's what I run: across 30 Google Ads accounts I manage, 30 days to October 21, 2026, 50 of 51 Search campaigns used Presence. **Optimized targeting on Display and Demand Gen.** On by default. The label sounds like an improvement. The behavior is audience expansion off the seed. On a clean account with strong conversion signal it can help. On an account with weak signal or a young conversion model, it spends budget learning from bad data. On one Display campaign labeled retargeting that I audited in May 2026, it left only 14 of 1,247 clicks going to past site visitors. The other 1,233 went to cold traffic, and [the retargeting warning signs](/wasted-ad-spend/retargeting-ads-not-converting-warning-signs/) shows where to find the split. Turn it off until the conversion model has stabilized on real volume. **Advantage+ audience on Meta.** The default on many new ad sets. It treats most audience signals as suggestions and expands when the model predicts better performance. The seed matters less every week the campaign runs, and spend drifts toward whichever audience is cheapest. Switch to original audience options for any campaign where the buyer profile is narrow, the AOV is high, or conversion volume per ad set is thin. Check the optimization goal at the same time. In my Q3 2026 Meta benchmark, 9 of 18 accounts, carrying 42% of spend, optimized for no conversion event at all (N = 18), and Reach ad sets returned a 0.06% link CTR against 2.84% for conversion ad sets. An audience with no conversion goal behind it drifts toward whoever is cheapest to reach. The placement rows are in my [Meta CPM by placement benchmark](/benchmarks/meta-ads-cpm-by-placement/). **Device bid modifiers left at zero.** The default treats every device as equal value, which is rarely how the data reads, and the direction depends on the business. In my Q3 2026 benchmark, mobile converted better than desktop in 21 of 23 lead-gen accounts with clean tracking (N = 23), and mobile clicks cost more in 16 of 21 (N = 21). On all 4 purchase-tracked stores, desktop clicks cost more and desktop converted better (N = 4). Without modifiers, spend distributes by impression availability rather than by conversion economics. On manual bidding, set modifiers from the last ninety days of device data and adjust monthly. On Smart Bidding, device modifiers are ignored except -100%, so segment campaigns instead. **Content exclusions at default coverage.** The default inventory settings leave a lot of sensitive and low-quality content eligible. The result is placements no founder would approve if shown a screenshot: parked domains, UGC pre-roll with no editorial review, Performance Max inventory inside mobile game ad units. Review the content suitability settings at the account level and exclude every category that doesn't match the brand, so the exclusions cover Display, Video, Demand Gen, and Performance Max together. **Ad schedule left at 24/7.** Every new campaign runs all day, every day, ignoring the conversion-by-hour pattern the account already shows. B2B service campaigns usually convert nine to five on weekdays. [Furniture and home decor](/for-home-brands/furniture/) carries volume on evenings and weekends. A local service business with a phone-call goal collects voicemails after hours that never get returned. Pull the day and hour report, and any block converting far below the campaign's best block is a candidate for a pause or a negative modifier. ### Five reports, one sitting Run these five together and the leak stops feeling diffuse. 1. **Search terms filtered to zero conversions, sorted by cost.** Ninety days. Any single query that has spent more than three times target CPA with zero conversions is flagged for review. Many queries at that level is structural. Read this one carefully before adding negatives: zero conversions is a spend fact, not a relevance verdict. A query returns nothing either because it should never have matched, or because it matched correctly and then failed on a product page, a price, a stock level, or a tag that stopped firing. Check the proper nouns against the live catalog or service list rather than from memory. A brand the business stocks or a practitioner it employs can sit in that list, and blocking it on spend alone removes the buyer instead of the waste. 2. **GA4 paid engagement rate.** Reports, Acquisition, Traffic acquisition, filtered to paid. Read it against organic engagement on those pages. Segment by campaign and landing page, because the leak concentrates in one or two combinations rather than spreading evenly. 3. **The Performance Max placement report.** Insights, then Where your ads showed, then asset-group placements. Any placement taking a visible share of budget without a conversion is excludable. The report is buried on purpose. Find it anyway. 4. **The geographic report against the service area.** Covered above. The fix is two settings: presence-only location, and campaigns segmented by country or service region rather than negative location targeting, which the algorithm routinely overrides in favor of conversion volume. 5. **The placement report read a second time for bot patterns.** Same report, different question, using the three signatures above. The reason to run all five together is to find the short list of line items carrying the waste. A short list is a fix you can finish. How much of this you can even see has shrunk. In Q3 2026 the search terms report showed 52% of account spend for the typical account, across the 11 accounts I could pull completely (N = 11). Across the 25 clean-tracking accounts with search-term data, the typical one put 30.7% of the spend the report did show into terms with 2 or more clicks and zero conversions (N = 25). The method and every N are in my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/). ### Two accounts where the settings were the story On one account I diagnosed, ninety days of Meta spend had bought hundreds of thousands of impressions at a click-through rate close to zero, because the platform was distributing budget across audiences with no purchase intent for the product. Meta has no keyword layer to negative against, so the fix was structural: an audience signal rebuilt from the brand's CRM segments, the objective moved from awareness to conversions, and a frequency cap to stop a broad audience absorbing impressions at zero engagement. On a regional service account I work with, ninety days of Google Ads spend looked healthy on the dashboard. The settings audit caught what the dashboard wouldn't. Location targeting sat on presence or interest, nobody had read the device split against where the conversions came from, and content exclusions were at default coverage on a brand that should never have run on parked domains or embedded UGC video. ### Fixing the misfit structurally Audience misfit is rarely solved with one toggle. The structural fix is three decisions in sequence. Layered audience signals at the campaign or asset-group level, so the algorithm starts from a precise seed. Exclusions at the demographic, geographic, and placement level, so the seed doesn't drift. And a [tracking stack](/frameworks/tracking-stack/) clean enough to tell wrong-audience signal from mismeasured-conversion noise. Audit the six settings on the first of every month. Confirm location is presence only, that expansion features match the documented audience strategy, that device modifiers reflect the last ninety days of conversion data, that content exclusions are at full coverage, and that the schedule reflects the conversion-by-hour pattern. Run the [25-page Google Ads Setup Checklist](/audit/) against the live account before changing targeting, because many cases that look like audience misfit are tracking misfires the audience reports are reflecting. It is free and ungated. When two of the seven signs fire on the same account, or the diagnosis points at something deeper than a settings change, [request a free ads audit](/audit-request/). I write it myself from read-only access and send it back within 24 hours, no call. [The wasted ad spend hub](/wasted-ad-spend/) indexes the adjacent diagnostics. ## When is it appropriate to pause underperforming keywords or ad groups? URL: https://connercrowe.com/wasted-ad-spend/when-to-pause-underperforming-keywords-ad-groups/ Published: 2026-05-25 | Updated: 2026-10-22 Answer: Verify tracking before you pause. On one home-services account, a renamed form button had killed the only website conversion, and 6 of 29 lead-gen accounts I benchmarked counted a non-lead action or click proxy as primary. With tracking clean, pause at twice target CPA with zero conversions across 100 clicks, conversion rate under half the campaign average, or negative margin. ### Check the tracking before any pause rule Every rule below reads the conversion column. If that column is wrong, the rules pause the wrong keywords with full confidence. On one home-services account I worked on, the only website conversion was wired to a Tag Manager trigger that fired on a button reading "Send." Someone renamed the button to "Submit," and the conversion went Inactive without an error anywhere. Every form lead read as zero, and Smart Bidding steered budget toward people who tapped the call button. A zero-conversion rule run on that account would have paused keywords that were producing form leads. After I rebuilt the trigger on the form's own success event, the ninety-day view at the follow-up audit read 582 clicks and 42 conversions, with two keyword themes carrying 86% of them. Both themes were protected during the cleanup. The full case is in [the conversion that died when a button changed its name](/results/dead-form-conversion-fix/). It isn't rare. In my Q3 2026 benchmark, 6 of 29 lead-gen accounts counted a non-lead action or a click proxy as a primary conversion (N = 29), from my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/). In my [home services benchmarks](/home-services-paid-search-benchmarks/), 2 of 11 accounts had tracking too broken to report a cost per lead at all in Q3 2026 (N = 11). Open Goals, then Conversions, read the Status column, and confirm what sits in Primary before you trust a zero. ### Why pause decisions need rules, not instinct A keyword looks expensive in the morning, someone pauses it, and Smart Bidding loses a learning signal on the ad group for the next two weeks. Pausing without a rule is more expensive than pausing too late, because every paused keyword shifts budget into the next worst keyword in the auction. The same goes for un-pausing. On my own account I culled the keyword set on 14 August 2026, and click volume fell, which was the point. An audit a week later read the fall as an over-correction and tried to re-enable four of the paused keywords. I reverted all of it the same day. A drop in clicks you caused on purpose isn't a reason to undo it. The six rules below cover most of the pause decisions a search account needs across a year. Apply them in order. One caveat, about statistical significance on small accounts, sits underneath all six. ### Rule 1: spend hits twice the target CPA with zero conversions across 100 clicks If a keyword has spent twice the target CPA and produced no conversions across at least 100 clicks, pause it. The 100-click floor matters. At 30 clicks a keyword that converts at the campaign average of 3 percent has a 40 percent chance of showing zero conversions through pure variance. At 100 clicks that probability drops below 5 percent. Two-times-CPA at 100 clicks is the point where the keyword has earned the right to be called a loser rather than unlucky. Exception: assist-heavy queries that show conversions in a path report but not a last-click report. Check the path and attribution reports before pausing anything that looks like an upper-funnel term. Brand keywords shouldn't be in the account for this rule to judge. My default is no bidding on a client's own brand: the brand name and its misspellings go in as phrase negatives. The one exception is when a competitor's ad holds the brand search above the organic result, and then I bid on the brand on purpose and judge it by that test, not by this rule. ### Rule 2: conversion rate below half of campaign average across 100 clicks If a keyword converts at less than half the campaign average over 100 clicks, pause it or move it to a lower-bid ad group. Half the campaign average is the threshold where Smart Bidding starts subsidizing the keyword by pulling budget from better performers. A campaign converting at 4 percent overall is paying twice as much per acquisition on a keyword converting at 1.8 percent. The platform will keep bidding on it as long as the target CPA looks like it might still be hit at the campaign level. The lower-bid ad group move is the softer option. Build a "watch list" ad group with manual CPC bids capped at half the campaign average CPC. Move marginal keywords there. If they recover over 60 days, promote them back. If they stay flat, pause. ### Rule 3: Quality Score under 4 for 30 consecutive days If a keyword sits at Quality Score 3 or below for 30 straight days, pause it or rebuild the ad group around it. Quality Score below 4 means the keyword, the ad and the landing page are badly matched. The score itself isn't an auction input, but the same three signals feed Ad Rank in real time, so treat it as a mismatch to fix and judge it by cost per conversion, not as a CPC penalty. Thirty days is the window where short-term ad copy variance gets averaged out and the score reflects the structural mismatch between keyword, ad, and landing page. This rule can't reach a large share of most accounts. In my Q3 2026 benchmark, a median 38% of keyword spend ran on keywords with no Quality Score reported at all (N = 31). Across 30 Google Ads accounts I manage, 30 days to October 21, 2026, 28.8% of pooled keyword spend had no score. For unscored keywords, Rules 1, 2, and 5 do the work. Three diagnostic columns tell you which lever to pull before pausing: Landing Page Experience, Ad Relevance, Expected CTR. Two columns at "below average" usually means the ad group is the wrong home for the keyword. The rebuild is faster than the pause-and-replace cycle on a keyword with real intent. ### Rule 4: zero impressions for 14 days on a budgeted ad group If an ad group has shown zero impressions for 14 consecutive days while the campaign is hitting daily budget, pause the ad group. The 14-day window filters out seasonality and weekend lulls. A budgeted campaign with a frozen ad group means the bid is too low to enter the auction, the keywords have been outcompeted on relevance, or the targeting has collapsed. Pausing returns the budget to ad groups that can spend it. If the campaign isn't hitting daily budget and an ad group has zero impressions, the diagnosis is different. The keywords are too narrow or the bids are below the first-page threshold. That's a rebuild, not a pause. ### Rule 5: search terms 80 percent irrelevant after a negative-keyword pass If a keyword's last 90 days of search terms show 80 percent or more irrelevance after a fresh negative-keyword pass, pause the keyword. Eighty percent is the threshold where the keyword is no longer matching the intent it was named for. A broad match keyword that triggers ten relevant queries and forty irrelevant ones is functioning as a different keyword. Adding negatives slows the bleed, but a keyword that needs a hundred negatives to behave is a keyword that should be paused and replaced with the exact-match versions of the queries that converted. A quick way to run it: in the search terms report, filter to the keyword and sum the cost of the queries that converted at least once in the trailing 90 days. If only a small share of the keyword's spend went to queries that ever converted, it's a pause candidate. ### Rule 6: contribution margin turns negative on the keyword If a keyword's revenue minus cost of goods minus ad spend goes negative across a meaningful sample, pause it regardless of conversion volume. This is the rule that catches the keyword that converts well but on the wrong product. A keyword pushing low-margin SKUs at a target CPA that assumed average-margin SKUs is losing money on every click, even though the campaign dashboard shows it hitting target. Pull conversion value by SKU mix on the top thirty spenders every quarter. Pause the negative-margin keywords or route them to a campaign with a tighter ROAS target. ### Pause decision rules summary | Trigger condition | Threshold | Action | |---|---|---| | Spend without conversions | 2x target CPA on 100+ clicks | Pause keyword | | Conversion rate floor | Under half campaign average across 100 clicks | Pause or move to watch-list group | | Quality Score | Under 4 for 30 consecutive days | Rebuild the ad group, or pause the keyword | | Impression freeze | Zero impressions for 14 days on a campaign hitting budget | Pause ad group | | Search-term irrelevance | Above 80 percent irrelevant after negatives | Pause keyword, add its converting queries as exact match | | Contribution margin | Negative on 30-day rolling sample | Pause or reroute to tighter ROAS target | ### The statistical-significance caveat for small accounts Every rule above assumes the account generates enough clicks for the thresholds to mean anything. Small accounts often don't. My own lead-generation account took 75 clicks from 1 to 25 August 2026. One of its landing pages took 43 clicks and $272.16 for zero leads in that window, which is a real signal about the page and still under the 100-click floor for a keyword verdict. I read it at the page level, which is where the fix sits, and the numbers are in [why I leave about $50 a day unspent in my own Google Ads account](/blog/budget-stopped-being-the-constraint/). An account spending three thousand dollars a month on a long-tail keyword set may have keywords that take four months to reach 100 clicks. Pausing at 30 clicks because spend hit two-times CPA throws away the signal before it could form. The fix is to apply the rules at the ad group level on small accounts, not the keyword level. Aggregate ten keywords into one themed group, run the rules against the group's totals, and pause the group if it fails the test across the same thresholds. A rough proxy: if a keyword has spent less than the cost of one conversion at target CPA, the data is too thin to act on. Wait or consolidate. ### Ad group level vs keyword level: which lever to pull The pause lever changes shape depending on what the data tells you. Pause at the keyword level when one or two keywords inside a healthy ad group are dragging the average down. The other keywords keep the ad group's learning signal alive and the budget routes to them. This is the default pause for Rules 1, 2, 5, and 6. Pause at the ad group level when the failure is structural: most keywords in the group share the same Quality Score problem, the same intent mismatch, or the same zero-impression freeze. Pausing one keyword at a time inside a broken ad group buys two weeks before the next one fails the test. Pause the whole group, rebuild the keyword list around the queries that converted, and relaunch with three new responsive search ads. This is the default for Rules 3 and 4. When two of the six pause rules fire on the same ad group, the [free 25-page Setup Checklist](/audit/) is the next read, because the settings it covers are often why; the [contact form](/contact/) is open when the diagnosis points at something deeper than a settings change. The full [wasted-ad-spend library](/wasted-ad-spend/) covers the rebuild patterns once the pause decisions are clean. ## Which Google Ads match type mistakes waste ad spend? URL: https://connercrowe.com/wasted-ad-spend/typical-mistakes-keyword-matching-wasted-spend/ Published: 2026-05-25 | Updated: 2026-10-30 Answer: Six match-type mistakes leak search spend, starting with broad match and no negative list. The typical account I manage put 30.7% of visible search-term spend into terms with 2+ clicks and zero conversions in Q3 2026. Tighter doesn't mean cheaper: in one law firm account, phrase cost 23% to 68% more per click than broad on the same keywords. ### Why match-type waste hides longer than it should Bad keywords get caught in the search-terms report. Bad match types get caught in the conversion data three months later, after Google has already spent the budget. The platform changed how every match type behaves between 2018 and 2022, and an account built on the old playbook still bids as if it hadn't. Six patterns cause almost all of it. Each has a fix that takes under an hour, and two checks at the end of this page show which ones your account has. ### The six mistakes at a glance | Mistake | Where it shows up | The fix | |---|---|---| | Broad match with no negative list | Search terms full of unrelated queries | Campaign-level negative list, reviewed every quarter | | Phrase match in unthemed ad groups | Low Ad Relevance, weak cost per conversion | Three to five keywords per ad group, grouped by intent | | Exact match close variants nobody reads | Exact keywords matching queries you never chose | Monthly close-variant review: promote or negate | | Mixed match types in one ad group | Broad taking clicks exact should have won | One match type per ad group, cross-negated | | Negative lists nobody prunes | The same junk queries back every quarter | One shared list, one owner, a review date | | Broad plus Smart Bidding on thin data | Spend climbs, then conversion rate falls | Exact and phrase first, broad after 30 conversions a month | ### Mistake 1: broad match running without a negative-keyword list Broad match without negatives is the most common. The ad group sits at default settings, Google matches the keyword to anything it considers thematically related, and the search-terms report fills with queries that have no commercial connection to the product. Google's [broad match algorithm](https://support.google.com/google-ads/answer/2497828) has shifted toward intent-based matching since 2021, which sounds tighter and expanded the surface area in practice. A broad-match keyword for "leather sofa" now triggers on queries about furniture care, sofa repair tutorials, and competitor brand names. None of that converts on a commercial landing page. None of it gets blocked unless a negative-keyword list catches it. For home and furniture clients I run [a separate playbook](/for-home-brands/furniture/) on the negative-list patterns that repeat across the vertical. To find it, filter the keywords view to match type equals broad and group by ad group. Open the search terms tab on each one and read the last sixty days. If the ad group has no associated negative-keyword list, or the list hasn't been touched in a quarter, you have found a leak. The fix is a campaign-level negative-keyword list attached to every campaign that runs broad match, reviewed every quarter against a fresh search-terms pull. Accounts that skip that discipline keep paying for queries that never had a chance of converting. ### Mistake 2: phrase match inside ad groups with no shared theme Phrase match works when the ad group has a tight thematic anchor. Three to five phrase-match keywords that all describe the same intent, paired with ad copy written to that theme, gives Google a clean signal about what the ad group serves. The mistake is dumping fifteen unrelated phrase-match keywords into one ad group because they sat near each other in a keyword tool export. Google now has to judge ad relevance against fifteen different intents using one set of responsive search ads. Ad Relevance sags, and the ad group looks like it's performing because impressions stay high. Don't expect the damage to show up in CPC. Of 30 Google Ads accounts I manage, 7 had enough clicks on both bands in the 30 days to October 21, 2026. In the typical one, keywords on Quality Score 1 to 5 cost 0.98x what QS 7 to 10 keywords cost per click, and they cost more in only 3 of the 7 (N = 7). A low score earns attention when the clicks it buys don't convert, so read it next to cost per conversion. The diagnostic columns tell you which lever to pull: low Ad Relevance points at the copy, low Landing Page Experience at the page, and low Expected CTR at intent mismatch, which usually traces back to this mistake. [Ad copy relevance and click-through rate](/wasted-ad-spend/ad-copy-relevance-poor-click-through-rates/) covers the copy side. The opposite error costs the same money. Filter the keywords view to ad groups with one active keyword and count them. A few single-keyword ad groups inside a deliberate SKAG structure is fine. Dozens of them usually means the account was broken apart for tidy-looking reports and now runs thin ad copy against keyword counts too low for Google to learn on. Both fixes land in the same place. Group keywords by intent rather than tool category, aim for three to five per ad group, write three responsive search ads tuned to that theme, and read Ad Relevance and cost per conversion again after two weeks. ### Mistake 3: exact match expanded by close variants without monitoring Exact match no longer means exact. Google's [exact match documentation](https://support.google.com/google-ads/answer/2497825) states that exact match matches queries with the "same meaning" as the keyword, not the same words. Plural forms, synonyms, reordered terms, paraphrases, and intent-equivalent rewrites can all qualify under close-variant matching. The cost shows up in two places. First, exact-match keywords trigger on queries the account owner never approved, and those queries skip the scrutiny exact match was meant to provide. Second, close-variant expansion overlaps with phrase and broad match in the same campaign, so Google decides which keyword to attribute the click to, and the auction logic favors the keyword with the higher bid. The fix is to read the search-terms report against exact-match keywords every month. Tag any query that isn't the literal keyword. If the close variant converts well, keep it and add it as its own exact-match keyword. If it converts poorly or carries the wrong intent, add it as a negative at the exact-match level. Sum the close-variant cost against total exact-match cost each month to see how much of the spend isn't literal. ### Mistake 4: mixing match types in one ad group Putting broad, phrase, and exact versions of the same keyword in one ad group looks tidy on a spreadsheet. Inside Google's auction it creates internal competition. The platform picks one keyword by its own prioritization rules and Ad Rank, then attributes the click and conversion to that keyword. The result is messy reporting and unstable bid signals. Broad match cannibalizes exact match impressions. Exact match starves of conversion data because broad caught the query first. Smart Bidding can't read clean signal off any of the three. The fix is one match type per ad group. Build a tight-themed exact-match ad group for the queries that already convert, a phrase-match ad group for adjacent intents, and a broad-match ad group only once a negative-keyword list and a Smart Bidding strategy are both in place. Cross-negate between them so the same query can't trigger more than one ad group. Then judge each match type by cost per conversion, not by CPC. Here's one law firm account I manage, over the 30 days to October 20, 2026: on six keywords running in both broad and phrase, [phrase cost 23% to 68% more per click than broad](/wasted-ad-spend/can-i-see-competitor-cpc-google-ads/). Part of that gap is broad entering looser auctions, which also brings in more irrelevant searches, so the cheaper click only wins if it converts at a similar cost. A cheaper click can still cost more where it counts. On one service-business account, a [14-day AI Max test](/blog/google-ads-search-term-matching-ai-max/) cut CPC from $3.89 to $2.26, down 42%, while cost per qualified lead rose from $178 to $313, up 76% (N = 1). [AI Max expanded matches](/wasted-ad-spend/what-are-ai-max-expanded-matches-and-landing-page-matches/) are the newest version of this problem. ### Mistake 5: negative lists nobody has pruned Negative lists rot. Queries flagged in a past search-term review often never make it onto the list. The fix happened in a sheet that nobody pushed into the account, or the list owner left, or the list was attached at the ad-group level instead of the campaign level and only protected one of seven campaigns. Google doesn't flag missing negatives. The optimization score recommends adding broad match keywords, not blocking the queries those keywords pulled in last quarter. The miss compounds. Every ninety days the account spends more on queries that should have been negated two audits ago. The fix is one shared negative-keyword list attached at the account or campaign level, owned by one person, reviewed every quarter. Pull the last ninety days of search terms, add the new irrelevant queries, then prune the old negatives that no longer match served queries. A 2023 list blocking phrases nobody searches anymore costs nothing. A list missing the twenty queries Google started routing to the account last month costs real money. Document the date the list was last reviewed. The [wasted-ad-spend library](/wasted-ad-spend/) covers the list-management workflow in depth. ### Mistake 6: broad match plus Smart Bidding before conversion data stabilizes Google recommends broad match paired with Target CPA or Target ROAS bidding inside the optimization score. The recommendation only works when the conversion data feeding the bidding algorithm is clean and stable. In Q3 2026, 6 of 29 lead-gen accounts in my benchmark counted a non-lead action or a click proxy as a primary conversion (N = 29), and Smart Bidding on broad match learns from whatever that column counts. When broad match runs with Smart Bidding on thin conversion data, the algorithm chases noise. It bids up on queries that converted once by accident, then burns budget while it learns. The first sixty to ninety days look productive because spend climbs. The next sixty to ninety days look terrible because conversion rate collapses on the queries the algorithm bet on. The fix is sequencing. Run exact and phrase match on Manual CPC, or Maximize Clicks with a CPC cap, until the campaign has at least thirty conversions per month against stable creative. Then layer in broad match with Target CPA, give Smart Bidding two weeks to learn, and read the search-terms report weekly for the first month. Pull broad match if cost per conversion climbs above the manual baseline and stays there for two weeks. ### Two checks that find the leak in your account The six mistakes explain why an account keeps refilling with the same queries. These two checks show how much it's costing you right now. #### Check 1: ninety days of search terms sorted by cost Open the search-terms report at the account level, set the date range to the last ninety days, and sort by cost descending. Export the top three hundred queries to a sheet, with the match type of the keyword that caught each one. Filter for queries with zero conversions and more than one hundred dollars in spend. On a mid-sized account that single filter can return dozens of rows. In my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/), the typical account put 30.7% of the search-term spend Google showed into terms with 2 or more clicks and zero conversions, with a range of 10.0% to 56.5% (N = 25). For the typical account, the report showed only 52% of account spend (N = 11), so the leak you can read is part of the leak you have. Group the rows by match type and the mistake behind them usually names itself. Tag each query as relevant, irrelevant, or wrong intent, and add the irrelevant and wrong-intent queries to a campaign-level negative list before anything else moves. Tag proper nouns against the live catalog or service list rather than from memory. A brand the business stocks or a staff member it employs can sit in the zero-conversion list for reasons that have nothing to do with relevance, and negating those blocks the buyer instead of the leak. On one law firm rebuild, the first ninety-day pull surfaced budget going to DIY-contract researchers, real-estate agents shopping for marketing services, and people researching unrelated legal issues. Three hundred negatives went in over the first thirty days. Within ninety days spend was down 50% and qualified signups had doubled, though the negatives were one of four changes in that window and I didn't isolate which did the most. [Law firms hit this differently](/for-service-brands/law-firms/), and the negative-list build looks nothing like the ecommerce template. #### Check 2: your own brand name inside non-brand keywords Run the search-terms report across every campaign and filter for queries containing the brand name or any misspelling of it. Each row is a click the organic listing would have delivered free, matched in by broad, phrase, or a close variant. Those clicks also convert well, which flatters the keyword that caught them. Compare conversion rate by match type inside each campaign. If broad match converts at double the rate of exact, brand interception is a likely cause. My default is to stay out of that auction. The brand name and its misspellings go in as phrase negatives across the account, and the organic listing takes those searches. The one exception is a competitor's ad holding your brand search above your organic result. Then a separate brand campaign is worth testing, kept out of every non-brand number. Move the brand negatives from the account level to a shared list applied to every non-brand campaign, and remember the brand campaign's own search-terms report needs the same filter in reverse, because non-brand queries slip into it at brand bids. Performance Max does the same thing on a store. On one ecommerce store I manage, a single Performance Max campaign was serving on brand searches and claiming credit for buyers who would have closed through organic anyway. Reported ROAS looked healthy and non-brand performance was invisible underneath. Excluding the brand term from the campaign was the first time the two could be read separately. Informational queries on commercial campaigns ("how does", "what is", "ideas for") are the other leak these checks turn up. [Signs your ads target the wrong audience](/wasted-ad-spend/warning-signs-ads-targeting-wrong-audience/) covers that one, and [how to audit a digital ad account](/wasted-ad-spend/best-practices-audit-digital-advertising-account/) covers the full account pass beyond keywords. ### The audit cadence Run both checks once across every campaign. Fix the worst three mistakes they point to. Then re-run the search-terms and negative-keyword review every ninety days. The accounts that stay clean are the ones that treat negative-keyword hygiene as a recurring calendar item rather than a one-time cleanup. I run this exact pass against client accounts as the opening step of any paid engagement. The [services overview](/services/) covers what the rebuild looks like, and the [contact form](/contact/) is where the conversation starts when the leak is bigger than a weekend fix. ## Which metrics should I monitor to catch poor ad performance? URL: https://connercrowe.com/wasted-ad-spend/key-metrics-monitor-detect-poor-ad-performance/ Published: 2026-05-25 | Updated: 2026-10-30 Answer: Watch four numbers weekly: CPA trend, spend pacing, zero-conversion spend share, and impression share lost to budget versus rank. In 23 of 31 Search accounts I manage, budget cost more impression share than rank in Q3 2026 (N = 31). Monthly, check what the conversion column counts: all conversions typically ran 2.3x primary across 22 lead-gen accounts. ### The Monday written summary Every account I run gets a Monday morning written summary. Four numbers, one paragraph, sent before anything else opens. The point isn't the document. It's reading the same numbers the same way every seven days, so a drift registers as a drift instead of a month-end surprise. Each number below names its report. ### The four weekly numbers #### CPA trend Cost per acquisition at the account level, then for the top three campaigns by spend, on a rolling seven-day window against the prior seven days and the trailing twenty-eight. One week in isolation is noise; the trend across three windows is signal. First move: that campaign's search terms report, before bids. #### Spend pacing Month-to-date spend divided by days elapsed, multiplied by days in the month, against budget. Pacing well over budget in the first ten days means a bid strategy off its leash. Pacing well under means starving the auction. First move: change history for the prior fourteen days. #### Zero-conversion spend share The report: campaigns or ad groups view, cost descending, conversions filtered to zero, last thirty days. Total every ad group that spent over a hundred dollars and produced nothing, divided by total account spend, and watch the trend week to week. For a reference point, the typical account put 30.7% of its visible search-term spend into terms with 2 or more clicks and zero conversions, across the accounts in my [Q3 2026 Google Ads benchmarks](/benchmarks/google-ads-account-benchmarks/) (N = 25). The cause is usually three or four ad groups the bid strategy won't defund because its conversion signal arrives too rarely to retrain the model. #### Impression share, split by budget and rank The report: campaigns view with Search Lost IS (budget) and [Search Lost IS (rank)](https://support.google.com/google-ads/answer/2497703) added, week over week. Lost-to-budget means the campaign hits daily cap and Google would spend more. Lost-to-rank means it can't win the auction at the bid and Quality Score it has. The two dictate completely different fixes: impressions lost to budget on a healthy conversion rate leave revenue on the table, while a large loss to rank means you're paying up for the impressions you do win. Budget is the more common ceiling in my book. In 23 of 31 Search accounts in my [Q3 2026 benchmarks](/benchmarks/google-ads-account-benchmarks/), impression share lost to budget was larger than impression share lost to rank, and, taking the middle account on each measure, 37.9% was lost to budget against 26.2% to rank (N = 31). Read both columns before you raise a bid. Read a sudden drop in impression share as an instruction to look rather than an instruction to bid, and read it alongside the budget column, because the three impression-share figures sum to 100% and a campaign losing most of its impressions to budget has a hard ceiling on what its rank column can report. Fixing the budget loss makes the rank number climb on its own. Lost-to-rank climbing while lost-to-budget falls toward zero can also be the correct resting state on an account whose bids were capped on purpose, and clearing the first-page floor is worth doing only when the implied cost per lead, divided by your close rate, still clears what a sale earns. [Should I raise bids when the budget isn't spending](/wasted-ad-spend/should-i-raise-bids-when-the-budget-is-not-spending/) works that comparison through on an account where the answer was no. ### Search campaigns: seven numbers behind three reports The default view rewards more match types and more spend. These seven sit further in. #### Search-term irrelevance rate The report: search terms, last ninety days, cost descending. Tag the top two hundred queries relevant or irrelevant, divide irrelevant cost by total cost. Know what the report leaves out before you trust the ratio. Across the 11 accounts where I pulled every search-term row in Q3 2026, the terms Google showed covered 52% of account spend for the typical account, with a range of 20% to 71% (N = 11). The irrelevance rate you calculate describes the visible half. Tight match types keep it low, and broad match needs closer reading. The fix is rarely a longer negative list. It's tighter match types, audience signals on Performance Max, and a decision about where broad match operates. #### Top-spend keyword performance On most Search campaigns a handful of keywords carry most of the cost. Read them weekly: spend, clicks, conversions, conversion rate, CPA. One going off the rails masks a healthy account average for two or three weeks, so reading the ten weekly catches it on day seven instead of day twenty-one. First move: the search terms it triggered, and its match type. #### Quality Score on top-spend keywords The report: keywords view with [Quality Score](https://support.google.com/google-ads/answer/6167118), Landing Page Experience, Ad Relevance, and Expected CTR added, cost descending. A score below 5 on a keyword taking more than a few percent of campaign spend is a warning sign. The score itself isn't an auction input, but it summarizes the same three quality signals Ad Rank reads in real time. Don't read it as a CPC verdict. Across 30 Google Ads accounts I manage, 30 days to October 21, 2026, the 7 accounts with enough clicks in both Quality Score bands (1 to 5 and 7 to 10) typically paid 0.98x as much per click on the low-score keywords, and low-score clicks cost more in only 3 of the 7 (N = 7, range 0.89x to 3.86x). Under Smart Bidding the bid moves too, so judge a weak score by cost per conversion. Three below 5 in the top ten by spend calls for a landing-page and ad-copy rebuild. Two is a warning. One is noise. Monthly, widen to the top fifty and bucket them seven-or-higher, four-to-six, three-or-lower, plus a fourth bucket for keywords with no score. That last bucket is often the biggest. In my Q3 2026 benchmarks, the typical account had 38% of its keyword spend on keywords with no Quality Score reported (N = 31). Counted separately, the typical account had 4% on keywords scored 7 to 10 (N = 31). A growing share slipping into the bottom bucket means the account's ads and landing pages are drifting away from the queries it pays for. #### Click-through rate on exact-match keywords The report: keywords view filtered to match type equals exact, impressions descending. Read each keyword against its own history. A CTR that collapses with real impression volume means the keyword is matching queries it no longer represents, or the ad misses the intent. #### Conversion rate by match type The report: dimensions or segment view, segmented by match type at ad group level. Exact should convert highest, phrase lower, broad meaningfully lower than phrase. When the order inverts the data is lying, usually from a tracking de-duplication failure or searches for your own brand name slipping into broad. A broad-match rate that doubles the exact rate on the same keyword root often means broad is picking up people who were already looking for you. My default is that no campaign bids on the client's own brand: the brand name and its misspellings sit as phrase-match negatives at the account level, and the one exception is a competitor's ad holding the brand search above the organic result. The [Tracking Stack reference](/frameworks/tracking-stack/) covers the de-duplication contract. #### Impression share by audience segment The report: audiences view inside a campaign, impression share segmented by audience. [Customer Match](https://support.google.com/google-ads/answer/6379332) lists fed from email and SMS should clock the highest impression share and conversion rate, because logged-in past purchasers searching category terms are the cheapest conversions in the account. If they sit low while cold audiences absorb the spend, layer the audience as a bid modifier and raise the floor. #### Top impression rate against rising CPC The report: search campaigns view with Top Impression Rate, Absolute Top Impression Rate, and Average CPC added, last thirty days against the previous thirty. When top impression rate falls while CPC rises, you're paying more per click for worse placement. You're outbid on rank, not on budget, and more bid won't recover the position. Rising CPC alone isn't the alarm. In my [Google Ads CPC trend](/benchmarks/google-ads-cpc-trend/), 16 of 22 accounts paid more per Search click in Q3 2026 than a year earlier (N = 22). ### Display and Performance Max: six numbers Google buries Search has the search-terms report, where a bad query carries a name and a cost. Display has no equivalent. Google files placements under Insights, and Performance Max scrubs most of it into a "Placements where your ads appeared" download with no spend column. These placements can absorb a large share of the budget, and they rarely get audited. #### Placement category share The report: Insights or Where ads showed, sorted by cost, mobile-app placements broken out. Most meaningful engagement should come from a short list of placements. A thousand placements taking ten dollars each is the canonical Performance Max display leak. First move: export placements, sort by cost, exclude the long tail below a minimum click threshold. Thirty minutes in the account-level exclusion list at the MCC. #### Viewability rate The report: campaigns view with Active View Viewable Impressions divided by total impressions, or the Insights tab on Performance Max. Google's [Active View documentation](https://support.google.com/google-ads/answer/7029393) defines a viewable display impression as fifty percent of the ad on screen for one second, and a viewable video impression as fifty percent on screen while playing for two seconds. Track it against its own history, and exclude the placements that drag it down. #### Mobile-app placement spend The report: placement type filter set to mobile-app in the Where ads showed view. The highest-friction inventory in the system: children's games and lock-screen tap traps. Click-through rates look exceptional, conversion rates collapse, session duration sits under three seconds because most clicks were accidental. Exclude mobileappcategory::69500 (Games), then the Display Network mobile-app exclusion at account level. [Performance Max](https://support.google.com/google-ads/answer/10724817) takes the same exclusions through account-level placement exclusions, which most accounts never open. #### View-through to click ratio The report: conversions segmented by attribution type, view-through against click-through. A view-through conversion fires when someone saw the ad and converted later without clicking. Useful as a directional signal, dangerous as a KPI. When view-through conversions dwarf click conversions, the platform is claiming conversions off impressions that may never have been seen. Keep the view-through window short and keep those conversions out of any ROAS used to set bids. Four-times return on clicks and twelve-times on view-through is a campaign that needs the click number. #### Audience overlap The report: audience insights at campaign level with overlap percentage against Customer Match and remarketing lists. On Meta, the audience overlap tool. A prospecting campaign that overlaps the remarketing list is buying impressions on people who already converted, and retargeting would have reached them at a lower CPM. Performance Max does this constantly, firing at warm audiences because they convert faster and make the asset group look efficient. Meta fails differently: two prospecting ad sets on overlapping audiences bid against each other while both report as functional. First move: consolidate or rebuild into distinct buckets. #### Engagement and session duration on paid traffic The report: GA4 traffic-acquisition view, source filtered to paid, [engaged sessions and average session duration](https://support.google.com/analytics/answer/12253918) against total sessions. GA4 counts a session as engaged if it lasts ten seconds or longer, fires a conversion event, or includes two or more pageviews. Read paid engagement against organic engagement on those same URLs. A paid segment far below organic means broad-match queries on irrelevant pages, app placements, or a page that renders slowly on mobile. Segment to display and read session duration: a cluster of sessions lasting a few seconds means accidental clicks dominate. ### The monthly revenue check Five numbers on the first Monday, all asking whether the spend turned into money. #### Blended ROAS against Shopify Compare platform-reported results against what the store recorded for the same range: Shopify Analytics, Reports, Sales by traffic source, and the Shopify orders GA4 attributes to paid search on a last-non-direct basis. The gap is the metric. Google Ads conversions within ten percent of those paid-search orders says the tracking stack is honest. On a multi-channel account, GA4 paid-channel conversions should land within fifteen percent of the sum of Google Ads and Meta reported conversions. Wider than either band says the platform is claiming results the store can't see, and no bid change fixes a tracking problem. [The Google Ads versus Shopify reconciliation](/conversion-tracking/google-ads-conversions-dont-match-shopify-orders/) lists the causes in the order I rule them out. Sales by referrer adds the second layer, showing referring URLs, and catches a click parameter stripped at checkout. First move: server-side conversions, a GA4 enhanced measurement audit, and a recheck of whether the consent banner is blocking the pixel. #### New-customer ROAS Total ROAS hides repeat purchases. A returning-customer order through a paid click is a margin loss most months, because that customer would have bought through email or direct. New-customer ROAS, isolated through Shopify customer reports or the platform's new-customer acquisition column, says whether paid pays for itself. A blended ROAS of three-point-five against a new-customer ROAS of one-point-two means paid is subsidizing the existing-customer base rather than building one. #### Primary conversions against all conversions The report: conversions view with the Conversions and All conv. columns side by side, last thirty days, then the conversion actions list with every action's category and primary or secondary status. The gap shows how much the account records that it doesn't bid on, and it's where non-lead actions hide. Across 22 lead-gen accounts in my [Q3 2026 benchmarks](/benchmarks/google-ads-account-benchmarks/), all conversions typically ran 2.3x primary conversions, and 5 of the 22 ran above 10x (N = 22). 6 of 29 lead-gen accounts counted a non-lead action or a click proxy as a primary conversion (N = 29). Read the category of every primary action, not its name. Broken conversion tracking is the most common finding in the audits I ran in 2026, and this screen is where it shows first. #### Cost per conversion against target CPA The report: campaigns view with Cost per Conversion and Target CPA visible, segmented by week for the last ninety days. A single week over target is noise. Several consecutive weeks well above target is structural. Both platforms let actual CPA roam above target on the belief a future conversion is likely, and three weeks of being wrong calls for a manual reset, not more learning time. The fix is a tighter conversion definition, a lower target, or a pause on the worst ad group. #### Conversion volume against rising spend The report: campaigns view with Cost and Conversions plotted as a time series, last six months, weekly. In a healthy account the lines move together; in a decaying one spend climbs while conversions flatten. The cause is audience saturation, creative fatigue past recovery, or a tracking change that broke attribution. Check whether the new cost per conversion still clears margin before you add budget. ### The four reports that explain the numbers Metrics say something moved. These four say why, in this order. **Auction insights**, campaign and ad group level, top three spenders. Watch for impression share lost to budget rising on campaigns that used to clear inventory, and a new competitor in the top three with a high overlap rate. Fix: bid cap review, ad rank diagnosis, and a check on a competitor's new price promotion. **Change history**, the most under-used report in Google Ads. Filter the break period by change type: budget, bid strategy, Performance Max assets, conversion goals. Often one change explains the entire drop, usually a bid strategy switched to maximize conversions on thin conversion data. Fix: revert it and run the original strategy for a full conversion cycle. **The Performance Max asset report**, next to the placement report. The asset report often shows the algorithm fixated on one headline or image that produces clicks and no orders. Fix: placement exclusions and asset group splits by intent. **GA4 landing pages**: Pages and Screens, Session source / medium secondary, filtered to google / cpc, read against engagement rate. A page taking a big share of paid traffic at weak engagement and a low conversion rate is the leak, not the keyword that sent the click. Fix: rebuild it or reroute the traffic. ### The cadence is the product On a home-services client I monitor weekly, the Meta dashboard showed a large click count at a very low cost per click across the trailing ninety days. It looked excellent by every surface metric. The cadence caught what the dashboard didn't: the campaigns were lead-form objectives where every form open registered as a click, and only a small fraction of those opens became completed forms. The founder would have kept scaling spend against the leak. A related gap is common: 9 of the 18 accounts in my [Q3 2026 Meta benchmarks](/benchmarks/meta-ads-cpm-by-placement/), carrying 42% of spend, optimized for no conversion event at all (N = 18). The metric that matters is rarely the metric the platform highlights. If I open one number first it's the zero-conversion spend share, the fastest read on whether the problem is tuning or structure. On Search that job belongs to search-term irrelevance, which distorts every metric downstream. On Display it's placement category share: the long tail explains the collapsed viewability, the mobile-app spend, and the three-second sessions. The numbers matter less than reading them on the same day, in the same order, every week. Run the [free 25-page Setup Checklist](/audit/) for the diagnostic written out, or read the [wasted-ad-spend library](/wasted-ad-spend/). The [services overview](/services/) walks the structural fix. Founders who want me running this cadence can start with my [free written audit](/audit-request/), sent with read-only access. ## Which tools can help analyze ad spend efficiency in paid search campaigns? URL: https://connercrowe.com/wasted-ad-spend/tools-to-analyze-ad-spend-efficiency/ Published: 2026-05-25 | Updated: 2026-10-30 Answer: I start with the free native reports, which hide more than people think. In my Q3 2026 benchmark, the typical account’s search terms report showed 52% of spend (N = 11), and the typical account put 30.7% of what it showed into terms with 2+ clicks and no conversions (N = 25). Paid tools earn their fee at higher spend. ### Free native tools inside Google Ads The platform ships with three reports that diagnose most wasted-spend problems if you read them honestly. The [search terms report](https://support.google.com/google-ads/answer/2472708) is the single most-used report in the account, and it shows less than people assume. In my [Q3 2026 benchmark](/benchmarks/google-ads-account-benchmarks/) of the Google Ads accounts I manage, it showed 52% of account spend for the typical account (N = 11), because Google hides low-volume searches. Of the spend it did show, the typical account put 30.7% into terms with two or more clicks and no conversions (N = 25). Roughly half the spend never shows up in that report, and close to a third of what does show went to terms that clicked twice or more and never converted. That's why I read it first. Filter to ninety days, sort by impressions, and read the top hundred queries the way a stranger would. If a quarter of the list describes products you don't sell or jobs you don't staff, broad match is teaching the algorithm the wrong business. [Auction insights](https://support.google.com/google-ads/answer/2579754) tells you who else is bidding against you and at what overlap rate. A sudden jump in impression share lost to budget, paired with a new competitor in the top three, usually explains a CPC spike that the recommendations tab won't. The [recommendations tab](https://support.google.com/google-ads/answer/3448398) is where founders get burned. Google's [optimization score](https://support.google.com/google-ads/answer/9061546) rises as you apply or dismiss its recommendations, each weighted by Google's own estimate of its impact, so it measures how much of Google's suggested work you've handled rather than whether the account is profitable. Most of those recommendations push toward broader match, higher budgets, and more campaigns. Accepting blind is a reliable way to add waste. Read each recommendation against the search terms report before clicking apply. If the recommendation expands match types on a campaign already leaking irrelevant queries, decline it. ### Shopify is the ledger, every ad platform is a claim Google Ads reports its own attribution. Meta reports its own. GA4 reports a third. They almost never match, and the gap is where the answer lives. Orders inside the Shopify admin are the source of truth no ad platform can overrule. Every other platform is reporting an attempted attribution of those orders. If Google Ads says one hundred conversions, Meta says eighty, and Shopify shows one hundred forty paid orders for the period, the gap is either organic, email, direct, or double-counting between paid channels. That gap is the question worth answering. Google Ads conversion tracking, set against a server-side tag and deduplicated against Shopify, reports how many orders the platform thinks it earned, by campaign, by keyword, by device. Meta's Events Manager does the same job for Facebook and Instagram with one caveat. Since iOS 14.5, Meta's reported conversions tend to run optimistic against Shopify's order log, by more on stores with a high iOS share. The platform isn't lying. It's modeling. Read Meta conversions as directional, not as ledger truth. ### GA4 cross-checked with Shopify or your CRM | Comparison | Should land within | |---|---| | Google Ads conversions against Shopify orders attributed to paid search, last-non-direct in GA4 | ten percent; a gap over twenty-five percent is a tracking problem | | GA4 paid-channel conversions against the sum of Google Ads and Meta reported conversions | fifteen percent | | GA4 paid total against Shopify paid-source orders | twenty percent | GA4 is the lens founders most often ignore, because the interface is rougher than Universal Analytics was. Its free funnel exploration and attribution reports do what paid SaaS charges three hundred dollars a month for: last-non-direct attribution across paid search, paid social, organic, email, and direct, rolled up against actual Shopify orders. Two thresholds matter. Google Ads conversions should land within ten percent of Shopify orders attributed to paid search through a last-non-direct model in GA4, and a gap larger than twenty-five percent is a tracking problem, not a performance problem. No bid adjustment fixes it. On a multi-channel account, GA4 paid-channel conversions should land within fifteen percent of the sum of Google Ads and Meta reported conversions, and the GA4 paid total within twenty percent of Shopify's paid-source orders. Wider than that is the same diagnosis. The [Tracking Stack reference](/frameworks/tracking-stack/) covers the deduplication contract and the GA4-to-Shopify reconciliation that closes the gap. The full band table, with the measurement basis for each pair, sits on [why Google Ads conversions don't match Shopify orders](/conversion-tracking/google-ads-conversions-dont-match-shopify-orders/). The cross-check takes fifteen minutes a week in a spreadsheet. Five numbers: Google Ads conversions, Meta conversions, GA4 paid sessions and conversions, Shopify paid orders, Shopify revenue. The pattern across weeks tells the story no single dashboard does. Funnel exploration also answers what paid-search tools can't: whether clicks land on the right page and add to cart, or bounce inside fifteen seconds. That's a landing-page problem masquerading as a media problem, and the diagnosis is free. ### Microsoft Advertising and the publisher report [Microsoft Advertising](https://learn.microsoft.com/en-us/advertising/msa-help/hlp_ba_proc_createreport) ships the same three core reports. Run the search terms report on a ninety-day, top-hundred read. Broad match leaks Bing budget the way it leaks Google budget. Google now lists search partner sites, but with impressions only. Microsoft's website URL (publisher) report shows spend and conversions by site, which is what you need to cut waste. Microsoft serves ads across Bing, partner search sites and the Audience Network, and there's no clean opt-out from Audience ads on Search campaigns. The sites that take spend that return nothing go on a [website exclusion list](https://learn.microsoft.com/en-us/advertising/msa-help/hlp_ba_proc_addexclusions). Google Import is the other Bing-only leak: by default, a scheduled import can overwrite the bids, budgets and statuses you changed in Microsoft Ads. Both checks, plus Copilot placements and UET, are on [what a free Microsoft Ads audit should check](/wasted-ad-spend/free-microsoft-ads-bing-audit/). My free custom audit covers Microsoft Advertising accounts on the same terms as my Google Ads audit. [Microsoft Clarity](https://clarity.microsoft.com/) is free, with no traffic cap on heatmaps or session recordings. Most Bing advertisers skip it because they already run GA4. GA4 reports what happened in aggregate. Clarity shows where the visitor clicked, how far they scrolled and which sessions had rage clicks or dead clicks. A Google campaign and a Bing campaign can both convert at one percent in GA4 and still behave differently in Clarity, because the same page can work for Google's visitors and stall Bing's. If you install it, name it in your privacy policy. A UET tag can also load Clarity on its own, so check whether yours [already does](/blog/bing-ads-tag-loads-clarity-session-recording/) before you add a second copy. ### Paid third-party audit platforms [Optmyzr](https://www.optmyzr.com/help/) is the best-known rule-based auditor, priced by ad spend. I don't run it. My own checks run through scripts I built against the Google Ads API: Monday reports pulled straight from the API, and a weekly negative-keyword pass that waits for my approval before anything is written. That's the choice an operator with many accounts makes. A single-account founder needs neither, because the free reports plus a fifteen-minute weekly read cover the same ground. [Adalysis](https://docs.adalysis.com/manage/ads/rsas/overview) is the closest competitor and stronger on responsive search ads. If RSAs are the bottleneck (low ad strength, weak headline variance), it's the more targeted of the two. Pricing is set by a spend slider, with no flat starting figure published. SEMrush's PPC toolkit is competitor research rather than account hygiene: the keyword gap report and the ad copy library show what competitors run and how their headlines test against yours. Treat it as a research tool. Most paid PPC audit tools price for agencies running fifty accounts, and a single-account founder paying for Optmyzr buys capacity they never use. The [audit pricing breakdown at /google-ads-audit-cost](/google-ads-audit-cost/) walks the market tiers for a one-time senior human review of the same account. Mine is free. | Tool | Category | Cost | Best for | What it catches | |---|---|---|---|---| | Optmyzr | Paid SaaS auditor | Paid, scales by ad spend | Accounts above $50K/mo spend | Bid drift, quality score decay | | Adalysis | Paid SaaS auditor | Priced by ad spend | RSA-heavy accounts | Ad strength, headline variance | | SEMrush PPC toolkit | Competitor research | Paid plans | Researching competitor ads | Keyword gaps, competitor copy | | Google Ads native reports | Free platform tool | Free | Single-account founders | Search terms, auction shifts | | Microsoft Advertising native | Free platform tool | Free | Any Bing-active account | Partner and publisher spend | | Microsoft Clarity | Free analytics | Free | Any site taking Bing traffic | Heatmaps, session recordings | | Google Ads Setup Checklist PDF | Free DIY workbook | Free | Founders auditing their own setup | Ten settings and tracking red flags | ### Multi-channel attribution platforms [Triple Whale](https://kb.triplewhale.com/) is one of the most widely used attribution platforms in the Shopify ecosystem. It connects Shopify, Meta, Google Ads, Klaviyo, and TikTok, then reconciles them into a single revenue view with post-purchase survey data on top. That survey, the one asking new customers where they first heard about the brand, is the strongest part of the product. It now has a free tier, and paid plans scale with revenue, so check current pricing before you budget for it. [Northbeam](https://docs.northbeam.io/docs/what-is-northbeam) targets the same buyer with a first-party tracking layer and an MTA model (multi-touch attribution) some operators prefer over Triple Whale's pixel reconciliation. It earns its line item where paid social assists and paid search closes, and the founder needs to credit both honestly. Polar Analytics sits below both on price, trading attribution depth for a cleaner dashboard. All three earn the line item above fifty thousand a month with multiple active channels. Under that the math rarely works: a two-thousand-a-month account paying three hundred for Triple Whale spends fifteen percent of media on tooling for marginal gains. [Hyros](https://hyros.com/) serves a different buyer: info products, coaching, agencies, and long-sales-cycle lead-gen where the conversion is a call booked, a lead captured, or a trip-wire that earns the real revenue weeks later. It tracks a click from ad to lead to sale across email opens, CRM events, and call bookings, then credits the original ad source when the close lands forty-five days later. On a Shopify checkout one click from the ad, most of that machinery is overkill. Pricing is custom, from the low-to-mid hundreds per month into the thousands at scale. Law firms read attribution through [a different lens](/for-service-brands/law-firms/), because the close lands weeks after the click and the closed-case feedback loop is what makes Hyros worth the line item. | Platform | Function | Price tier | Right at spend tier | |---|---|---|---| | Google Ads + GA4 + Shopify | Native attribution stack | Free | Under $50K/mo | | Triple Whale | Pixel attribution + survey | Paid, check current pricing | $50K to $200K/mo | | Northbeam | First-party MTA model | Paid, check current pricing | $50K to $500K/mo | | Polar Analytics | Dashboard layer + light attribution | Paid, check current pricing | $30K to $150K/mo | | Hyros | Long sales cycle attribution | Low hundreds to thousands/mo | Lead-gen, info products, high-ticket | ### Custom diagnostic workbooks This is the category most founders skip. A workbook does what a SaaS tool can't: it forces the operator to type the numbers in, which is the moment most leaks become obvious. The [free 25-page Google Ads Setup Checklist](/audit/) is the workbook I give away: ten settings and tracking red flags, each with the menu path to fix it, plus a four-week clean-up plan. A founder running it line by line surfaces most of what a one-thousand-dollar agency audit would, and owns the diagnosis at the end. It doesn't replace a rule engine running across dozens of accounts. For one account, it covers most of the same ground. The same applies on Meta. In my [Q3 2026 Meta benchmark](/benchmarks/meta-ads-cpm-by-placement/), 9 of 18 ad accounts were optimizing for no conversion event at all, and they carried 42% of the spend (N = 18). No paid tool is needed to see that. Ads Manager shows the optimization goal on every ad set, and a spreadsheet of goal against spend catches it in one row. ### Matching the stack to the spend, then reading it weekly My rule of thumb: under fifty thousand a month with one or two channels, native reports plus the GA4 cross-check are enough, and paid SaaS is a distraction because the bottleneck is creative and offer. Above that, with three or more channels, the post-purchase survey is the reason to pay for Triple Whale or Northbeam, because it shows which channel takes more credit than it earns. Where the click and the revenue are separated by weeks, Hyros is right regardless of spend. The stack on [furniture and home accounts](/for-home-brands/furniture/) holds the same shape against a longer purchase window. No tool replaces the human read. Open the account once a week. Read the search terms report. Read the top five ad creatives against the landing pages they point at. Confirm the conversion counts in Google Ads, GA4 and Shopify sit inside the bands above. Twenty minutes a week, done with attention, beats any dashboard nobody opens. Send the account read-only through the [contact form](/contact/) if you want the platform-fit decision run against live data. The [services overview](/services/) covers how the Google and Microsoft audit passes land inside a paid build, [my pricing page](/pricing/) covers the engagement levels, and [my process page](/process/) covers the walk itself. [The wasted ad spend hub](/wasted-ad-spend/) covers the other patterns in the same shape. --- # Blog ## Do Directory Profiles Help AI Search? Mine Backfired First URL: https://connercrowe.com/blog/directory-profiles-ai-search-backfired/ Published: 2026-10-30 I set up directory profiles so AI assistants could check me. Claude started citing them instead of my site, then used them to say I have no reviews. ### Do directory profiles help AI search? They get read, which can help or hurt. In October I added directory profiles so AI assistants would have somewhere besides my own site to check me. By my 30-day re-test, engines were citing them. The profiles had no reviews on them yet, and Claude used that to say I have no reviews. ### What happened when I added the profiles? | Claude, 17 questions about me | Sept 27 (before profiles) | Oct 28 (30 days) | | --- | --- | --- | | Answers citing my Clutch or DesignRush profile | 0 | 12 of 17 | | Answers citing my own site | 12 of 17 | 0 of 17 | | Answers saying I have no reviews or no Google profile | 6 of 17 | 9 of 17 (too small a change to tell from luck) | On September 27, 2026, before the profiles, I asked AI assistants a locked set of questions about me and my business. Claude cited my own site in 12 of 17 answers about me. A Clutch search turned up nothing for me, and a BBB search found no match. My DesignRush profile went live on October 5, and my Clutch and BBB profiles on October 14. None of them had reviews yet. On October 28, 2026 I asked the same questions again: - Claude cited my Clutch or DesignRush profile in 12 of 17 answers about me, up from 0. It cited my own site in 0 of 17, down from 12. - Outside Claude, my Clutch profile also showed up as a source in logged-out ChatGPT, Copilot and Perplexity answers. So the profiles worked as sources. Engines were citing them two to three weeks after they went live. ### Why did Claude say I have no reviews? Mostly because the profiles it was reading had none. At baseline, Claude already said I had no third-party reviews and no Google Business Profile, in 6 of 17 answers about me. I have a Google profile. It had a 5.0 rating from 9 reviews then and 11 by October 28. I expected outside profiles to fix that. Instead, my new Clutch profile and DesignRush page gave Claude fresh evidence: business pages with zero reviews on them. At 30 days, 9 of 17 Claude answers about me said I had no reviews or no Google profile. That's up from 6 of 17, too small a change to tell from luck. In 7 of those 9 answers Claude had read my Clutch or DesignRush page. In the other 2 it found nothing about my reviews at all. One logged-out Copilot answer said I had no third-party reviews too, and Google's AI Mode said independent reviews of me are sparse. Other engines did report my 5.0 rating: Gemini, Copilot, Perplexity and Google's AI Mode. On questions about me, they mostly got it from the reviews page on my own site, which went live on October 27. At 30 days Claude cited none of my pages, so it worked from directory pages that show zero reviews. Losing Claude's citations to my site is what cost me the rating in its answers. Claude ran signed in to my own account on both dates, and at 30 days many of its answers drew on that account instead of searching. The drop in its citations to my site is too big to be luck. Read the rest of the Claude numbers as a direction. ### Should you set up directory profiles for AI search? Yes, but in a different order than I did: 1. **Line up the first reviews before the profile goes live.** Ask your recent clients for an honest review, whatever they thought, and send the link the day the profile publishes. Never pay, discount or trade anything for a review, and don't only ask the clients you expect to be happy. 2. **Fill in every field.** Engines lift services, location and pricing straight from these pages, and fill any blank from another source, which may be out of date. 3. **Tell one location story.** Use the same business name and contact details everywhere, and make sure your site names the city your listings show. Mine show Spokane Valley, where I work, while my site only said Spokane, so I updated my site to say both. 4. **Keep a reviews page on your own site.** It's where the engines that got my rating right found it. 5. **Re-check what AI says about you a few weeks after launch.** That's how I caught this. My own next step is the first one, starting with Clutch. Across my test questions engines cite it more than any other profile, and The Manifest pulls from it. For the full before-and-after on everything I changed, see [my 30-day re-test](/blog/how-long-does-ai-search-optimization-take-30-day-retest/). If you want to know what AI assistants say about your business right now, my [AI answer audit](/ai-answer-audit/) page has a free snapshot, or you can run a few questions yourself with the free [What AI says about you](/what-ai-says-about-you/) template. ## How Long Does AI Search Optimization Take? I Checked at 30 Days URL: https://connercrowe.com/blog/how-long-does-ai-search-optimization-take-30-day-retest/ Published: 2026-10-30 I re-ran my own AI-visibility audit 30 days after fixing it. Gemini's factual errors about me fell from 7 of 17 answers to 0. Buyer questions haven't moved. ### How long does AI search optimization take? In my own re-test, one engine stopped getting my facts wrong inside 30 days. Being named when someone asks who to hire hasn't moved yet. Gemini's factual errors about me went from 7 of 17 answers to 0 of 17, and Gemini started citing my own site on 16 of 17 questions about me, up from 4. On buyer questions, the change was too small to tell from luck. I audit what AI assistants say about businesses for a living, so I ran the same audit on myself. On September 27, 2026 I asked a locked panel of 53 questions across ChatGPT, Claude, Gemini, Perplexity, Microsoft Copilot and Google's AI Mode and AI Overviews. I made most of the fixes between October 3 and October 8, and added a few more through October 27. On October 28 I asked the same 53 questions again. ### What counted as a real change? On October 4, after the baseline and before the re-test ran, I wrote down the rules for this re-test: - A change only counts if it's too big to be luck. In practice: the 95% confidence ranges before and after can't overlap. - At 30 days, the number that matters is how often answers about me get a fact wrong. - Being named on buyer questions is a 90-day number. A 30-day move gets reported, never called a win. ### The numbers | What I measured | September 27, 2026 | October 28, 2026 | Verdict | |---|---|---|---| | Gemini answers about me with a factual error | 7 of 17 (41%, 95% CI 21.6% to 64.0%) | 0 of 17 (0%, 0% to 18.4%) | Real change | | Gemini answers about me citing my site | 4 of 17 (24%, 9.6% to 47.3%) | 16 of 17 (94%, 73.0% to 99.0%) | Real change | | Claude answers about me citing my site | 12 of 17 (71%, 46.9% to 86.7%) | 0 of 17 (0%, 0% to 18.4%) | Real change, the wrong way | | All answers about me with a factual error | 15 of 45 (33%, 21.4% to 47.9%) | 12 of 86 (14%, 8.2% to 22.8%) | Not yet | | Buyer questions naming me, same question slots | 3 of 204 (1.5%, 0.5% to 4.2%) | 7 of 204 (3.4%, 1.7% to 6.9%) | Not yet | The 45 and 86 differ because Copilot and Perplexity were scored this time, and ChatGPT and AI Mode hit usage limits at baseline, so fewer of their answers were scored then. The 204 compares only the slots that ran on both dates. My [baseline post](/blog/six-ai-search-audits-crawler-access/) reported 2 of 203 across every buyer question, which is a slightly different cut of the same run. ### What changed in AI answers within 30 days? Gemini did. At baseline it got something wrong about me in 7 of 17 answers. At 30 days it was 0 of 17, and it cited my own pages in 16 of 17 answers. My fixes in that window were on my own site and on directory profiles: correcting prices, stating what I offer in plain words at the top of the main pages, and adding the profiles engines could check me against. Old bio pages also dropped out. Three outdated third-party bios were cited 21 times at baseline, 20 of them by Claude. At 30 days they were cited zero times. Only one of the three was a removal I asked for: one page was already gone, and another turned out to belong to someone else with my name. On questions about me, Claude now cites my new DesignRush and Clutch profiles in 12 of 17 answers. It also stopped citing my own site, and those new profiles had a cost, which I get to below. ### What didn't change in 30 days? **Buyer questions.** 3 of 204 to 7 of 204 looks like progress, and the ranges overlap. Three of those seven came from Claude answering from my own signed-in account, so the clean count is closer to 4. I'll know more at 60 and 90 days. **Overall accuracy.** Errors across all engines fell from 15 of 45 answers to 12 of 86. The ranges overlap, so I'm not calling it. Nine of the 12 were Claude. The other three were ChatGPT quoting a price I took off my site on October 8, 2026. **New claims are slow.** I put a free audit with a 24-hour turnaround on my site on October 3, 2026. On October 28, 1 of 85 answers about me mentioned both. **Copilot cited my pages less.** Bing's 30-day count of Copilot citations for my site fell from 699 to 373. I had set holding steady as a 30-day target, and I missed it. That count is slippery, and I wrote about [how I read it](/blog/copilot-citations-drop-wrong-baseline/). ### Can new directory listings make AI answers worse? They can for a while. At baseline, Claude said I had no Google Business Profile. I have one, with a 5.0 rating from 11 reviews. I expected new directory profiles to help. They were brand new with no reviews on them, so they gave engines fresh evidence that I have none. At 30 days Claude still said it in 9 of 17 answers about me, one logged-out Copilot answer said it too, and Google's AI Mode said independent reviews of me are sparse. In my Spokane test questions, Claude's search never surfaced my Google profile. If you set up new listings so AI engines can check you, have a plan for their first reviews before they go live. ### What changed between the two runs besides my fixes? The questions and the scoring rules stayed the same. How some answers were collected did not: - Google answers came from a regular browser at baseline and from a search tool set to Spokane at 30 days. - About a third of ChatGPT answers came through a similar third-party tool from a U.S. server, not set to Spokane. - Copilot and Perplexity had no usable baseline, so they appear only in the 30-day totals. - Claude ran signed in to my own account, and its answers often drew on that account instead of searching. I flag every Claude number above for that reason. The Gemini result is the cleanest one here: same engine, same questions, same collection both times. ### What I'm changing - At 30 days I measure accuracy first, and I don't expect buyer questions to move. - I check each engine on its own. Gemini reflected my fixes within four weeks. ChatGPT still quoted an old price 20 days after I removed it. - Every re-test now locks how answers are collected and graded (logged in or out, location, tool, grading rules), not only the questions. The 60-day run is set for November 26, 2026. I'll post it whether it moved or not. ### Want to know what AI says about your business? If ChatGPT is quoting an old price or telling people you have no reviews, customers hear it before they reach your site. My [AI answer audit](/ai-answer-audit/) page has a free snapshot: I run the questions your buyers ask and send what the engines say within 24 hours. To run a few yourself first, use the free [What AI says about you](/what-ai-says-about-you/) template. For how often AI answers name a business at all, see my [AI answers field report](/benchmarks/ai-answers-field-report/). ## Impressions Up 22%, Clicks Up Three (Why I Audit My Own Reports First) URL: https://connercrowe.com/blog/impressions-up-clicks-up-three-auditing-my-own-report/ Published: 2026-10-30 My search report: impressions up 22%, average position nearly halved, clicks from 18 to 21. What I reported, and the four claims an audit caught. About six weeks after I rewrote a batch of pages on this site, I pulled a month of Search Console numbers to see whether the work had paid off. Impressions were up 22%. Average position went from 16.5 to 8.7. That's the kind of slide that opens an agency's monthly report. Clicks went from 18 to 21. ### Quick Take My own recheck got every number in the table right: 2,179 impressions became 2,669, position nearly halved, and click-through held at 0.8%. The first draft of the summary still had four high-severity errors, plus smaller ones. A second reviewer pulled the data again, checked the PDFs itself, and caught all four. ### What the numbers said Both windows are 28 days, with the day the changes shipped left out. | Window | Clicks | Impressions | CTR | Avg position | |---|---|---|---|---| | Aug 23 to Sep 19 | 18 | 2,179 | 0.8% | 16.5 | | Sep 21 to Oct 18 | 21 | 2,669 | 0.8% | 8.7 | More people saw the site, Google ranked it higher, and the share who clicked didn't move. Twenty-one clicks is too thin to call a result. Only 4 of them came from a search I could see as my name. The other 17 show up at page level only, behind queries Search Console doesn't report. The write-up calls it a direction. More pages changed on October 21 and 22, after the window closed. They show up only in an overlapping October 1 to 28 pull, and a re-measure already set for November 20 will read them cleanly. ### The page that ranks and doesn't get clicked The biggest winner shows the gap best. One post, about how [Google's AI Max matches search terms](/blog/google-ads-search-term-matching-ai-max/), went from 158 impressions at position 33.3 to 575 at position 5.8. On one family of searches it now sits at position 2 or 3: people pasting rows straight out of a Google Ads export, like "total: ai max expanded matches". Before a decoder section went into the post, those searches found it somewhere between position 40 and 100. It got 1 click in each window. Google's report on its AI search features shows the same post at 100 impressions, up from 5. That report counts impressions. It doesn't say the page was quoted, and none of it is traffic. My guess is that an AI Overview answers the pasted row on the results page, so nobody needs to click through. That's an untested hypothesis. That page is on my fix list: a one-line hand-off in the section people land on, so anything that quotes the page carries the next step with it. ### The four sentences the audit corrected The recheck went to a second reviewer, an AI agent that pulled Search Console again with its own script, over the same windows, before it saw the draft. Then it checked the draft against its own pull. Round one, every site total and every page and query number matched the draft exactly. It still came back with four high-severity findings: 1. The draft said almost all the clicks were brand. Four of 21 visibly were. 2. It said six of nine assistant-style searches were math word problems landing on my calculator, like "a car dealership has $5,601 to spend on newspaper ads". Three of nine were. 3. It said one page was still sliding. It had dropped from position 11.2 to 40.1, and in a later, overlapping window (October 1 to 28) it was back up to 33.3. 4. It said an old set of map coordinates was still inside my free PDFs. Text extraction found them in neither file. Round two found one medium issue: the draft said its list of winners covered every page that moved, and the list missed three. Round three came back clean. ### Why the sentences go wrong The table came straight out of the export, so it matched. The counts and claims in the sentences had never been checked against a fresh pull or against the files themselves. An agency writing its own monthly update has the same gap, because the people who write the sentences also check them. The check I trust is a reviewer who didn't write the claims and who pulls the data again from the source instead of reading the author's table. I run [the same kind of adversarial check on my case studies](/blog/44-agents-fact-checked-my-case-studies/), and on that run it caught three high-severity problems in work I'd signed off on. ### What skipping the check costs If the first draft had gone out, it would have called 21 clicks mostly brand, called a recovering page still sliding, and pointed at a PDF fix for a problem that didn't exist. ### How to check your own monthly report Five questions I'd ask of any report before spending money on it: - What are the clicks next to the impressions, on every page the report calls a win? - What share of the clicks came from your brand name? - Did anything else change inside the measurement window? - Who checked the summary, and did they pull the data themselves? - Is the conversion count underneath it trustworthy? [The Tracking Stack](/frameworks/tracking-stack/) is how I check that one. If you want that read on your Google Ads account, my [free audit](/audit-request/) is written by me and comes back within 24 hours of read-only access. ## How to Check If AI Crawlers Can Reach Your Site URL: https://connercrowe.com/blog/how-to-check-if-ai-crawlers-can-reach-your-site/ Published: 2026-10-28 Read your live robots.txt, check what your CDN adds, fetch your pages as each AI bot, then read your logs. The four checks I run first in every audit. export const faqs = [ { question: 'What is an AI crawler?', answer: 'A bot an AI company runs to fetch web pages. Some collect pages for model training (GPTBot, ClaudeBot), some build the index an assistant searches when it answers (OAI-SearchBot, Claude-SearchBot, PerplexityBot), and some fetch a page because a user asked about it (ChatGPT-User, Claude-User, Perplexity-User). Each one is controlled by its own robots.txt token.', }, { question: 'Which AI crawlers should a small business allow?', answer: 'At minimum, the search crawlers: OAI-SearchBot for ChatGPT search, Claude-SearchBot for Claude, PerplexityBot for Perplexity, and Googlebot for Google Search, AI Overviews and AI Mode. OpenAI says a site that blocks OAI-SearchBot will not be shown in ChatGPT search answers. Training crawlers are a separate decision. Google, Apple and OpenAI say blocking their training tokens doesn\'t remove you from their search. Meta describes its training crawler as also indexing content for its products.', }, { question: 'Does blocking GPTBot hide my site from ChatGPT?', answer: 'Not from ChatGPT search, according to OpenAI. OpenAI says its GPTBot and OAI-SearchBot settings are independent: you can disallow GPTBot to opt out of training and still allow OAI-SearchBot to appear in search answers.', }, { question: 'How do I block AI bots in robots.txt?', answer: 'Add a group for each bot by its token, for example "User-agent: GPTBot" on one line and "Disallow: /" on the next. Robots.txt is a request. OpenAI says robots.txt may not apply to ChatGPT-User and Perplexity says Perplexity-User generally ignores it, so enforcing a block on those takes a firewall rule using the IP ranges they publish.', }, { question: 'Does Cloudflare block AI crawlers by default?', answer: 'It depends on when the domain was added and what is switched on. Cloudflare announced in July 2025 that it was changing its default to block AI crawlers. Its documentation says new domains from September 15, 2026 get defaults that block training and agent bots on pages with ads while search crawlers stay allowed, and that crawlers combining search and training are affected by the training block. Check the AI and bot settings on your own zone instead of assuming either way.', }, { question: 'How long does a robots.txt change take to work?', answer: 'The robots.txt standard (RFC 9309) says crawlers should not use a cached copy for more than 24 hours. OpenAI says its search systems can take about 24 hours to adjust after a robots.txt update, and Meta says its crawlers may cache the file for up to 24 hours.', }, ]; ### Quick take If an AI assistant can't fetch your pages, it can't read or quote them. The check takes four steps: read the robots.txt your site serves today, look for rules your CDN or platform adds on top of it, fetch your pages with each bot's user agent, and look for the bots in your logs. It's the first thing I check in every [AI answer audit](/ai-answer-audit/), and it's usually the shortest part of the report. Here is the path a bot takes to your page, and what the checker tests at each stop: If you want the fast version, the [AI crawler checker](/tools/ai-crawler-checker/) reads your robots.txt for the bots in my [AI crawler directory](/tools/ai-crawlers/). This post is the manual method, including the parts a robots.txt reader can't see. ### Know which bots you're checking for "AI crawler" covers three kinds of bot, and they don't carry the same stakes. **Search crawlers** build the index an assistant searches when it answers: [OAI-SearchBot](/tools/ai-crawlers/oai-searchbot/), [Claude-SearchBot](/tools/ai-crawlers/claude-searchbot/) and [PerplexityBot](/tools/ai-crawlers/perplexitybot/). Block one and you can drop out of that assistant's search answers. [OpenAI says so directly](https://developers.openai.com/api/docs/bots) for OAI-SearchBot. Google's AI Overviews and AI Mode don't have their own bot. Google says a page only needs to be indexed and eligible to show with a snippet, [with no additional technical requirements](https://developers.google.com/search/docs/appearance/ai-features), so Googlebot is the one that counts there. **User-triggered agents** fetch a page because a person asked the assistant to: [ChatGPT-User](/tools/ai-crawlers/chatgpt-user/), [Claude-User](/tools/ai-crawlers/claude-user/) and [Perplexity-User](/tools/ai-crawlers/perplexity-user/). **Training crawlers and tokens** decide whether your pages feed model training: [GPTBot](/tools/ai-crawlers/gptbot/), [ClaudeBot](/tools/ai-crawlers/claudebot/), [Google-Extended](/tools/ai-crawlers/google-extended/), [CCBot](/tools/ai-crawlers/ccbot/), [Applebot-Extended](/tools/ai-crawlers/applebot-extended/) and [meta-externalagent](/tools/ai-crawlers/meta-externalagent/). Google, Apple and OpenAI say blocking their training tokens doesn't remove you from their search, and Meta says its crawler also indexes content for its products. [Google says Google-Extended](https://developers.google.com/search/docs/crawling-indexing/google-common-crawlers) doesn't affect inclusion in Google Search, and [Apple says](https://support.apple.com/en-us/119829) pages that disallow Applebot-Extended can still show up in Apple's search results. For a local business or a store, the search crawlers are the ones to protect. The training bots are a judgment call. ### Step 1: read the robots.txt your site serves today Open `yoursite.com/robots.txt` in a browser. Read the live URL, not the file in your theme or repo. Step 2 explains why they can differ. Robots.txt is made of groups. Each group starts with one or more `User-agent` lines and lists `Allow` and `Disallow` rules under them. The standard, [RFC 9309](https://www.rfc-editor.org/rfc/rfc9309.html), sets how a bot picks its group. Under it, a bot obeys the group that names it and falls back to the `User-agent: *` group only when no group names it. Say your file has this: ``` User-agent: * Disallow: /checkout Disallow: /account User-agent: GPTBot Disallow: /blog ``` GPTBot now follows only its own group. It's allowed into `/checkout` and `/account`, because the `*` rules no longer apply to it. Every time you add a group for one bot, copy in the general rules you still want it to follow. What to look for: 1. A `User-agent: *` group with `Disallow: /`. That blocks every bot without its own group, the search crawlers included. 2. A group naming OAI-SearchBot, Claude-SearchBot, PerplexityBot or Googlebot with a broad `Disallow`. 3. A block of rules you didn't write, added by a plugin, an app or an old template. Read every bot name in it. 4. Googlebot rules, if you care about Apple. [Apple says](https://support.apple.com/en-us/119829) that if your file doesn't mention Applebot but does mention Googlebot, Applebot follows the Googlebot rules. Also check the status code. Under RFC 9309, a robots.txt that returns a 4xx error means bots may crawl everything. One that returns a 5xx error, or doesn't respond, means they must assume they can crawl nothing. A flaky server can block every bot without a single rule in the file. ### Step 2: check what your platform and CDN add The file you wrote isn't always the file that gets served. On Shopify, every store [starts with a default robots.txt](https://help.shopify.com/en/manual/promoting-marketing/seo/editing-robots-txt), and you change it by adding a `robots.txt.liquid` template. Shopify warns that a wrong edit can cost you all your traffic, so read the live URL after any change. On Cloudflare, [managed robots.txt](https://developers.cloudflare.com/bots/additional-configurations/managed-robots-txt/) can prepend Disallow rules for AI crawlers to your existing file when you turn it on. The example list in Cloudflare's docs covers training bots such as GPTBot, ClaudeBot, Google-Extended and CCBot. Then there's the firewall, which matters more. Robots.txt asks a bot to stay out. A firewall rule turns it away, and it never shows up in your robots.txt. Cloudflare [announced in July 2025](https://blog.cloudflare.com/content-independence-day-no-ai-crawl-without-compensation/) that it was changing its default to block AI crawlers. Its [current docs](https://developers.cloudflare.com/bots/additional-configurations/block-ai-bots/) say new domains from September 15, 2026 get defaults that block training and agent bots on pages with ads, while search crawlers stay allowed. Cloudflare's [changelog adds a catch](https://developers.cloudflare.com/changelog/post/2026-07-01-ai-traffic-options/): "multi-purpose crawlers that combine Search and Training will be affected by the new defaults to block Training." If a bot you care about does both, check it by name. In [AI Crawl Control](https://developers.cloudflare.com/ai-crawl-control/features/manage-ai-crawlers/), blocking a crawler creates a WAF rule on your zone. So log in and look. In Cloudflare, check AI Crawl Control and the bot settings. On any other CDN or security plugin, search the settings for "AI", "bot" and "crawler". In June my own Cloudflare firewall was returning 403 errors to five AI crawlers, and nothing in my analytics showed it. I wrote that one up in [six AI search audits](/blog/six-ai-search-audits-crawler-access/). ### Step 3: fetch your pages as each bot Now ask your server directly. This `curl` command requests your home page with OAI-SearchBot's published user agent and prints only the status code: ``` curl -s -o /dev/null -w "%{http_code}\n" -A "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36; compatible; OAI-SearchBot/1.4; +https://openai.com/searchbot" https://yoursite.com/ ``` On Windows, run it in PowerShell as `curl.exe` and change `/dev/null` to `NUL`. Swap in the user agent for each bot you care about. The directory lists the full string for every bot whose owner publishes one. Run it on your home page and two or three pages you want quoted, like a service page and a product page. A `200` is what you want. A `403`, `429` or `503` means something turned the bot away. A `200` can still be a challenge page, so drop `-o /dev/null` once and check that the HTML is your page. This test has one limit. The request comes from your IP address, not the bot's. A firewall rule that checks verified bot IP ranges, or blocks data-center traffic, can treat the real bot differently from your test. In my [AI answers field report](/benchmarks/ai-answers-field-report/), all 530 fetches across 9 business sites came back 200, and they came from a residential IP, which is why I report them with this caveat. A clean `curl` result rules out rules keyed to the user agent. Step 4 covers the rest. ### Step 4: look for the real bots in your logs The last check is whether the bots show up. Search your server or CDN logs for the token, like `OAI-SearchBot` or `PerplexityBot`, and look at the status codes they got. In Cloudflare, AI Crawl Control shows which AI crawlers visited and how often. Anyone can fake a user agent, so treat the token as a claim. OpenAI, Perplexity, Anthropic, Common Crawl and Apple publish IP lists or reverse DNS rules you can check a request against. Each bot's page in the directory links to its list. If a search crawler never shows up, that doesn't prove a block. A small site may not get crawled often. If it shows up and gets `403`s, you've found your problem. ### User-triggered agents don't follow the same rules Don't count on robots.txt to control the user agents. [OpenAI says](https://developers.openai.com/api/docs/bots) robots.txt rules may not apply to ChatGPT-User, because a user started the fetch. [Perplexity says](https://docs.perplexity.ai/docs/resources/perplexity-crawlers) Perplexity-User generally ignores them. [Anthropic says](https://support.claude.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler) its bots, Claude-User included, honor robots.txt. For a local business or store, the better move is to let these through. They're how an assistant reads your page when a customer asks about you. ### What a clean result does and doesn't tell you Passing all four steps means the assistants can read you. It doesn't mean they'll name you. In the field report, every crawler fetch got through, and businesses still went unnamed on most questions that didn't include their name. The ones that got named were the ones other sites talked about. So run this check, fix anything it finds, and move on. The free [AI Visibility Check](/tools/ai-visibility-check/) covers access along with the other things a tool can read from outside your site. The next step is finding out what the answers say, and I wrote up [how I do that for local businesses](/blog/how-to-show-up-in-chatgpt-and-ai-overviews-local-business/). To have me look, start with the free AI Answer Snapshot on my [AI answer audit page](/ai-answer-audit/). ### Frequently asked questions ## How to Show Up in ChatGPT and AI Overviews When You're a Local Business URL: https://connercrowe.com/blog/how-to-show-up-in-chatgpt-and-ai-overviews-local-business/ Published: 2026-10-28 The method I use to help local businesses get named in ChatGPT and Google's AI Overviews: let the right bots in, fix your facts, test the questions, get cited. export const faqs = [ { question: 'How do I get my business recommended by ChatGPT?', answer: 'Make sure OpenAI\'s search crawler, OAI-SearchBot, can reach your site, since OpenAI says sites that block it are left out of ChatGPT search answers. Then make your business facts match everywhere they appear, test the questions your buyers ask, and get listed or mentioned on the directories, review sites and local articles those answers cite. Getting named on other sites takes the longest.', }, { question: 'Do I need special markup or an llms.txt file to show up in AI Overviews?', answer: 'No. Google says there are no additional requirements to appear in AI Overviews or AI Mode, and that you don\'t need new machine readable files, AI text files, markup or special schema. A page needs to be indexed and eligible to show in Google Search with a snippet.', }, { question: 'Will blocking AI training bots hurt my visibility in ChatGPT or AI Overviews?', answer: 'According to their owners, no. OpenAI says GPTBot and OAI-SearchBot are set independently, so blocking GPTBot does not remove you from ChatGPT search. Google says Google-Extended does not affect inclusion in Google Search, and its AI Overviews guidance points to Googlebot and the normal snippet controls.', }, { question: 'Does my Google Business Profile matter for AI answers?', answer: 'Keep it accurate either way. Google says its local results weigh relevance, distance and prominence, and that prominence includes how many websites link to you and how many reviews you have. In my field report, google.com and facebook.com showed up in the cited sources for nearly every business I checked. A wrong fact on a profile is one more version for the assistant to pick from.', }, { question: 'How long until changes show up in ChatGPT?', answer: 'For crawler access, OpenAI says a robots.txt change can take about 24 hours to reach its search systems. For everything else, nobody publishes a timeline. I re-run the same questions a month after a fix, asked the same way, and compare.', }, { question: 'How do I check if ChatGPT mentions my business?', answer: 'Write ten questions a buyer would ask, most of them without your business name. Ask each one in ChatGPT in a private, logged-out window, three times, because answers change from run to run. Log who gets named, which facts are wrong and which sites are cited.', }, ]; ### Quick take When someone asks ChatGPT or Google's AI who to hire nearby, the answer leans on two things: pages the assistant can read, and what other sites say about you. Most local businesses I check pass the first and come up short on the second. Here is roughly how that answer gets built: The order I work in is the same every time. Let the right bots in. Make your facts agree everywhere. Ask the questions your buyers ask and write down what comes back. Then get named on the sites those answers already cite. This post walks through each step, the way I run it in my [AI answer audit](/ai-answer-audit/), at a size you can do yourself. ### Ranking in ChatGPT isn't one thing "Showing up" covers two very different questions. The first is a buyer who types your name. In my [AI answers field report](/benchmarks/ai-answers-field-report/), assistants named the business 90.2% of the time when the buyer asked about it by name (751 of 833 answers). That part mostly works. The second is a buyer who describes a problem and names nobody. "Who can fix a leaking water heater in Spokane this week?" On questions like that, the business was named 2.6% of the time (16 of 623 answers). That's where new customers come from, and it's the number that stays low. Everything below is aimed at the second number. ### Step 1: let the right bots in OpenAI uses its search crawler, [OAI-SearchBot](/tools/ai-crawlers/oai-searchbot/), to surface websites in ChatGPT search. [OpenAI says](https://developers.openai.com/api/docs/bots) sites that opt out of it won't be shown in ChatGPT search answers. Google's AI Overviews and AI Mode don't have a separate crawler. [Google says](https://developers.google.com/search/docs/appearance/ai-features) a page needs to be indexed and eligible to show in Search with a snippet, and that there are no additional technical requirements. If you show up in normal Google results with a description under your link, you're eligible. Blocking the training bots doesn't change either of those. OpenAI sets GPTBot and OAI-SearchBot independently. [Google says Google-Extended](https://developers.google.com/search/docs/crawling-indexing/google-common-crawlers) doesn't affect inclusion in Google Search. So the check is short. Is OAI-SearchBot allowed in your robots.txt? Is Googlebot? Is a firewall or a CDN setting turning either away? The [AI crawler checker](/tools/ai-crawler-checker/) reads your robots.txt for the bots in my [AI crawler directory](/tools/ai-crawlers/), and I wrote up [the full manual check](/blog/how-to-check-if-ai-crawlers-can-reach-your-site/) for the parts a robots.txt reader can't see. For most businesses this comes back clean. In the field report, all 530 crawler fetches across 9 sites returned a 200. Run it anyway. When it isn't clean, fix it first. OpenAI says sites that opt out of OAI-SearchBot won't be shown in ChatGPT search answers. ### Step 2: make your facts agree everywhere An assistant that finds two sets of hours has to pick one. In the shorter snapshot checks behind my field report, 4 of 4 businesses had facts that disagreed between their own website and their listings, at least 6 facts each. The method is a spreadsheet. 1. **Rows:** business name, address, phone, hours, service area, services, prices if you publish them, license number, year founded. 2. **Columns:** your website, your Google Business Profile, Yelp, Facebook, Bing Places, Apple Business Connect, and every industry directory you're listed on. 3. **Fill every cell.** Copy what each one says today, word for word. 4. **Fix the source.** Every row that disagrees is a fix. Correct each listing that's wrong, then your site. My free [check log and ledger](/what-ai-says-about-you/) has this sheet built, along with the questions in step 3. The Business Profile deserves extra care. [Google says](https://support.google.com/business/answer/7091) its local results weigh relevance, distance and prominence, and that prominence includes how many websites link to you and how many reviews you have. That's Google's local pack, not the AI answer. But in my field report, the domains cited in answers for the most businesses I checked were facebook.com, google.com, linkedin.com, reddit.com, bbb.org, yelp.com, youtube.com. Your profile and your listings are some of the pages being read. ### Step 3: ask the questions your buyers ask Write ten questions. A few should include your name: "Is [business] any good?" or "What are [business]'s hours?" Most shouldn't. "Who's the best [service] in [city]?" "Who can fix [problem] near [neighborhood] this week?" "How much does [service] cost in [city]?" Write at least one the way a customer would describe the problem to a friend, with no trade words. Then ask them. 1. **Use a clean session.** A private window, logged out where the assistant allows it. Your own chat history can leak into the answer. 2. **Ask each question three times.** The same question gets different answers from run to run, and one answer proves very little. 3. **Check ChatGPT and Google first.** On Google, AI Overviews appear on a normal search, and AI Mode is its own tab. 4. **Log three things per run:** who got named, which facts about you were wrong, and which sites the answer cited. This is the core of my audit at a smaller size. The full version, which is the first thing I ship in paid work, runs {SNAPSHOT_FACTS.fullQuestions} questions on {SNAPSHOT_FACTS.fullEngines} assistants, {SNAPSHOT_FACTS.fullRuns} runs each, with the question list locked before the first run so a re-test compares like with like. Ten questions on two assistants will still show you the pattern. ### Step 4: get named where the answers already look Your log from step 3 has a list of sites in it. Those are the sources the assistant trusts for your kind of business in your town: a directory, a review site, a local "best of" article, a forum thread, a trade association listing. That list is your work plan. Go down it in order of how often each site was cited. 1. **Listings you can claim:** claim them, fill every field, and match the facts from step 2. 2. **Review sites:** ask happy customers to review you on the platforms that keep showing up. In my snapshot checks, 4 of 4 businesses had fewer Google reviews than the median competitor the answers named instead. That's a small sample and it doesn't prove reviews cause naming, but it lines up with what the answers cite. 3. **Articles and roundups:** contact the writer with something worth adding, like a local price guide or a fact they got wrong. 4. **Local organizations:** chambers, trade groups and sponsors that list their members. This step is slow. It's also the one that moved naming in my audits. When I audited my own site, the engines could read every page and still named me on almost no hiring questions. Directories and roundups held the spots, and I had no presence on them. I wrote that up in [six AI search audits](/blog/six-ai-search-audits-crawler-access/). ### Step 5: write the page the answer wants to quote For each service you want to be named for, have one page that answers the buyer's question in its first two sentences: what the service is, where you do it, what drives the price, and how soon you can start. Put a date and your name on it. Keep the facts identical to your listings. In one audit I ran, the pages that won the comparison questions all shared a format: a verdict at the top, a comparison table, a recent date and a named author. The business had no page in that format. Skip the shortcuts. Google says you don't need [AI text files, markup or special schema](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) to appear in its AI features. An [llms.txt file](/blog/what-llms-txt-does-and-whether-you-need-one/) won't get you named on its own either. ### Where to start today Run the free [AI Visibility Check](/tools/ai-visibility-check/) on your site for a quick read of what a tool can see from the outside. Then do steps 2 and 3 by hand this week. A tool can't tell you what the answers say about you. To have me look first, start with the free AI Answer Snapshot on my [AI answer audit page](/ai-answer-audit/): {SNAPSHOT_FACTS.questions} questions on {SNAPSHOT_FACTS.engines}, back {SNAPSHOT_FACTS.duration}. ### Frequently asked questions ## What llms.txt Does, and Whether Your Business Needs One URL: https://connercrowe.com/blog/what-llms-txt-does-and-whether-you-need-one/ Published: 2026-10-28 llms.txt is a proposed file that gives AI tools a map of your site. No AI search crawler documents reading it. When it's worth writing, and how I write one. export const faqs = [ { question: 'What is llms.txt?', answer: 'A markdown file at the root of a website, at /llms.txt, that gives AI tools a short summary of the site and a hand-picked list of links to its most useful pages. Jeremy Howard proposed it on September 3, 2024. The only required part is an H1 with the site or project name.', }, { question: 'Is llms.txt an official standard?', answer: 'No. The specification at llmstxt.org calls itself a proposal and is open for community input. No standards body has adopted it, and none of the AI crawler owners I track has said its crawler reads it.', }, { question: 'Does ChatGPT read llms.txt?', answer: 'OpenAI\'s documentation for its crawlers, GPTBot, OAI-SearchBot and ChatGPT-User, doesn\'t say any of them reads llms.txt. OpenAI publishes llms.txt files for its own developer docs, but it has not said ChatGPT search uses llms.txt to decide what to show or cite.', }, { question: 'Does llms.txt help with Google AI Overviews?', answer: 'No. Google\'s guide to its generative AI features names llms.txt and says Google Search ignores AI text files: creating one will neither harm nor help your visibility in Search. A page only needs to be indexed and eligible to show in Search with a snippet.', }, { question: 'What is the difference between llms.txt and robots.txt?', answer: 'Robots.txt controls access: it tells crawlers what they may fetch, and the major AI companies document honoring it for their crawlers, with exceptions for some user-triggered fetchers. Llms.txt is a guide to content: it suggests what to read, and nothing enforces it. The proposal describes llms.txt as a complement to robots.txt, not a replacement.', }, { question: 'Can I block AI bots with llms.txt?', answer: 'No. Llms.txt has no allow or block rules. To block or allow a crawler, use its token in robots.txt, and for user-triggered agents that may not follow robots.txt, use a firewall rule.', }, ]; ### Quick take llms.txt is a proposed markdown file at the root of a website. It gives AI tools a short summary of the site and a list of the pages worth reading. Jeremy Howard [proposed it in September 2024](https://llmstxt.org/), and it's still a proposal. Developer documentation sites have taken to it. None of the AI search crawler owners I track says its crawler reads it, and Google says Google Search ignores the file. Check yours with the [free llms.txt checker and validator](/tools/llms-txt-checker/). For most local businesses and stores, it's a small file with no documented payoff. I publish one on this site anyway. Below is what it does, who uses it, and how I'd write one if you decide it's worth twenty minutes. ### What the file is The [proposal at llmstxt.org](https://llmstxt.org/) lays out a markdown file served at `/llms.txt`. In order, it has: 1. An H1 with the name of the site or project. This is the only required part. 2. A blockquote with a short summary. 3. Optional paragraphs or lists with more detail, without headings. 4. H2 sections, each a list of markdown links with an optional note after a colon. 5. An optional section titled "Optional" for links an AI tool can skip when it needs to save space. The proposal also asks sites to offer a clean markdown copy of each page at the same URL with `.md` added. That part matters for documentation sites. For a ten-page service business site, I'd skip it. A short file for a made-up plumber looks like this, part by part. The business is fictional: ### Why it exists The proposal's reasoning is about context windows. An AI tool can't read a whole website at once, and pages built for people are full of navigation and markup that make them hard to turn back into clean text. A short file of hand-picked links points the tool at the pages that matter. It's also explicit about when the file gets used. The author says he expected llms.txt to be useful mainly for inference rather than training: an agent reads it on demand, when it needs to know about a topic while it helps a user. It's meant to sit beside robots.txt. The proposal says it can give context for content robots.txt already allows. ### Who uses it What I can verify from primary sources, as of October 2026: **Developer docs publish it.** [OpenAI](https://developers.openai.com/llms.txt), [Anthropic](https://platform.claude.com/llms.txt) and [Perplexity](https://docs.perplexity.ai/llms.txt) all serve llms.txt files for their developer documentation, and Perplexity's crawler page carries a banner pointing AI agents to its file. That fits the use case. A coding assistant reading API docs is the exact situation the proposal describes. **The search crawlers don't mention it.** For my [AI crawler directory](/tools/ai-crawlers/), I read the crawler documentation from OpenAI, Anthropic, Perplexity, Google, Apple, Meta and Common Crawl. None of those pages says its crawler reads llms.txt or uses it to choose what to show or cite. That doesn't prove no system ever fetches the file. It means no AI search company has said it uses one, so you can't count on it. **Google says Search ignores it.** Google's [guide to optimizing for its generative AI features](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide), updated July 2026, names llms.txt directly. It says you don't need AI text files to appear in Google Search, that Search ignores them, and that creating one will neither harm nor help your visibility or rankings there. The same guide leaves the door open: "It's completely fine if you decide to create and maintain LLMS.txt files (or other similar files) for other services or systems that use these files." Its [AI features page](https://developers.google.com/search/docs/appearance/ai-features) adds that a page only needs to be indexed and eligible to show in Search with a snippet. I've seen adoption counts for llms.txt quoted around the web. I'm leaving them out, because I can't trace any of them to a method I can check. If you want an adoption number that means something for your own site, look in your server logs for requests to `/llms.txt` and see who's asking. ### When it's worth having Write one if your site has a lot of material an agent could get lost in: 1. **Documentation or a knowledge base.** This is the case the proposal was built for. 2. **A large catalog with policies that matter.** For a store, shipping rules, return windows, sizing and warranty terms are what a shopping assistant needs to get right. 3. **Facts that AI answers keep getting wrong.** A short file that states your service area, hours and prices in one place costs nothing. It won't fix the wrong answer by itself, but it's one more clean source. Don't count on it if your problem is that assistants never name you. In my audits, the businesses that got named were the ones other sites talked about. No file on their own site changed that. I covered that in [how to show up in ChatGPT and AI Overviews as a local business](/blog/how-to-show-up-in-chatgpt-and-ai-overviews-local-business/). ### How I write one This is the same order I use when an [AI answer audit](/ai-answer-audit/) turns up wrong facts. 1. **Start from what the answers get wrong.** Ask ChatGPT and Google's AI a few questions about your business and note every fact they miss. Those facts go in the blockquote, stated plainly. 2. **Make it match everything else.** The hours, address and service area have to be the same as your site and your Google Business Profile. A file that disagrees with your listings gives the assistant a third version to choose from. 3. **Link only pages you want quoted.** Services, prices, service area and policies. Leave out tag pages, carts, account pages and anything thin. 4. **Write one plain line per link.** Say what's on the page in words a customer would use. 5. **Serve it at the root and check it.** It should load at `yoursite.com/llms.txt` as plain text. Check how your platform lets you serve a file at the root before you write one. The free [llms.txt checker](/tools/llms-txt-checker/) validates the file you have, flags structure problems, and can generate a starting draft if you don't have one. 6. **Update it when the facts change.** A stale llms.txt is one more wrong version for an assistant to find. Then go back to the work that has documented effects. The free [AI Visibility Check](/tools/ai-visibility-check/) reads your llms.txt, your robots.txt and the rest of what a tool can see from outside your site. Make sure the crawlers can reach you. I wrote up [how to check that](/blog/how-to-check-if-ai-crawlers-can-reach-your-site/). After that, fix your facts at the source and get named on the sites the answers already cite. ### Frequently asked questions ## On 4 of 32 Sites Running the Bing Ads Tag, It Was the Only Thing Loading Clarity URL: https://connercrowe.com/blog/bing-ads-tag-loads-clarity-session-recording/ Published: 2026-10-27 I logged what 32 sites' Microsoft Ads tags loaded. On 4, the ads tag was the only thing loading Clarity, the session recorder. The two-minute check and the fix. On October 27 I opened 131 well-known retail, home-goods, home-services, legal and medical sites in a clean browser and logged every request to Microsoft's ad and analytics hosts. The Microsoft Ads tag loaded on 32 of them on two separate visits. On 7 of those 32, the tag also asked for Microsoft Clarity, the session-recording tool, on both visits. Three of the 7 also loaded a separate Clarity tag. On the other 4, nothing else on the page loaded Clarity on either visit, so the ads tag was the only route in. On 1 of those 4, the recorder started on the page itself and sent session data to Clarity within 15 seconds, on both visits. ### Quick Take Microsoft's Universal Event Tracking (UET) tag can load Clarity by itself, so a site can record visitors' sessions with no Clarity code in its source or its tag manager. If you run Microsoft Ads, open your browser's network panel, reload your homepage and filter for `clarity.ms/tag/uet/`. If it's there and nobody on your side chose it, your privacy policy probably doesn't mention it, and a cookie banner that gates replay tools in your tag manager won't be the thing deciding whether it runs. The fix is a setting on the tag. If the live page still loads it afterward, a short head script blocks it. ### How an ads tag ends up recording sessions The UET tag is Microsoft's equivalent of the Google Ads tag. It loads `bat.js` from `bat.bing.com`, and `bat.js` fetches a small config file for your tag ID. When Clarity is enabled on the tag, that config inserts a second script from `www.clarity.ms/tag/uet/`, and that one loads `clarity.js`, which records the session. Each step is a script inserting a script, so none of it shows in your page source or your tag manager, only in the browser's network panel. This is a documented feature. Microsoft's Clarity docs describe using "a single UET tag" for heatmaps, session playbacks and conversion tracking. In October 2021 Microsoft announced it would migrate existing UET tags to include Clarity from mid-November that year, with some accounts excluded, and would turn Clarity on by default for newly created tags from November. So any tag created or migrated since late 2021 may be recording today without anyone at the business having decided it should. ### What the spot check shows The request chain, and on one site the recording itself. On the 7 sites, the browser asked for the `clarity.ms/tag/uet/` address on both visits, which only happens when the UET config tells it to. Of the 4 where the ads tag was the only route, one is a Shopify store that runs one of its UET tags as a custom pixel. There the whole chain, recorder included, ran inside Shopify's sandboxed pixel frame, which isn't the page the shopper sees. On the other 2, the recorder didn't send any session data inside my 15-second visits, so I'm not counting them as recording. Headless visits are noisy. A few tags loaded on one visit and not the other, so I only counted sites that behaved the same way twice. ### The check, and the fix The check takes two minutes. Open your site in a private window, open the browser's network panel, reload, wait ten seconds, and filter for `clarity`. A request to `clarity.ms/tag/uet/` means the ads tag brought it in. A request to `clarity.ms/tag/` followed by something other than `uet` means someone installed Clarity on its own. If you want Clarity, keep it on purpose. I recommend it on top ad landing pages, because ten recordings show a broken form faster than any report. Name it in your privacy policy, and hold the recording until the visitor clicks a consent choice, which is how my own site runs replay. Keep it off health and legal intake pages unless your counsel has signed off. Recordings can capture what people type if masking is loosened, and Microsoft's own 2021 announcement says Clarity shouldn't be used on sites with sensitive data, health care included. If you don't want it, turn it off on the tag. Microsoft's FAQ says you "might uncheck the Enable Microsoft Clarity box" under the tag's settings, and newer screens label the toggle "Enable Clarity". Before you do, check your conversion goals: Microsoft lets you build goals with Clarity's code-free event setup, and I'd expect those to stop counting when Clarity stops loading. Then reload the live page and run the same filter, because that's what your visitors load. While you're in the tag's settings, look for a second one. Microsoft documents a separate feature called UET Insights, announced in May 2023 with its own toggle. It loads from `bat.bing.com/p/insights/`, so a filter for `clarity` won't show it. Four to eight of the sites in my check requested it, depending on the visit. I haven't tested what it collects, so I'm not calling it recording, but if you want the tag to measure conversions and nothing else, check both settings. If the page still loads it, or you don't want the ad account deciding what runs on your site, block it on the site. Here's a head script that does it, tested in a browser before I published it. Place it before the tag manager so it's in place before the UET tag runs: ```js (function () { function isClarity(url) { try { var host = new URL(String(url), location.href).hostname; return host === 'clarity.ms' || host.endsWith('.clarity.ms'); } catch (err) { return false; } } var srcProp = Object.getOwnPropertyDescriptor(HTMLScriptElement.prototype, 'src'); if (srcProp && srcProp.set) { Object.defineProperty(HTMLScriptElement.prototype, 'src', { configurable: true, enumerable: srcProp.enumerable, get: srcProp.get, set: function (url) { if (!isClarity(url)) srcProp.set.call(this, url); } }); } var originalSetAttribute = Element.prototype.setAttribute; Element.prototype.setAttribute = function (name, value) { var isScriptSrc = this instanceof HTMLScriptElement && String(name).toLowerCase() === 'src'; if (isScriptSrc && isClarity(value)) return; return originalSetAttribute.apply(this, arguments); }; window.clarity = function () {}; })(); ``` It refuses any script whose address is on a `clarity.ms` host, through the two routes the UET chain uses to set one, and it gives `window.clarity` an empty function so the tag's own calls to it land somewhere harmless. The UET page-load beacon keeps firing. It also blocks a Clarity install you made on purpose, so don't use it if you want Clarity. It doesn't touch UET Insights. And it only protects the page it runs on: if your UET tag runs as a Shopify custom pixel, a script in the theme's head never reaches it, and the setting on the tag is the control that counts. It has no backslashes for a reason. If your site assembles inline scripts as JavaScript template literals, a regular expression like `/^https?:\/\//` loses its backslashes on the way out, and the build passes a script that either won't parse or quietly matches something else. Read what the build emitted, and parse it with `new Function()`, before you deploy. ### What I did not claim - **That any of these sites did something wrong.** Loading Clarity is often deliberate, and some of these businesses may name it in their privacy policies. I can see the requests, not the decision behind them, which is also why I haven't named the sites. - **A prevalence rate.** This is 32 tags on sites I picked, not a random sample, checked twice in one day, so 4 of 32 describes this set and nothing wider. - **That the off switch fails.** I'm describing the documented switches and telling you to check the live page after using them. - **What UET Insights collects.** I saw its requests and checked its script's size and name. I didn't test what it records. - **Anything about the law.** I'm not a lawyer. Whether a gap between what your page loads and what your policy says is a legal problem where you operate is a question for your counsel. ### Keep going If this hit, the next three pieces in the same universe: - **[The Bing pixel that said YOUR_VALUE_HERE](/blog/bing-pixel-your-value-here/)**. Another UET tag that looked configured and wasn't: three conversion tags sending placeholder text for six months. - **[What your website sends before anyone says yes](/pre-consent-tracking/)**. The five-minute browser test for every tag on your site, not only UET, and a dated table of where tracking claims stand. - **[The STACK Audit](/blog/google-ads-tracking-audit-guide/)**. The five tracking layers I check before touching a bid, and where a misbehaving tag shows up. Free PDF: **[The 25-page Tracking Stack](/freebies/tracking-stack.pdf)**. Where a tag like UET fits in the eight-layer stack, and the dedup key it has to match. [No email gate](/blog/free-pdfs-without-an-email-gate/). If you run Microsoft Ads and haven't looked at what your UET tag loads, [request the free ads audit](/audit-request/) and say Microsoft Ads in the form. I'll email the read-only access steps, start with the UET tag on the account and on your live page, and write it up within 24 hours. No call. ## Does Google AI Mode Show Ads? 42 of 100 Local Hiring Answers Had One. URL: https://connercrowe.com/blog/google-ai-mode-ads-local-hiring-answers/ Published: 2026-10-27 | Updated: 2026-10-29 Yes. I asked Google AI Mode 100 'who should I hire' questions in 5 U.S. cities. 42 answers had an ad. Plumber questions drew one in all 5. CPAs got none. > **Update, October 29, 2026:** I'm re-running this test, and the first repeat found far fewer ads. On October 29 I asked the same locked questions again, with each search set to its own city instead of run from my location near Spokane. So far 4 of 75 answers had a sponsored block (5.3%, 95% CI 2.1% to 12.9%), against 42 of 100 in the original run. Spokane went from 12 of 20 to 2 of 20. The gap is too big to be chance alone, but I don't know the cause yet. Three things changed between the runs: the location setting, the day, and the AI Mode page itself, which now usually loads without the "may include mistakes" footer. In one answer the sponsored block sat after the follow-up questions. Until the repeats are done, read the figures below as one run on October 27, not a typical rate. A third run is set for October 31, and I'll rewrite this post with all three runs after it. ### Does Google AI Mode show ads in local answers? Yes. On October 27, 2026, I asked Google's AI Mode 100 "who should I hire" questions, the same 20 questions in each of five U.S. cities: Spokane, Seattle, Denver, Dallas and Kalispell, Montana. 42 of the 100 answers had a block labeled "Sponsored" inside the answer. Home services drew ads in 34 of 55 answers. Professional services, like lawyers, dentists and CPAs, drew one in 30. If you're a plumber, electrician, exterminator, house cleaner or mover, that means a competitor can pay to sit inside the answer when someone asks the AI who to hire. ### How I tested it I wrote 20 questions a local buyer would ask, like "Who is the best plumber in Denver, CO?" or "Best CPA for a small business in Dallas." Eleven are home services. Six are professional services: a family dentist, a personal injury lawyer, a real estate attorney, a CPA, a chiropractor and a marketing agency. Three are auto repair, a vet and a furniture store. The furniture question is a shopping question more than a hiring one, so I grouped it with auto repair and vets instead of home services. Each city got the same list with the city name swapped. The one exception is Spokane's electrician question, which named Spokane Valley. I saved each city's list and its SHA-256 hash before that city's first question ran, and I haven't changed them since. The lists and hashes are at the end of this post. I asked each question once, signed out, in a desktop browser, from my own location near Spokane. I counted an answer as having an ad when a block labeled "Sponsored" appeared in the answer text before Google's "AI responses may include mistakes" footer. I read the text around all 42 hits to confirm each was an advertiser block. I also searched the other 58 answers for any other ad label and found none. I started with Spokane, where I live, and Kalispell, a small city in northwest Montana. Then I added Seattle, Denver and Dallas: three larger markets, two of them outside the Northwest. ### How often did AI Mode answers include an ad? | City | Answers with an ad | 95% confidence interval | |---|---|---| | Spokane, WA | 12 of 20 (60%) | 38.7% to 78.1% | | Seattle, WA | 9 of 20 (45%) | 25.8% to 65.8% | | Denver, CO | 7 of 20 (35%) | 18.1% to 56.7% | | Dallas, TX | 7 of 20 (35%) | 18.1% to 56.7% | | Kalispell, MT | 7 of 20 (35%) | 18.1% to 56.7% | | All five | 42 of 100 (42%) | 32.8% to 51.8% | Every city's interval overlaps every other's, so I wouldn't read anything into the city order. Spokane also isn't a fair comparison. It's where I searched from, and it's the only city I ran in the morning, so local advertisers could have reached me by location as well as by the city in the question. The intervals treat each answer as independent. They aren't fully: the same 20 questions repeat in each city, and many came out the same everywhere. The real uncertainty is wider than the table shows, and it doesn't account for answers changing from one run to the next. ### Which local businesses get ads in AI Mode answers? The type of business mattered much more than the city. | Business type | Answers with an ad | 95% confidence interval | |---|---|---| | Home services (11 questions per city) | 34 of 55 (62%) | 48.6% to 73.5% | | Auto repair, vet, furniture store (3 per city) | 7 of 15 (47%) | 24.8% to 69.9% | | Professional services (6 per city) | 1 of 30 (3%) | 0.6% to 16.7% | By city, home services drew ads in 9 of 11 questions in Spokane, 7 of 11 in Seattle and 6 of 11 in Denver, Dallas and Kalispell. By question, here's how many of the five cities had an ad in the answer: | Question | Cities with an ad (of 5) | |---|---| | Plumber | 5 | | Electrician | 5 | | Pest control | 5 | | House cleaning | 5 | | Moving company | 5 | | Roofing | 4 | | Furniture store | 4 | | Window replacement | 2 | | Veterinarian | 2 | | HVAC repair | 1 | | Landscaping | 1 | | Kitchen remodel | 1 | | Auto repair | 1 | | Real estate attorney | 1 | | Concrete work | 0 | | Family dentist | 0 | | Personal injury lawyer | 0 | | CPA | 0 | | Chiropractor | 0 | | Marketing agency | 0 | The only professional-services ad in the whole test was a law firm's text ad on the Dallas real estate attorney question. ### How does this compare to other AI Mode ad studies? The biggest study I know of is [SE Ranking's](https://seranking.com/blog/google-ai-mode-ads/), published July 14, 2026. They checked 50,032 U.S. commercial keywords on June 30 and found AI Mode text ads on 29.45% of them. Healthcare had the lowest ad presence in their data, 2.64%, and the low-ad niches leaned toward informational and "your money or your life" topics. That lines up with dentists, chiropractors, lawyers and CPAs drawing almost nothing here. The two numbers aren't directly comparable. Their sample is 500 times bigger and covers a broad mix of commercial searches. Mine covers one narrow, high-intent kind: asking who to hire in a named city. They also counted text ads only. Counting just the text ads here, 14 of my 100 answers had one. Most of the ads I saw were business listings. ### What do the ads look like? 28 of the 42 ads were business cards or short business lists, usually under a line like "To get a quote for your project, you can connect with these local plumbers." One, in Seattle, was a national furniture retailer rather than a local business. The other 14 were single text ads with the advertiser's name, a line of ad copy and their website. I sorted them by reading each one. In all 42, the "Sponsored" label sat in the last 40% of the answer text, after most of the AI's own answer. No answer had more than one sponsored block. I counted that from the saved answer text, which I'm not publishing because it names the advertisers. I'm leaving the businesses in the answers unnamed. The point is how Google builds the answer. ### Are AI Mode's recommendations themselves sponsored? The paid placements I saw were labeled and sat in their own block. I didn't measure whether the advertisers also showed up in the AI's unpaid picks, so I can't say whether paying for the ad has any effect on those. ### What this means if you own a local business 1. **The AI answer now has paid space in it.** For plumbing, electrical, pest control, cleaning and moving questions, my one run drew an ad in all five cities. It's labeled, but it sits inside the answer a buyer is reading. If you already pay for those clicks on regular search, my [home services cost-per-lead benchmarks](/home-services-paid-search-benchmarks/) show what plumbing and roofing leads cost there. 2. **If you're in professional services, there was almost no paid space to compete with.** Dentist, CPA and chiropractor questions drew no ads in any city, and lawyer questions drew one in ten. That leaves the unpaid picks. In my [six-audit study](/blog/six-ai-search-audits-crawler-access/), the businesses engines named were the ones directories, reviews and roundups vouched for. 3. **Ask whoever runs your Google Ads whether you're eligible to show here, and whether they can see impressions from it.** Google has changed which campaign types can serve in AI Mode more than once, so [check the current rules](https://searchengineland.com/google-ai-mode-ads-everything-you-need-to-know-494628) against your own setup. Full disclosure: I run Google Ads for clients, so I have a stake in this. ### What I don't know - **One run per question.** AI answers change from one run to the next, and so do ads. A second run the same day could land on different numbers. - **One window of time.** All 100 questions ran between about 11 a.m. and 2 p.m. Pacific. I didn't test whether time of day matters. - **One engine.** I only checked Google's AI Mode. ChatGPT, Gemini and Perplexity weren't part of this. - **One location.** Someone searching from a phone in Dallas could see something different. - **Small samples.** 20 questions per city is enough to see the pattern by business type, not to rank cities. ### The data - [All 100 answers, one row each](/benchmarks/data/ai-mode-ads-2026-10-27/results.csv): city, question, ad or no ad, ad format and where the label sat. Advertiser names are left out. - The five question lists, with the SHA-256 hash I saved before each city ran: - [Spokane](/benchmarks/data/ai-mode-ads-2026-10-27/spokane-questions.txt): `4560206cbfbfcdb9d26977d231a9340906eb7328fa449bb583d89e01bb8ea9f4` - [Kalispell](/benchmarks/data/ai-mode-ads-2026-10-27/kalispell-questions.txt): `d3d34f5ce3a40d8e56f13998ea90efebc25bf096a079f040549f30916faf4caa` - [Seattle](/benchmarks/data/ai-mode-ads-2026-10-27/seattle-questions.txt): `58dfef61cc40b0e6a3626c66523d94c910f5c671eb1fb576735fa077fb327d1c` - [Denver](/benchmarks/data/ai-mode-ads-2026-10-27/denver-questions.txt): `7aeaa05e31e6d8514cb865d30606c2f6bcbae555d843157fd30f826a974893df` - [Dallas](/benchmarks/data/ai-mode-ads-2026-10-27/dallas-questions.txt): `4586871f8be7273c34c43de099c50839cc37155124f73e20420cfe28b9255c73` If you want to know what AI assistants say when someone asks who to hire in your category, I put together a free snapshot for that on the [AI answer audit](/ai-answer-audit/) page. For how often AI answers name a business at all, and how often they get its facts wrong, see my [AI answers field report](/benchmarks/ai-answers-field-report/). The free [What AI says about you](/what-ai-says-about-you/) template lets you run the questions on your own business. ## I Measured My Copilot Citation Drop Against Its Best Five Weeks URL: https://connercrowe.com/blog/copilot-citations-drop-wrong-baseline/ Published: 2026-10-19 I called a 45% fall in my Copilot citations. Against June and July, the real fall was smaller and started a week later than I said. How I read the report now. On September 27 I exported Bing Webmaster Tools' AI Performance report for my own site and wrote one line in my audit notes: Copilot citations were "decaying ~40% fortnight over fortnight." The window behind that line was a day off. Fixed, the two fortnights averaged 32.2 and 17.7 citations a day, a 45% fall. The window was the small mistake. The comparison was the big one: the fortnight I measured against sat inside the busiest five weeks in the report's 90-day view. ### Quick Take I called a 45% fall in my Copilot citations. Against June and July it was about 16%, and it didn't start where I thought. The first week of my "decay" fortnight sat right at the early-summer level. The second week was well under it, and a later pull says the fall carried into October. So there's a real decline in here, smaller and later than my note said. Before I call a citation drop now, I find the last stretch that wasn't a high, deal with the report's zero days, and write the comparison down before I look. ### What the AI Performance report counts Microsoft put AI Performance into Bing Webmaster Tools as a public preview in February 2026. It counts how often your pages appear as sources in Microsoft Copilot, Bing's AI summaries and some partner surfaces. You get total citations, the average number of your pages cited per day, the pages cited, and the grounding queries, which are the searches the AI ran behind the scenes to find sources. It has no click or traffic figures. The tables are a sample, and the report says so on the page. My 30-day per-page table came back "No results found" while the 30-day total read 699, so I treat every figure in it as a floor. And it has zero days I can't explain. My 90-day series reads exactly zero on July 27, August 1 through 6, and August 22 and 23. Nothing on my site was down on those days. I cannot tell whether Copilot stopped citing me or the report stopped counting, and the export doesn't say. In a 14-day window, two zero days cut the average by a seventh on their own. ### Where my comparison went wrong Three windows from the 90-day series: | Window | Days | Citations per day | |---|---|---| | June 28 to July 26 | 29 | 21.1 | | August 7 to September 11, excluding two zero days | 34 | 37.8 | | September 12 to 25 | 14 | 17.7 | The middle row is the outlier. It holds the report's nine highest days, including two at 74. It began on August 7, four days after I shipped a new cluster of answer pages, one of which became my second most-cited page, and it peaked during a round of indexing and internal-link fixes on August 11 to 13. That launch landed in the middle of a six-day run of zeros, so I can't tie the high to it, or to the fixes, or to anything else I did. My note compared the bottom row against the tail of that run, so what it measured was the end of a high. I picked the June and July window after seeing the data too, so I checked whether the answer depends on it. It does, a little. Against late June and early July (17.5 a day) the fortnight is flat. Against the second half of July (24.9) it's 29% down. Split the fortnight in half and the picture gets clearer: September 12 to 18 averaged 21.7, right at normal, and September 19 to 25 averaged 13.7, which is 22 to 45% under every June or July window I tried. My note got the size of the fall wrong and the start of it wrong. It didn't get the direction wrong. I also tried the obvious control, Bing's ordinary search numbers, to see whether the whole site was losing visibility or only the AI layer. My Search Performance report showed 396 impressions and 6 clicks over roughly the same three months. At about four impressions a day, it's too thin to tell me anything either way. My own working notes leaned on that same series to rule out an indexing problem, and I'm withdrawing that conclusion here. ### Before I call a citation drop I check in this order now, and I'll use it on the next client report that shows a fall. If someone hands you an AI visibility report with a two-week change in it, these are the questions to ask them. 1. **Find the last window that wasn't a high.** Plot the full daily series and pick a stretch with no spike in it, at least four weeks long. Then try a second reasonable window and see whether the answer changes. 2. **Mark the zero days and leave them out of both sides.** A run of zeros can be a reporting gap, and an average that includes them drops for reasons that have nothing to do with the site. 3. **Open every cited URL and confirm it still resolves.** I did this on October 19 for the 40 pages the report had cited. 33 still build as pages, and the other 7 redirect to live pages that absorbed them. Two of the 33 are blog posts I noindexed on October 11, which held 16 of the 1,515 sampled page citations between them. That's mine to own, but it isn't the fall. 4. **Write the comparison down before the next pull.** Which windows, which metric, and what size of change counts as real. I write that down before I open the next export. ### October, and the edits I stacked on it A later pull, taken October 18, put my October daily rate below the June and July baseline. That fits the slide that started around September 19. I have not re-measured it for this post, so I'm not printing the figure, and I don't know the cause. I also made it harder to find out. Between October 3 and October 19 I changed the four pages Copilot cited most 16 times, across 9 commits. Those four held 964 of the 1,515 sampled page citations, about 64%. Some of those edits were fixes Copilot's own grounding queries pointed at, and one put my name into the paragraph each page leads with. The last was a one-line price update. With that many edits stacked on four pages, I won't be able to point at the one that moved the number, whichever way it moves. So apart from price changes, I have stopped editing those four pages until the 30-day re-test of my AI search audit on October 28. Changing the thing you're measuring while you measure it is the same mistake as picking the baseline after you've seen the number. ### What I did not claim - **That the decline isn't real.** It is. It started later and ran smaller than my note said, and the October pull says it kept going. - **A cause for the August high or the fall after it.** Both overlap with work on my site, and the zero days sit in the way of tying either to a change. - **That the counts are complete.** Every figure here comes from a report whose tables Microsoft labels a sample, and some of its zero days may be missing data rather than real zeros. - **Anything about other sites.** This is one site's report. I'd run the same checks on a client's, but I would not expect these numbers. ### Keep going If this hit, the next two pieces in the same universe: - **[I ran six AI search audits. Crawler access was almost never the problem.](/blog/six-ai-search-audits-crawler-access/)** The buyer-question side of the same measurement, and why a citation count isn't the same as getting named. - **[Migrating a site AI assistants already cite](/blog/migrating-a-site-ai-assistants-already-cite/).** What to protect before a rebuild touches the pages engines quote. Citation counts measure how often an engine leaned on your pages. Whether it named you when a buyer asked who to hire is a separate number. If you run a practice or a law firm, the free [AI answer snapshot](/ai-answer-audit/) is a first look at that one: ten real buyer questions put to three engines, with a ranked fix list. It's a single run, so treat it as a starting read rather than a trend. The full audit in paid work locks its question panel before the first run, and every re-test reuses the same file. Since this post, I've pooled the buyer-question side across every audit and snapshot into one [AI answers field report](/benchmarks/ai-answers-field-report/), with counts and intervals. To run the same check on your own business by hand, use the free [What AI says about you](/what-ai-says-about-you/) template. ## I Ran Six AI Search Audits. Crawler Access Was Almost Never the Problem. URL: https://connercrowe.com/blog/six-ai-search-audits-crawler-access/ Published: 2026-10-04 | Updated: 2026-10-29 3,623 graded AI answers across six businesses. Every AI crawler I tested could read every site. Who got named came down to what other sites said. > **Update, October 29, 2026:** I re-ran my own audit 30 days after fixing what it found. What moved, what didn't, and how the collection changed between runs are in [my 30-day re-test](/blog/how-long-does-ai-search-optimization-take-30-day-retest/). > **Update, October 27, 2026:** my [AI answers field report](/benchmarks/ai-answers-field-report/) supersedes the client figures in this post. It pools 9 businesses instead of five, keeps my own site out, drops answers the audits flagged as contaminated, and leads with problem questions (the buyer describes a problem and names nobody) instead of every question that doesn't name the business. On that measure the business was named 2.6% of the time (16 of 623). On October 27, 2026 I also rounded the per-business range endpoints below outward and removed one sentence that gave single-business figures. The current pooled figures are in the field report. ### Quick Take I audited how AI assistants answer buyer questions for six businesses, 3,623 graded answers in all. Every AI search crawler I tested could read every site. In five of the six, what held the business back was corroboration: what other sites say about it, and whether that's true. The sixth was missing the page format engines quote. ### What I measured Between late September and early October I ran the same audit on six sites. Five were clients, a mix of online stores and local service firms. The sixth was my own site. Client figures in this post are reported only in aggregate. Per-client results will be published only with each client's written permission. My own site's numbers are mine, so those appear in full. For each business I built a panel of about 50 to 60 buyer questions from real demand: converting search terms, Search Console queries and customer phrasing. I locked each panel and recorded its SHA-256 hash before the first run, so I can't quietly swap out a question that makes a result look bad. The re-tests reuse the same file, byte for byte. The engines were ChatGPT, Claude, Gemini, Perplexity, Google AI Mode, Google AI Overviews and Microsoft Copilot, though not every engine ran in every audit. Quotas, CAPTCHAs I didn't solve and a Microsoft terms wall got in the way. The plan was three runs of every question on every engine, because the same question asked twice gets different answers. Quotas cut that on several engines, and each report says where. A separate verifier pass re-ran a sample in fresh sessions. Anything that didn't reproduce got dropped or downgraded. Every rate below carries its raw count and a 95% Wilson interval. A percentage built from a few hundred answers that change run to run doesn't mean much without its range. ### What I expected vs what I found I went in testing two ideas. Most AI search advice says to check crawler access first. My notes from earlier audits said access was rarely the problem and corroboration usually was. I tested both instead of assuming either. Access was the one that had already bitten me. In June my Cloudflare firewall was returning 403 errors to five AI crawlers, and nothing in my analytics showed it. I fixed it then. This time access was clean everywhere. I fetched each home page and a set of money pages with the major AI crawlers' user agents. All 546 fetches came back 200. All six sites graded A on access. One caveat. Those fetches sent the crawlers' user-agent strings from a residential IP, not from the crawlers' own IP ranges. A firewall rule keyed to verified bot IPs could behave differently, and I haven't tested that. ### The pattern: engines read you, then ask someone else I grade each audit on six layers, from L1 (can the crawler get in) to L6 (can you measure the traffic). L5 is corroboration: anything outside your own site that confirms you exist and are good. In five of the six audits, L5 was the binding constraint. Across the five client audits, the share of non-brand answers that named the business ran from about 5% to 25%. Those are the questions a buyer asks without typing the business name, which is where new customers come from. Even at the top of that range, the business went unnamed in about three answers out of four. My own site sat below the whole range: 1.0% (2 of 203, 95% CI 0.3% to 3.5%). When someone typed my name, engines found me 85.4% of the time (41 of 48, 95% CI 72.8% to 92.8%). They could read me fine. When someone asked who to hire, they reached for marketplaces, directories and roundup articles. The reason was plain once I looked. I had two independent referring domains. The engines drew on 81 slots across roundups and results pages for my buyer questions, and I held none of them. Both of my non-brand mentions came on the same question, which matched one of my pages almost word for word. ChatGPT's held up on a re-run. Google AI Mode's didn't. That's about as far as on-site work goes alone. The client audits showed the same mechanism. Engines cited a business's own pages to explain how something works, then named someone else when the question turned to who to buy from. For that answer they trusted directories, review sites, editorial roundups and forums. Google now sells space in that answer too: when I asked its AI Mode 100 local hiring questions, [42 of the answers had a paid ad inside](/blog/google-ai-mode-ads-local-hiring-answers/). ### The exception One audit didn't fit. Off-site trust wasn't what stopped its comparison answers. The pages winning those comparisons weren't from bigger or more trusted sites. They shared a format: a verdict up top, a comparison table, a recent date and a named author. The business had no live page in that format, so its binding layer was L4, content format. That changed my thinking. Trust signals decide questions about who you are. Comparison questions go to whoever has the page in the right format. ### The accuracy finding Getting named is half the job. The other half is what the engine says once it names you. Three of the client audits counted wrong facts per answer, and in each a sizable share of the answers that named the business got at least one fact wrong. The [field report](/benchmarks/ai-answers-field-report/) has the current per-business range. The audits drew the base a little differently, so read it as a range, not a ranking. The other two counted errors in ways that don't convert to a per-answer rate, so I left them out. The errors had sources. Most traced to a stale directory profile, an old listing, a profile nobody maintained anymore, or the business's own pages saying two different things. Plenty of reviews and listings didn't protect a business either, because corroboration only helps when it's right. My own site had its share. Claude told people I don't have a Google Business Profile. I do, with reviews, and the claim reproduced on a re-run. I'd rank accuracy ahead of getting recommended, for two reasons. A wrong fact in a brand answer reaches the person who already knows your name and is checking you out, which is the person closest to buying. And accuracy is the faster fix, because most errors trace to a page someone can correct or claim. That's why brand fact-error rate is the 30-day metric in my re-test plan, and non-brand mention rate is the 90-day one. ### What this means if you own a business To run the by-name check on your own business, use the free [What AI says about you](/what-ai-says-about-you/) template. 1. **Check access once, then move on.** Confirm robots.txt allows the AI crawlers and fetch a few pages with their user agents. In six audits it came back clean every time. 2. **Ask the engines about you by name.** Read the answers for wrong facts, then find where each one came from. It's usually a directory or an old listing. Fix the ones you control first. 3. **Find the sources engines cite for your buyer questions.** Get onto the real ones. Don't seed reviews. The FTC's fake-review rule covers that. 4. **Put your name in the passage engines lift.** If your page gets cited and your name doesn't, the paragraph that gets quoted doesn't say who you are. 5. **Publish the format for comparison questions.** Verdict first, a table, a date, a named author. If you're rebuilding soon, read my notes on [migrating a site that AI assistants already cite](/blog/migrating-a-site-ai-assistants-already-cite/). ### What I don't know yet - **No client re-tests are published yet.** My own 30-day re-test is out: [what changed after 30 days](/blog/how-long-does-ai-search-optimization-take-30-day-retest/). The client figures here are first baselines, not before-and-after numbers. I wrote down the success rules on October 4, before any re-test: a change only counts when the 95% intervals don't overlap. - **Panel answers aren't total AI visibility.** These rates count answers to my locked buyer questions. Bing Webmaster Tools reports citations across Copilot answers in general, not only my questions, and on one audit it showed far more citations than the panel caught, led by its how-to guides. Its counts are a sample, though, and they swing hard week to week. I [misread a drop in my own](/blog/copilot-citations-drop-wrong-baseline/) by comparing against a high. Read the rates as share of buyer answers, not overall reach. - **Engines answer differently run to run.** On my own audit, the verifier's fresh sessions matched 21 of the 23 sampled runs it could complete (91.3%, 95% CI 73.2% to 97.6%). One of the two misses was a recommendation of me that vanished on the re-run. - **Location.** Most runs came from my location in Washington. I wrote the city into local questions, and the reports flag answers that localized to me anyway. - **Signed-in leakage.** On several audits my own signed-in assistant accounts carried my context into some runs. Those runs were excluded or flagged. - **Coverage.** My own site got 251 of 721 planned runs before quotas stopped me. I ran 225 more after I had started fixing things, and left them out of the baseline. - **Sample.** Six sites isn't a market study. Five are clients, picked because they're my clients, and one is me. If you want this run on your own business, the method and what you get back are on the [AI answer audit](/ai-answer-audit/) page. If the fix turns out to be the site and its listings, that's what my [service-business websites](/service-business-websites/) build covers. ## The Fields You Most Want to Change Are the Ones That Vanish URL: https://connercrowe.com/blog/fields-you-most-want-to-change-vanish/ Published: 2026-09-28 A Google Ads mutate can return success and change nothing. The standard field mask drops any value equal to its default, and off is a default. I moved a lead conversion action's default value from `1.0` to `0.0` and the API told me it worked. It hadn't. The value was still `1.0` a minute later, and the account carried on assigning a dollar of invented revenue to every lead it counted. That was one of three writes in five weeks, across three accounts, that came back successful and changed nothing. ### Quick Take A 200 is not proof that a URL is good, and a success response is not proof that a write landed. Same rule, [one layer further in](/blog/a-200-is-not-proof-redirects-that-eat-the-gclid/). On the Google Ads API the reason is specific. The helper Google documents for building a field mask compares values rather than presence, so a field you set to `false`, `0`, `0.0` or `""` looks unchanged to it and never reaches the mask, and Google ignores any field the mask doesn't name. The writes that vanish are the ones setting something to off, zero, or empty. ### Why off, zero and empty are the dangerous values Start with the part Google publishes itself. The Python client library's field-mask page hands you a helper, `protobuf_helpers.field_mask`, and documents what it does when the first argument is `None`. The mask it returns "will just contain all of the fields on the second protobuf object that are not set to their default value." Read that against what you were trying to do. You set `target_cpa_micros` to `0` because zero is how you say there is no target. The helper walks your object, finds `0`, and reads it as a field you never touched. It leaves it out. Google's side then ignores any field the mask doesn't name, even though you sent the value, and returns a success, because the request was legal. Legality was never the thing in doubt. Underneath that helper is a protobuf rule that explains part of it. In proto3 a plain scalar field carries no record of whether anyone set it, so "the default value is synonymous with 'not present' for purposes of serialization." A `0` you meant and a `0` nobody touched are the same bytes. That covers the bid-strategy write, where `target_cpa_micros` is a plain `int64`. It does not cover the other two, and this is the part worth knowing. Both of those fields are declared `optional` in the Google Ads protos, which means proto3 does track their presence and protobuf would have serialized them. They died anyway. Read the helper's source and you find it comparing `getattr` values rather than calling `HasField`, so a tracked field you deliberately set to its default still reads as unchanged and still gets dropped. On the nested one it's worse: the helper recurses into `value_settings`, finds nothing that differs, and the whole subtree disappears. So the rule isn't that protobuf drops defaults. It's that the mask helper does, whether or not protobuf would have. Here is why it lands on the writes you care about. Intent maps onto defaults. Turning something off is `false`. Removing a target is `0`. Clearing a list is empty. Protobuf can't tell subtraction apart from doing nothing, so every write where you take something away is a write at risk. ### What it looked like in an account The `1.0` to `0.0` write above is the clean case. I ran the mutate, got the success back, and only found it because I'd started querying the object again afterwards. The value sat there unchanged while Smart Bidding kept treating a dollar a lead as revenue. The second one showed its own failure in the response and I nearly missed it. I wrote a boolean `false` to demote a conversion action, and the update mask that came back on the operation held one entry: `resource_name`. The field I cared about wasn't in it. The evidence was sitting in the response object the whole time, and I hadn't been reading response objects, because they said success. Worth saying plainly that this particular field is the wrong one to be reasoning about at all, for reasons I've [written up separately](/blog/primary-for-goal-is-a-legacy-field/). What makes it a receipt here isn't the field, it's the one-entry mask. None of this is unique to Google, and the neighbouring cases have their own causes. Staging a Tag Manager change through an API client, I sent four parameters on a Conversion Linker tag and read three back. The one that vanished was a plain string, not a boolean. The three survivors came back retyped as `template`, which stores a boolean `false` as the string `"false"`, and a non-empty string is truthy in JavaScript, so there's a live risk of the setting inverting. I reproduced the loss and the retype on demand rather than inferring them. I did not test what the container runtime does with the retyped value. A monitoring vendor's v2 API does a blunter version: it accepts an SSL reminder setting, returns `{"stat":"ok"}`, and the value never moves. Different causes, one class. The write is accepted and the system is unchanged. ### If you are the one paying for this You don't need any of the above. The question worth asking whoever touches your account, whether that's an agency, a tool, or an agent, is not whether they made the change. It's what they read back afterwards to confirm it, and when. "The API returned success" is an answer about their request, not about your account. A change that silently failed looks identical to one that worked, from the dashboard and from the outside, for as long as nobody goes and looks. This is the write-side half of the verification step in [the STACK Audit](/blog/google-ads-tracking-audit-guide/). That pass checks what an account currently reports. This one is about whether the last person to change it did. ### One footnote for engineers Three habits close it, and only the third proves anything. The first one contradicts Google's own Python sample, which builds the mask with `protobuf_helpers.field_mask(None, campaign._pb)`. That sample is fine for adding a value and unsafe for removing one. ```python from google.protobuf import field_mask_pb2 # client.get_type("FieldMask") does not exist # 1. State the paths. Never derive them when the value might be a default. # For a message that has subfields, use the subfield path: masking the bare # parent is rejected with FIELD_HAS_SUBFIELDS. Google's stated reason for # that error is to stop you clearing subfields you never populated. op.update.resource_name = svc.campaign_path(CID, CAMPAIGN_ID) op.update.maximize_conversions.target_cpa_micros = 0 client.copy_from(op.update_mask, field_mask_pb2.FieldMask( paths=["maximize_conversions.target_cpa_micros"])) # 2. Probe an unfamiliar shape instead of spending a live mutate on a guess. # validate_only belongs on the REQUEST OBJECT. As a method kwarg it raises # TypeError before the call leaves your machine, which reads exactly like # the API rejecting you. req = client.get_type("MutateCampaignsRequest") req.customer_id = str(CID) # proto-plus rejects an int here req.validate_only = True req.operations.append(op) svc.mutate_campaigns(request=req) # 3. Re-read the object with an independent query. This is the only proof. ``` Google's own docs land in the same place on the first habit: if you mean to clear a field, name it in the mask. And `validate_only` does less than the name suggests. It checks the request is legal, then skips the execution and returns an empty response. Note the shape of that. A successful validation returns nothing. So does a successful no-op. Emptiness is not information. ### What it costs to leave alone The cost isn't the failed write. It's the belief that it succeeded, which stops anyone from looking again. An account in this state reports itself as configured. A bid strategy that was supposed to change is still the old one, doing what it did before, sometimes on a budget that was raised to suit the new one. A lead action still carries a made-up value that Smart Bidding reads as revenue and buys more of. Both are live bidding inputs, and neither surfaces as an error anywhere, because there wasn't one. It compounds under automation. A script or an agent making twenty writes in a batch reports twenty successes, and the ones that quietly did nothing are indistinguishable from the ones that worked. The more of your account maintenance runs through software, the more of it rests on a response that means "legal request" and gets read as "done." ### What I did not claim I haven't measured what any of these cost in spend. I know the writes didn't land and I know the state they left the accounts in. I never ran a counterfactual on what different bidding would have bought. And the field-mask behavior is a property of proto3 and of how a mask gets built, not a bug in Google's API, which does what its documentation says it does. ### The Receipts The three Google Ads instances ran between 30 July and 5 August 2026 across three accounts, through the Python client library on API v24. The fields were `maximize_conversions.target_cpa_micros` set to `0`, `conversion_action.primary_for_goal` set to `false`, and `value_settings.default_value` moved from `1.0` to `0.0`. On the second, the returned mask was `['resource_name']`. The Tag Manager parameter loss was reproduced deliberately on 1 September 2026 in a throwaway workspace, four parameters sent and three returned, then reverted. The monitoring case is UptimeRobot's v2 `editMonitor`, tested 15 September 2026, which returned `{"stat":"ok"}` for an SSL reminder flag that never changed on read-back; their v3 `PATCH /monitors/{id}` sets it. ### Keep going If this hit, the next two pieces in the same universe: - **[A 200 Is Not Proof (The Redirect That Eats Your gclid)](/blog/a-200-is-not-proof-redirects-that-eat-the-gclid/)**. The read-side sibling, and where this rule got its name. A correct status code on a correct page, with the click identifier stripped on the way through. - **[The field your Google Ads audit is reading is the wrong one](/blog/primary-for-goal-is-a-legacy-field/)**. What happens when you read `primary_for_goal` instead of writing to it, and why it makes a clean account look broken. Free PDF: **[The 25-page Tracking Stack](/freebies/tracking-stack.pdf)**. Writes like these land on [Layer 05, the Google Ads layer](/frameworks/tracking-stack/), and show up wrong on Layer 08, reporting. No email gate. If software has been changing your ad account, the thing worth knowing is what signal that account is feeding Smart Bidding today, whoever or whatever last touched it. That's a read, not a write, and it's the free ads audit: [send me the account](/audit-request/) and you'll have it back in 24 hours. ## A 200 Is Not Proof (The Redirect That Eats Your gclid) URL: https://connercrowe.com/blog/a-200-is-not-proof-redirects-that-eat-the-gclid/ Published: 2026-09-15 | Updated: 2026-10-22 A redirect can return a 301 to the right page and still strip the gclid from every ad click. The conversion comes back attached to nothing. The check. Five redirects on a client's site were quietly deleting a quarter of his advertising measurement, and every check anyone had run on them came back green. The redirects were correct. Each one returned a 301 to exactly the page it should have. Click an ad, land on the right article, read it, call the office. Nothing to see. What they dropped was the query string. ### What the query string is carrying When somebody clicks a Google ad, Google appends a `gclid` to the destination URL. That parameter is the click. It's how a conversion recorded on your site gets matched back to the keyword, the ad, the device, the auction it came from. Microsoft uses `msclkid` the same way. Your UTM tags ride in the same place. All of it lives after the question mark, and a redirect that rebuilds the destination URL from scratch throws all of it away. Here's the shape of the failure: ``` Ad final URL: /old-page/?gclid=EAIaIQ...&utm_source=google&utm_medium=cpc Redirect: 301 -> /new-page/ Browser lands: /new-page/ ``` Status code 301. Correct target. And every conversion that visitor generates arrives in Google Ads attached to nothing, because the identifier that would have attached it was discarded one hop before the page loaded. ### Why nobody catches it Because everything a normal check looks at passes. Open the ad's final URL in a browser and you get the right page. Run a crawler over the site and the redirect resolves 200 with no chain and no loop. Look at Google Ads and the ads are approved, the landing page experience is fine, and the destination isn't flagged. Look at analytics and traffic is arriving. The only symptom is negative. Conversions get counted, but a growing share of them sit in campaigns and keywords that appear to have produced nothing. If the account is on automated bidding, the bidding is learning from a signal with a hole in it, and it'll quietly pull budget toward the segments whose conversions still make it home. On the account I found this on, the affected redirects sat under 860 of the 3,519 paid clicks in a thirty-day window, for months. The traffic was fine. The measurement was gone. ### The check One line, and it works on any platform. ``` curl -s -o /dev/null -w "%{http_code} -> %{redirect_url}\n" \ "https://example.com/old-page/?gclid=TEST123&utm_source=google&utm_medium=cpc" ``` You want the parameters back out the other side: ``` 301 -> https://example.com/new-page/?gclid=TEST123&utm_source=google&utm_medium=cpc ``` If `redirect_url` comes back bare, that redirect is eating your click identifiers. If there's more than one hop, run it again against each hop, because the loss can happen anywhere in the chain. For the final destination rather than the hop, follow the redirects and print where you ended up: ``` curl -s -o /dev/null -L -w "%{url_effective}\n" \ "https://example.com/old-page/?gclid=TEST123" ``` I ran this against every landing page on a site migration in September 2026 before letting a single ad repoint at it. The cutover script refuses to run otherwise. A 200 is a precondition, not the test. ### Where it comes from Four sources, in the order I find them. **Hand-written server rules.** Both Apache and nginx keep the original query string by default, right up until the replacement target carries a `?` of its own, and then they silently stop. In Apache, a `RewriteRule` whose substitution introduces its own query string replaces the incoming one unless you add the `QSA` flag. In nginx, `rewrite` preserves the arguments by default but `return 301 /path;` does not, so a `return` needs a trailing `$is_args$args`. The rule that looks tidiest, a literal destination with its own tracking parameters bolted on, is the one that drops the click. **WordPress redirect plugins.** Several of them normalize the destination and rebuild it clean. This is the most common source I see, because the plugin is doing what it thinks is tidy. **Marketing vanity URLs.** The `/promo` that forwards to the real page, set up in a registrar's forwarding panel years ago by somebody who has left. Domain forwarding at the registrar level almost never preserves the query string. **Host-level redirect files.** Netlify `_redirects` and Cloudflare Pages rules do preserve the query string by default, but a rule that hardcodes its own query parameters replaces rather than merges. ### The rule I'd write down Never accept a status code as proof that a URL is good. Verify that the thing you care about came out the other end. That's the same discipline as checking a form delivers mail rather than that it returned a success panel, and checking a tag fired rather than that the container published. The identity layer of [The Tracking Stack](/frameworks/tracking-stack/) is entirely made of parameters that survive or don't, and a status code never tells you which. A 200 tells you a server answered. It tells you nothing about what it answered with. If you're not sure your landing pages keep the click ID, the conversion side shows it: [the free ads audit](/audit-request/) reads what your ad account is actually counting, from read-only access. Back in writing within 24 hours. No call. ## Four Ways a Site Launch Lies to You URL: https://connercrowe.com/blog/four-ways-a-site-launch-lies-to-you/ Published: 2026-09-15 | Updated: 2026-10-22 A site can go live looking right while its search index is missing, generated files 404 and a build variable breaks a feature. The script that ended them. I put a client's new site live in September 2026 and it was broken for about an hour. Every page loaded. The design was right, the content was right, the redirects were right, and the thing rendered perfectly. Underneath, the site search told every visitor it was unavailable, two files the site links to from its own pages returned 404, and a build variable that the bundler never saw had left a feature on the page inert. I caught it, fixed it at 00:56, and then wrote a deploy script so it can't happen again. Four separate failures fired that night. None of them is exotic, and all four are invisible from the browser. ### One: the build command that skipped half the build The build was made with `npx astro build`. That runs the framework and nothing else. The `prebuild` and `postbuild` steps in `package.json` only run when the build is started through npm. On this site, every generated output is written in postbuild: the search index, the machine-readable files, the redirect markers. So the build succeeded, produced a directory full of correct-looking pages, and silently omitted everything that wasn't a page. The generalized version of this one isn't about Astro. Any modern site has a build pipeline with steps hanging off it, and calling the framework directly instead of the project's own build command skips them all without a warning. If your repository defines `npm run build`, that is the build. Reaching past it to the tool underneath is how you ship three quarters of a site. ### Two: an environment variable that was set, and wasn't The worst consequence that night came from a single build variable left unresolved, which meant a component that depends on it rendered nothing, and a feature built on that component sat inert on an otherwise perfect-looking page. I had set that variable in the shell. The shell reported it as set. The build never saw it. I still don't have a complete account of why, and I'd rather say that than invent one. Vite documents `process.env` as winning over the `.env` files, so setting it in the shell should have been enough, and on this build it wasn't. What I could establish is the ordering that fixed it: Vite loads `.env`, then `.env.local`, then `.env.[mode]`, then `.env.[mode].local`, each overriding the last. Moving the value into `.env.production` put it above the stale one in `.env.local` and the build picked it up. The comment inside `.env.local` telling you to use a shell variable is wrong for this project whatever the mechanism turns out to be. The lesson worth keeping: `echo $VAR` only ever reports on your shell. If a variable matters, assert on the built output, by grepping the artifact for the compiled value, rather than on the environment you built from. I wrote up the pre-campaign version of this check, where the flag was wrong on an ad destination rather than a deploy, in [The Preview Flag](/blog/preview-flag-ad-destination-launch-check/). ### Three: a fresh terminal that built a different site Environment variables are session-local, and I opened a new terminal. That build came out with the wrong build flags, which meant it declared itself `noindex` and generated no sitemap. It also got published. A site that tells search engines not to index it looks completely normal to a human. There's no visual difference, no error, no warning in the deploy log. You find out weeks later, when the traffic doesn't come back after a migration. ### Four: the host rebuilt it and undid everything The deploy went out with a plain `deploy --prod`. The hosting project has its own build command configured, so the host ignored the artifact I'd just built and rebuilt the site itself, without any of the variables that were only ever set on my machine. It then published that. This is the trap I would most expect to catch somebody who has done everything else right. You can build correctly, check the artifact carefully, and still have the host throw it away and substitute its own. The deploy has to be told to publish the directory as-is. ### What replaced all of it One script, and nothing else is allowed to deploy. It sets the variables, runs the full npm lifecycle rather than the framework directly, checks the artifact before it goes anywhere, deploys with the build step disabled so the host publishes exactly what was checked, and then verifies the live site afterwards. A deploy isn't finished when the upload succeeds. It's finished when the live domain has been asked whether the thing you expected is there. The script checks status codes on the pages that matter, samples the legacy redirects, confirms the generated files exist rather than 404, confirms the site declares itself indexable, and confirms the mail records still resolve to the mail host, because the one failure that would genuinely hurt this client is the one nobody would notice for a day. Ten checks against the live domain, seconds to run, every time. ### The pattern under all four Each of these failures produced a site that rendered perfectly, which is why they survive a visual review. The four questions I would now ask of any launch, in order: 1. Did the build run through the project's own build command, or did somebody call the tool underneath it? 2. Are the environment variables present in the compiled output, rather than in the shell that started it? 3. Does the live site declare itself indexable, and does its sitemap exist? 4. Did the host publish the artifact you checked, or did it build its own? I wrote separately about the fifth question, which is whether the redirects preserve the parameters your advertising depends on, over at [A 200 Is Not Proof](/blog/a-200-is-not-proof-redirects-that-eat-the-gclid/). Same family of problem. The page loads, so everyone stops looking. The tracking layers a launch can break are mapped in [the Tracking Stack framework](/frameworks/tracking-stack/). The full account of this particular migration, including what I haven't claimed about it yet, is in [the case study](/results/hand-wrist-institute-site-and-ads-cutover/). The launch checks above are yours to run. The ad account that will send traffic to the new site is the part I check for free: [send me read-only access](/audit-request/) and I'll write up what its tracking and campaigns are really reporting, back within 24 hours. No call. ## Migrating a Site That AI Assistants Already Cite URL: https://connercrowe.com/blog/migrating-a-site-ai-assistants-already-cite/ Published: 2026-09-15 | Updated: 2026-10-28 AI assistants that already cite URLs on your domain hold a stale index and give no warning when a citation breaks. The exposure most 2026 migrations miss. The client whose site I migrated in September 2026 found me through Grok. He asked it a question in my field, it named two providers, I was one of them, and he emailed me. That's recorded in my own notes as the attribution, because he told me on the call. Which means I spent a fortnight rebuilding the site of somebody who hired me because of an AI citation, on a domain that has hundreds of pages old enough and specific enough to be cited the same way. That reframed what the risky part of the migration was. ### Rankings are the exposure people plan for Everyone migrating a site knows to protect the rankings. You inventory the URLs, you map the redirects, you keep the titles, you watch Search Console for coverage errors, and if something drops you can see it drop. The feedback loop works. Search Console will tell you a page 404s. It'll tell you impressions fell. There's a dashboard, a date, and a graph with a cliff in it. AI surfaces give you none of that. ### What breaks, and how quietly An assistant that cites your site cites a specific URL. It either resolves or it doesn't, and there's no report anywhere that tells you it stopped. Four things go wrong in a migration, in rough order of how often I see them: **The URL changed.** Someone tidied `/carpal-tunnel-syndrome/` into `/conditions/carpal-tunnel-syndrome/` because it's a better structure. It is a better structure. It also breaks every citation pointing at the old one unless the redirect is there, and unlike a search engine, nothing comes back to re-crawl and forgive you on a schedule you can predict. **Two pages got merged.** This is the one that has no redirect fix. Content consolidation is usually good SEO hygiene, and it's the most destructive thing you can do to a cited corpus, because the passage that was being quoted no longer exists anywhere. A 301 sends the reader to a page that doesn't contain the answer they were cited to. **The headings moved.** Extraction works on structure. A page that answered a question under an `h2` phrased the way people ask it, rewritten into prose with a decorative heading, can stay live and stay ranked and stop being quotable. **The machine-readable files 404.** If your site publishes `llms.txt`, or a feed, or structured data that the page depends on, a build that omits them still produces pages that render correctly end to end. On a launch I did in September 2026, exactly that happened: the generated files were skipped by a build command that ran the framework instead of the project's build, and `llms.txt` returned 404 for about an hour on a site whose own pages link to it. I wrote that failure up in [Four Ways a Site Launch Lies to You](/blog/four-ways-a-site-launch-lies-to-you/). ### What I did instead The old site had 533 URLs. I crawled and inventoried all of them before a single page of the new site was written: status, final URL, title, description, H1, word count, canonical, indexable, in the sitemap. The new site was fitted to that inventory rather than the inventory being fitted to the new site. Three rules came out of it. **Consolidation is allowed, but only in one direction.** I want to be precise here, because my first draft of this post said nothing gets consolidated during a migration and that was a rule I had not actually followed. This migration retired 58 legacy URLs into the blog index and merged six more into surviving posts. What separates those from the failure above is what was in them. The retired set is almost entirely off-topic local filler that a previous content programme had produced: the best barbecue spots in town, the best boxing gyms, the farmers markets. None of it answers a question about hands or wrists, so none of it can be cited for anything the practice wants to be cited for. The six merges go the other way, folding three near-identical posts about food and joint pain into the one that covers it properly. The rule I would write down is narrower than "never consolidate." Do not merge or retire a page that answers a question in the subject you want to be known for. Everything else is fair, and a migration is a reasonable moment to do it, because you are rebuilding the redirect map anyway. What you cannot do is discover afterwards that the page you folded away was the one being quoted. **Every legacy URL resolves, and the redirects preserve the query string.** After cutover I sampled 40 rules against the live domain and checked both the status and the target. Query-string preservation matters for a different reason, which is that it carries the advertising click identifiers, and I've written that up separately in [A 200 Is Not Proof](/blog/a-200-is-not-proof-redirects-that-eat-the-gclid/). **Old media URLs resolve too.** Images embedded in other people's pages and in years of syndicated content are citations of a sort. The forced media redirects on this site land on the preserved originals, so they still resolve. ### How to check whether you have this exposure You can't query most of these surfaces directly, so the check is indirect, and it's a short one. Ask three or four assistants the questions your best pages answer, and write down every URL of yours they cite. That list is your real exposure, and it's usually a different set of pages than your top-ranking ones, skewed toward older content that answers a specific question and away from the commercial pages you'd protect first. For Copilot there's one partial exception. Bing Webmaster Tools' AI Performance report lists a sample of the pages Copilot cites, so add those to the list. Read its counts with care, because they swing hard week to week, and I've [misread a drop in mine](/blog/copilot-citations-drop-wrong-baseline/) before. Then look at your referrer data for assistant hostnames, and look for traffic to pages you'd forgotten about. On the migration I just did, the pages carrying the most of this exposure were condition articles written years before anyone was optimizing for a language model. Whatever that list contains, those URLs don't change, don't merge, and don't get their headings rewritten in the same release as everything else. ### The part I'd argue with myself about There's a fair objection here, which is that assistants re-crawl, indexes refresh, and a correct 301 will eventually be followed by everything that matters. Probably true, and I would still plan around the gap rather than the steady state. Nobody I've asked can tell me how long the refresh takes for any given surface, the pages most likely to be cited are the ones least likely to be missed in your own analytics, and the whole failure is silent. For the cost of an inventory and a redirect map you'd build anyway, the downside of being careful is nothing. The full write-up of that migration, including the eight ad landing pages and the 51% of conversions I deliberately didn't move, is in [the case study](/results/hand-wrist-institute-site-and-ads-cutover/). If you also run Google or Meta ads to the site you're migrating, the conversion tracking is the first thing a migration breaks. [The free ads audit](/audit-request/) reads it from read-only access, and I send the write-up within 24 hours. No call. Before and after a migration, the free [What AI says about you](/what-ai-says-about-you/) template gives you a fixed set of questions to ask the assistants, so you can compare like with like. For how often assistants name a business at all, see my [AI answers field report](/benchmarks/ai-answers-field-report/). ## The 51% of Conversions I Didn't Move to New Pages URL: https://connercrowe.com/blog/the-51-percent-i-didnt-move/ Published: 2026-09-15 | Updated: 2026-10-22 I repointed eight ad groups at new landing pages and held back the two carrying 51% of conversions. The eight are the cheap test before the expensive move. I spent a fortnight building eight landing pages for a client's search campaigns. The night the new site went live I repointed eight ad groups at them and left the two biggest ones alone. Those two carry 51% of the conversions in the account. They still point where they pointed before. Every instinct in paid media says finish the job. I want to explain why not finishing it was the decision that made the work readable. ### The spread that made eight of them obvious Thirty days before cutover, this account's conversion rate by destination looked like this: | Destination | Conversion rate | |---|---| | Site root (two ad groups) | 11.76% and 11.06% | | A legacy blog post | 0.88% | | A different legacy blog post | 1.11% | | A condition page | 8.65% | A tenfold spread inside one account, and not one of those ad groups had a landing page built for it. The ones at 0.88% and 1.11% were sending paid clicks to articles published in 2010, with a navigation menu, a sidebar, an author byline, and no obvious way to book anything. Moving those wasn't a judgment call. There's almost nothing below 0.88%, and the pages they were moving to were built for the search that triggered them. ### Why the two at 11% are a different question Because they were already the best thing in the account, and because the ground had just moved under them. The root those two point at is the home page, and the home page is brand new. Their 11% was earned on the old WordPress one. The moment the site went live, their baseline was invalidated whether or not I touched them. So the choice was between one untested change on 51% of the conversions, or two untested changes stacked on top of each other on 51% of the conversions, with no way afterwards to tell which one did what. If the number went down, I'd have no idea whether the new home page underperformed the old one or the landing page underperformed the home page. If it went up, same problem in the other direction. There's also a plain risk argument. A landing page that beats a 2010 blog post is a near-certainty. A landing page that beats a page already converting at 11% is a real question, and I would rather ask it with evidence than with confidence. ### The eight are a free experiment The eight ad groups I moved were the account's worst destinations. That makes them close to costless as a test: there's almost no downside left in a 0.88% conversion rate and a lot of headroom above it. And they're measuring the exact thing I need to know before touching the 51%, which is whether these landing pages convert at all. So the sequence is cheap evidence first, expensive decision second, with the cheap evidence bearing directly on the expensive decision. Moving everything at once would have spent the expensive decision to learn the cheap thing. ### Write the rule before you have the number The failure mode of holding something back is that the holding becomes permanent. Three weeks pass, the number is ambiguous, nobody wants to move 51% of an account on a feeling, and it sits there forever. So the decision rule went into the readout script the same day, before any data existed to argue with: - Moved-set conversion rate clearly above 11%: move the two held ad groups. - Between 8 and 11%: run them as a 50/50 experiment instead of a switch. - Clearly below 8%: leave them on the home page, and rethink the landing pages. It needs roughly 300 clicks across the moved set before it's allowed to read. That threshold is also written down, because the strongest pull in this situation is to look at the first day of data and see whatever you were hoping to see. Write the rule while you're still indifferent to the answer. It costs nothing on the day you make it and it's almost impossible to do later. ### What the client asked for His own framing, before any of this started, was that he didn't want to look up one day and find a sharp drop. Holding the 51% is the honest answer to that. Half of what's working keeps working while the other half proves itself, and when the change does come to the big ad groups it'll come with a number attached rather than an argument. I've written the rest of that engagement up in [the case study](/results/hand-wrist-institute-site-and-ads-cutover/), including the part where I publish no performance result at all, because there isn't one yet. If paid ads already feed the site you're about to change, [send me read-only access for the free ads audit](/audit-request/). I read the ad account and its [conversion tracking](/frameworks/lead-quality-stack/), and the write-up comes back within 24 hours. No call. ## Why I leave about $50 a day unspent in my own Google Ads account URL: https://connercrowe.com/blog/budget-stopped-being-the-constraint/ Published: 2026-09-06 | Updated: 2026-10-22 My campaign spends $11.71 of a $64.52 daily budget and loses 75.1% of impression share to rank. The bid that clears the floor needs a close rate I don't have. Between 21 and 25 August my own Google Ads campaign spent $51.04 across 8 clicks against a daily budget of $64.52. Read in the campaigns view with both impression-share columns showing, that window lost 0.0% of impression share to budget and 75.1% to rank. The 11 to 17 August window on the same two columns read 84.3% and 12.5%. Roughly $50 a day now goes unspent, I know why, and I'm leaving it there. ### Three changes, in order, and then the flip I made all three of these myself, in this sequence, and only the last one looks like an experiment. On 14 August I culled the keyword set. Click volume fell, which was the point. An audit I ran a week later read the fall as an over-correction and tried to re-enable four of the paused keywords. I reverted all of it the same day. On 17 August I wrote a bid plan against a daily budget of $14.61. Nearly every enabled keyword went to a $6.50 cap. I set it low on purpose, so one $30 click couldn't eat a whole day at that budget. On 21 August I moved the monthly target to $1,000 and the daily budget went to $64.52. My pacing tool, AdPulse, runs at a 2x limit, so that figure is exactly 2 x ($1,000 / 31). It's a ceiling rather than a computed catch-up number. I never re-tuned the bids underneath it, so the campaign spent the following week carrying a $14.61-a-day bid plan on a $64.52-a-day budget. Here's what the weekly review pulled on 25 August, three windows from the campaigns view with Search Lost IS (budget) and Search Lost IS (rank) added: | Window | Spend | Clicks | Leads | Lost IS, budget | Lost IS, rank | |---|---|---|---|---|---| | 11 to 17 August | $183.01 | 29 | 0 | 84.3% | 12.5% | | 18 to 24 August | $75.80 | 11 | 0 | 22.6% | 60.4% | | 21 to 25 August | $51.04 | 8 | 0 | 0.0% | 75.1% | One mechanical note before that table gets over-read. Impression share, lost to budget and lost to rank add up to 100%, so the 12.5% in the first row couldn't have exceeded 15.7% while the campaign sat out of money 84.3% of the time. What the third row shows is the rank loss with the budget ceiling taken off it. That's a clearer view of the same auction, not evidence of a harder one. The same weekly review puts the campaign at $11.71 a day on a trailing seven-day basis against that $64.52 budget. The budget increase revealed what the cull and the bid cap had already done to the account, and it took away the last excuse for the spend. Caps of $6.50 sit under first-page bid estimates of $13 to $32. The campaign can't enter the auctions it wants, however much money is parked behind it. So the question stopped being why spend stalled and became whether clearing the floor is worth doing. I decided it isn't. The reason is arithmetic rather than nerve. ### The arithmetic that stopped the bid raise August closed the 25th at $534.28 across 75 clicks with one genuine lead, which arrived on 9 August and reconciles against a real `generate_lead` event in my own analytics. One lead per 75 clicks is a 1.33% lead rate. That's the only conversion figure this account owns, it comes from the same account over the same month as the spend above, and everything below stands on it. Four steps, and I would run them in this order on any account where the rank column has taken over: 1. **Take the two lost-impression-share columns apart before touching anything.** Lost to budget and lost to rank point at opposite fixes. Mine moved from 84.3% budget in mid-August to 0.0% budget and 75.1% rank across the 21st to the 25th, which is a different account from the one the bid plan was written for. 2. **Pull the first-page bid estimate, not the market CPC.** Mine runs $13 to $32 across the enabled keyword set against a market CPC nearer $7. Google computes that estimate from the keyword's own Quality Score and from what competitors are bidding, so the gap between those two numbers is partly the price of my own account, and it's the number a bid has to clear. 3. **Multiply the floor-clearing bid by clicks per lead, and read the result as a ceiling.** At a 1.33% lead rate that's 75 clicks per lead. The weekly review put floor-clearing bids at $16 to $23, which lands cost per lead between $1,200 and $1,700. That's arithmetic off the bid cap rather than a realised CPC. You pay a cent over the next bidder rather than your own cap, so treat $1,700 as the top of the band. 4. **Divide by the close rate, because cost per lead isn't the number that decides this.** Cost per booked job is cost per lead divided by close rate, which is the formula I've already published for [home-services accounts](/blog/cost-per-booked-job-home-services/) and it applies to mine on the same terms. At the ceiling, an offer with a $5,000 floor needs a 24% to 34% close rate to return the offer price before I've done a minute of the work. My account has produced one lead. There's no close rate here to divide by. So the caps stay at $6.50 and about $50 a day goes unspent. The raise asks me to spend somewhere near $2,000 buying clicks to discover a close rate on a channel that's produced a single lead, and I would rather put that $2,000 of attention on the thing the account is measurably failing at. ### The lever is the landing page, and the account says so twice The Quality Score components here are unambiguous. Every keyword carrying a score reads `post_click_quality_score: BELOW_AVERAGE` alongside `creative_quality_score: ABOVE_AVERAGE`, which are the API's names for landing page experience and ad relevance. Google describes Quality Score itself as a diagnostic rather than an auction input, and the auction reads ad quality in real time through Ad Rank instead. So the ads are being read as relevant and the page behind them isn't. The first-page estimate moves with the keyword's Quality Score, which makes a below-average post-click read a plausible part of why my estimates sit at $13 to $32 against a market CPC nearer $7. No bid change touches it. Read by landing page through 25 August, `/ecommerce-website-design/` has taken 43 clicks and $272.16 for zero leads. `/shopify-store-design/` has taken 29 clicks and $243.07 and produced the one lead. More money has gone into the page that's never converted than into the page that has. Both of those are repository changes with an auto-deploy behind them. Neither is an ads write. That took me longest to accept, because the account is the surface I'm fluent in and the page is the surface that decides the price. ### The kill gate had to be rewritten too I set the original gate as volume: by 31 October, roughly $2,500 and roughly 375 cumulative clicks, and under 5 qualified leads means the channel stops. On 25 August I worked out that it could only ever return "kill." At the trailing seven-day rate the campaign lands nearer $1,320 and 190 clicks by 31 October, and five leads on 190 clicks needs a 2.6% lead rate against a measured 1.33%. The gate was going to fail on volume rather than on whether paid search works here. It was measuring the wrong thing. It now reads as a lead rate at about 200 clicks. Under 2% and paid search doesn't clear an offer at this price, so the money moves to outbound. Between 2 and 4% and the conversion rate gets fixed before a dollar of budget moves. Over 4% and it scales. The spoke version of this decision, written as the question people search, is [should I raise bids when the budget is not spending](/wasted-ad-spend/should-i-raise-bids-when-the-budget-is-not-spending/). ### What I did not claim This account produced one lead in the window I'm describing, on 9 August, and zero in the sixteen days after it. This isn't a success story and I'm not writing it as one. There's no before-and-after here and no lift to point at, only a decision with its arithmetic on the page. I'm also not claiming that budget increases never move spend. Mine didn't move because I'd already capped the bids under the first-page estimates four days earlier, which is a specific condition rather than a general rule. On a campaign losing impression share to budget, more budget is exactly the right lever, and my own 11 to 17 August window at 84.3% lost to budget is what that looks like. The 1.33% lead rate rests on one lead. I've never measured a close rate on this channel at all. That's enough to decline a bid raise, because declining costs nothing and the raise costs about $2,000. It's nowhere near enough to be a benchmark for anyone else's account, and the $1,200 to $1,700 figure is a ceiling computed off my bid cap rather than a bill I've paid. Finally, I haven't proven the landing page causes the low lead rate. What I have is a below-average post-click score on every scored keyword, an above-average creative score on the same keywords, and one page absorbing more spend than the other while producing nothing. That's where I'm looking next, not something I've already fixed. ### One thing to do next Open your campaigns view, add Search Lost IS (budget) and Search Lost IS (rank), and read the two columns for the last seven days against the seven before that. If rank has taken over from budget, the next number you need is the first-page bid estimate, and the number after that is your own measured close rate. If you want somebody to read the account settings underneath all of that, the ten I check first are in [the free 25-page Google Ads Setup Checklist](/audit/), or [request a free ads audit](/audit-request/) and I’ll read the account myself. ### Keep going If this hit, the next two pieces in the same universe: - **[The field your Google Ads audit is reading is the wrong one](/blog/primary-for-goal-is-a-legacy-field/)**. Same account, twelve days earlier, and a wrong finding I published against myself. - **[How ad fraud and click bots affect your budget](/wasted-ad-spend/ad-fraud-click-bots-budget-impact/)**. The opposite failure: a bid strategy buying all the volume it can, and none of it human. Free PDF: **[The 25-page Google Ads Setup Checklist](/freebies/google-ads-setup-audit.pdf)**. No email gate. ## Shopify Agentic Storefronts Run on Catalog Data URL: https://connercrowe.com/blog/agentic-storefronts-catalog-not-content/ Published: 2026-08-25 | Updated: 2026-10-22 Agentic storefronts are on by default for eligible Shopify stores. The catalog data layer decides which products get surfaced. What to check this week. ### Quick Take Shopify's agentic storefronts push your products into ChatGPT, Google AI Mode and Gemini, Microsoft Copilot, and Meta. Shopify's own documentation says the feature "is active by default for eligible stores," so for most merchants this is already running rather than pending a decision. The part worth internalizing is where the leverage sits. Products reach those surfaces through Shopify Catalog, which means the thing determining whether an agent can recommend you is your product data, not the answer pages and blog posts most brands are producing for AI search. I have spent a lot of the last year on content built to get cited by AI engines. It works for questions. It doesn't put a product in a shopping answer. ### It is already on Two facts from Shopify's help documentation are worth checking against your own store today rather than taking from me. The feature is on by default where you qualify. Shopify's wording is that agentic storefronts "is active by default for eligible stores," with merchants notified as availability expands. Google AI Mode and Gemini were still described as early access and not available to every store at the time I checked. Participation carries its own terms. Shopify requires you to read and agree to the Agentic Storefronts Supplemental Terms of Service, which is a separate document from the main terms. If nobody at your company has read it, that's the gap to close before you optimize anything. Go look at your sales channel settings and confirm what state you're in. This post is about what to do next, and that answer changes depending on whether you're live, eligible but not enrolled, or outside the eligibility window. ### What Shopify says makes a product legible to an agent This is the part I would not guess at, so here is what Shopify's own guidance names, in its words where I can quote it. **Structure that a machine can parse.** The goal is "structured, machine-parsable, real-time product information" an agent can "directly query, interpret, and act upon." Information that only exists inside a Liquid template or a JavaScript rendering path isn't information as far as an agent is concerned. **Pricing and inventory that are true right now.** Shopify's framing is that these "need to be accurate at the moment of the shopper's query," not at the moment something last crawled your site. This is the requirement that quietly rules out a lot of catalog setups. **Variants grouped under one parent.** "Group genuine variants under a single parent so an agent understands that it's one product with options." Stores that split colorways or sizes into separate products read as separate products. **Taxonomy specific enough to match a real query.** Shopify's example is worth copying exactly: use "men's insulated winter boots," not "footwear." Most catalogs I open are sitting at the footwear end of that. **Literal language instead of marketing copy.** The instruction is to "use literal, descriptive language in product fields rather than marketing copy." The adjective stack you wrote to sell the product to a human is working against you here. Notice what is missing from that list. There's no mention of blog volume, no mention of an answer-page library, no content-marketing lever at all. ### Where the effort usually goes instead Almost every brand I talk to about [AI search](/blog/six-ai-search-audits-crawler-access/) is thinking about it as a content problem. Write the guides, structure the answers, get cited. I run that play on my own site and I believe in it for a specific job, which is getting picked up when somebody asks a question. A product recommendation is a different transaction. When a shopper asks an agent for a 36 inch vanity under a thousand dollars, the agent isn't reading your buying guide and forming an opinion. It's querying a catalog. If your product type says "bathroom" instead of "36 inch bathroom vanity," you're not in the result set, and no amount of long-form content moves you into it. The uncomfortable version: the work that gets you cited and the work that gets you bought are two different projects, and most brands are funding only the first one. ### The trap that costs the most time Here is one I hit on a real catalog and would not have predicted. Shopify collection rules match on tags, title, or product type. They don't read your description text. That matters because the instinct is to fix agentic readiness by writing better descriptions, which is the most visible and most satisfying thing to improve. On a catalog where the differentiating attributes lived only in prose, none of that work was reachable by a collection rule, so it couldn't be used to build the structured groupings that make a catalog navigable. The attributes had to be promoted into tags and product types first. Descriptions came after, and they mattered less than the taxonomy work. The order is taxonomy, then tags, then copy. Doing it in the other order feels productive and moves nothing. ### What to check this week Open your store and answer five questions. None of this needs a developer. Confirm whether agentic storefronts is active in your sales channel settings, and whether anyone has agreed to the supplemental terms. Then pull ten products at random and read their product type field. If the values are category words rather than the phrase a shopper would type, that's your first project. Check whether your variants are grouped under one parent or split into separate products. Check whether your pricing and inventory are accurate in the catalog right now, not merely accurate on the page a crawler saw last week. And read one product description aloud. If it's selling rather than describing, rewrite it as a spec and see whether you lost anything a buyer needed. ### What I am not claiming I'm not claiming a ranking formula. Shopify publishes what makes product data agentic-ready, and it doesn't publish how the agents on those surfaces weight anything, so anyone selling you a ranking model for ChatGPT shopping is guessing. I'm also not claiming that content for AI search is wasted. It's the right investment for question intent, and I keep making it. The claim is narrower: it doesn't substitute for the catalog work, and if you have to pick one, the catalog is the one attached to a transaction. The requirements above come from Shopify's own documentation as it stood in August 2026, and this area is moving quickly enough that it's worth re-reading the source before you plan a quarter around it. If your conversion tracking isn't clean enough to tell you whether any of this worked, start with [the Tracking Stack](/frameworks/tracking-stack/) instead. If you want your own store read the same way, [request the free catalog audit](/audit-request/?type=catalog). For a home or furniture catalog I read the public store, so there's nothing to grant, and it's back in writing within 24 hours. Any other store gets [the free ads audit](/audit-request/) from read-only access. No call either way. If you'd rather see a catalog I built first, [Emma Grace Home is live](https://emmagracehome.com/): 150-plus products, every product image rendered in-house, and [the build is written up here](/results/150-sku-furniture-catalog-no-photographer/). To see what AI assistants say about your store by name, the free [What AI says about you](/what-ai-says-about-you/) template has the questions and a fact ledger. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Your Google Ads Audit Is Reading the Wrong Conversion Field URL: https://connercrowe.com/blog/primary-for-goal-is-a-legacy-field/ Published: 2026-08-25 | Updated: 2026-10-22 conversion_action.primary_for_goal is legacy and makes a clean account look broken. The conversion-goal layer decides what Smart Bidding buys. ### Quick Take If you're auditing what a Google Ads account bids toward, don't read `conversion_action.primary_for_goal`. It's a legacy field. An account can report `primary_for_goal: true` on actions that are not biddable at all, which makes a correctly configured account look broken. I know because I did exactly this on my own account in August and produced a wrong finding: I flagged two Google Business Profile local actions as poisoning the bidding signal when the goal layer had already excluded them. The setting that governs what Smart Bidding optimizes toward is the conversion-goal layer, and it has two levels that have to be read together. ### What the legacy field is doing `primary_for_goal` is a per-conversion-action boolean, and it reads like the answer to "does this count." It used to be closer to that. The account structure moved on and the field stayed, so what you're reading now is a flag that no longer expresses the thing you want to know. The specific failure is that it can't express a campaign-level override. A conversion goal can be biddable at the account level and switched off for one campaign, and no per-action boolean can represent that, because the state doesn't live on the action. ### The two levels that do decide Read both of these. Neither one alone is the answer. ``` SELECT customer_conversion_goal.category, customer_conversion_goal.origin, customer_conversion_goal.biddable FROM customer_conversion_goal ``` ``` SELECT campaign.name, campaign_conversion_goal.category, campaign_conversion_goal.origin, campaign_conversion_goal.biddable FROM campaign_conversion_goal WHERE campaign.id = ``` The first gives you the account default. The second gives you what a specific campaign actually does, including where it departs from that default. Resource names come back as `CATEGORY~ORIGIN`, so you will see values like `GET_DIRECTIONS~GOOGLE_HOSTED` and `SUBMIT_LEAD_FORM~WEBSITE`. Origin matters as much as category, because the same category arriving from a Google-hosted surface and from your website are different rows with different settings. ### The gotcha that will invert your reading `biddable` is omitted from the response when it's false. The API drops false and zero values rather than returning them, so a row with no `biddable` key means not biddable. If you parse this assuming a missing key is an error or a null to skip, you will either crash or, worse, silently treat non-biddable goals as unknown and drop them from your analysis. Read it as `.get('biddable', False)` and move on. This is the same omission behavior that bites elsewhere in the Google Ads API, and it's the single most common reason a script's picture of an account disagrees with the interface. The write side is worse. A field mask built from your object drops any value equal to its type's default, and the mutate returns success having changed nothing: [the fields that vanish](/blog/fields-you-most-want-to-change-vanish/). ### What still deserves a flag Getting the layer right doesn't mean the account is fine. It means you now have to check something harder, and `primary_for_goal` was never going to tell you this either. A goal can be correctly biddable while the conversion action underneath it fires on the wrong event. On my own account, `BOOK_APPOINTMENT~WEBSITE` is biddable, which is right. The action feeding it is "Call booked (Calendly)," which sounds right. It fires on a GTM trigger called `book_call_click`, which is the scheduling modal opening. Not a booking. Somebody who opened the modal and closed it counts the same as somebody who put a meeting on my calendar. Under Manual CPC that's a reporting annoyance. Under Smart Bidding it's a direction, and the system will go find more people who open modals. So the audit step is: for every biddable goal, open the GTM trigger behind the conversion action and confirm the event is the outcome, not the intent to start it. Names lie. `book_call_click` is honest about what it is if you read it, and the conversion action's display name is what hides it. ### What to do with this Three passes, in order. Pull `customer_conversion_goal` and write down which categories are biddable. Then pull `campaign_conversion_goal` for every enabled campaign and note where it diverges, because those overrides are usually deliberate and occasionally forgotten. Then, for each biddable goal, trace the action to its trigger and decide whether that trigger is a completed outcome. If the third pass turns up a soft trigger on a biddable goal, that's your finding, and it's a real one. It's also the finding I would have missed entirely if I had kept reading the legacy field, because the legacy field had me looking at two harmless local actions instead. ### What I am not claiming I'm not claiming `primary_for_goal` is meaningless everywhere. It still exists, older tooling still reads it, and in accounts that never diverged from account-level defaults it will often agree with the goal layer. The claim is that it's not a reliable basis for an audit conclusion, and agreement in the simple case is not evidence in the complicated one. I'm also not claiming this is the whole measurement picture. It tells you what bidding is aimed at. Whether the thing it's aimed at is being counted correctly is a separate question, and the architecture I work through for that is [the Tracking Stack](/frameworks/tracking-stack/). If you want the numbers behind how far apart platforms land on the same event, that's in [the Ecommerce Tracking Accuracy Benchmark](/benchmarks/ecommerce-tracking-accuracy/). When I audited [Emma Grace Home](https://emmagracehome.com/), its Google Ads was [optimizing toward add-to-carts](/results/unassigned-revenue-dual-ga4-fix/) instead of buyers. If you're not sure which of your conversion actions are actually biddable, that's one of the first things I check in [the free ads audit](/audit-request/). Send me read-only access and the write-up is back within 24 hours. No call. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## Q4 Carried the Most Volume and the Worst ROAS URL: https://connercrowe.com/blog/q4-roas-trough-not-peak/ Published: 2026-08-25 | Updated: 2026-10-22 Three years of purchase-only data from one account: the fourth quarter carried the most volume and the worst return on ad spend, both years I could measure. ### Quick Take On one ecommerce account I manage, I pulled three years of Google Ads performance computed on purchases only, with cart and checkout events stripped out of the value column. The fourth quarter came in below the full-year average both years I could measure it. Q4 2024 returned 4.80 against a 5.59 year. Q4 2025 returned 3.17 against a 3.54 year. That's 14% under the year in the first case and 10% under in the second, in the quarter that carries the most volume and gets planned around hardest. This is one account, N = 1, across two fourth quarters. It's enough that I now run the same check on every account before anyone writes a Q4 plan. ### Purchase-only is the whole trick Most accounts don't have a Conversions column made of purchases. When I measured this directly for the [Ecommerce Tracking Accuracy Benchmark](/benchmarks/ecommerce-tracking-accuracy/), two of the three accounts I could read carried add-to-cart inside the Conversions column at 16.6% to 17.2% of the total, and 19.8% to 25.8% of reported conversion value was cart value rather than revenue. The column still said Conv. value. At the far end of that, [Emma Grace Home's Google Ads was optimizing toward add-to-carts](/results/unassigned-revenue-dual-ga4-fix/) instead of buyers when I audited it. Cart events don't spread evenly across a year. More people add to cart and walk away in December than in June, because more people are browsing, comparing and price-checking in December. So the contamination concentrates in exactly the quarter you're trying to evaluate. A Q4 ROAS built on a value column that includes cart value reads high, and it reads highest in the quarter where abandonment peaks. The quarter looks like a peak because the measurement inflates most where the browsing is heaviest. ### The two fourth quarters Same account, same conversion action, purchases only. | Window | ROAS | Against the full year | Gap | |---|---|---|---| | Q4 2024 | 4.80 | 5.59 | 14.1% under | | Q4 2025 | 3.17 | 3.54 | 10.5% under | There's a second way to see it that doesn't depend on my quarterly cut being right. January through August, purchase-only, came in at 6.10 for 2024 and 3.70 for 2025. The full years landed at 5.59 and 3.54. The only thing that can pull a year below its own first eight months is what happened in the last four. ### The decline underneath the season The seasonal dip sits on top of a longer slide, and the two have different causes. January through August, like for like: | Year | ROAS | CPC vs prior year | Revenue per click vs prior year | |---|---|---|---| | 2024 | 6.10 | | | | 2025 | 3.70 | 1.77x | 1.07x | | 2026 | 3.63 | 1.94x | 1.90x | The 2024 to 2025 step is the entire collapse. Click costs went up 1.77x while revenue per click moved 1.07x, so efficiency had nowhere to go. From 2025 to 2026 both moved together and the ratio held, which is the healthier picture even though the absolute number is lower. One caveat on that last row: the 2025 to 2026 revenue-per-click jump is partly a tracking change on this account and reads closer to 1.55x once adjusted, so treat the direction as solid and the magnitude as soft. This separates two questions people usually merge. Seasonality is one. A structural change in what a click costs is another. If your year-over-year ROAS is down, run the CPC and revenue-per-click multiples before you conclude the algorithm forgot how to work. ### What I do differently because of this I stopped setting annual targets off a fourth quarter. The target comes off a full year, or off Q1 through Q3, and the fourth quarter gets planned as the period where I accept lower efficiency in exchange for volume. I still freeze structural changes before October. The reason changed. It's not about protecting a peak, it's that you can't read the effect of a campaign restructure against a baseline that is moving underneath you for seasonal reasons. And on a thin-margin catalog, Q4 became the quarter where I check whether the volume is still profitable at that efficiency, rather than assuming the quarter carries the year. ### Check yours in about twenty minutes Open Google Ads, go to the conversion actions view, and confirm which actions are inside the Conversions column. If add-to-cart or begin-checkout are in there carrying value, every historical ROAS figure you have is inflated by an unknown amount and the rest of this exercise won't work until you isolate the purchase action. Then pull ROAS by quarter for three years on the purchase action alone, and put each quarter next to its own full-year figure. You're looking for one thing: whether your fourth quarters sit above or below their years. If they sit below, your planning assumption is upside down, and the number to fix first is the target you carried into the quarter. The full architecture I rebuild accounts on, including how to get a single trustworthy purchase signal, is [the Tracking Stack](/frameworks/tracking-stack/). ### What I am not claiming This is one account and two fourth quarters. It's not a category rate, and I would not put it on a slide as an industry benchmark. One catalog, one competitive set, one set of margins. I'm also not claiming Q4 is the wrong quarter to spend into. Lower efficiency at much higher volume can be the right trade when the margin math works and the customers come back. The finding tells you what the number is likely to be. Whether to spend into it is a separate decision, and it's a margin question rather than a ROAS one. What I'm claiming is narrow and testable: on the account I could measure across three years, the quarter everybody plans around returned less than the year it sat inside, both times, and the reported figures said otherwise until the cart events came out. If Q4 is coming and you can't tell which of last year's numbers were real, [the free ads audit](/audit-request/) reads the account from read-only access. I write it myself and it's back within 24 hours. No call. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## The Foreign Hostname (Why I Check Where My Analytics Data Came From) URL: https://connercrowe.com/blog/foreign-hostname-analytics-check/ Published: 2026-08-17 | Updated: 2026-10-22 Four page paths in my analytics project have never existed on my site. A pitch mockup carried the same public key. The provenance check that finds it. On 11 August I opened the analytics project behind connercrowe.com and found four page paths that have never existed on this site. They had logged about 136 pageviews over the trailing ninety days, filed in the same project as my blog posts and my case studies. They came from a pitch mockup I had deployed on a separate domain, which shipped carrying the same public analytics key as the site. ### Quick Take An analytics property is scoped to a key, not to a site, and that key is public by design. Any page that carries it writes into the property, and the hostname is recorded on every event without anything in the default view marking it as foreign. So the page list in your analytics isn't an inventory of your site. It's an inventory of everywhere your key is installed, and I learned the difference because an audit had already built a recommendation on four pages that weren't part of the site it was auditing. ### The key is public on purpose PostHog documents the project key, the one starting `phc_`, as a public client-side key that is safe to expose in web and mobile code. It can't reach your stored data; it sends events and evaluates feature flags, and that's the whole of it. GA4's measurement ID is the same shape of thing, sitting in the page source of every site that uses it. Neither is a secret and neither can be, because the browser has to send the event and the browser holds nothing the reader can't open and read. What follows from that is the part I hadn't thought through. If the key is the credential, the origin is not. The hostname rides along as a property on the event, recorded rather than enforced. A browser collection endpoint holding a valid key has no reason to check whether the page that sent it belongs to you, and doesn't. The vendors do gate the server-side path, which makes the contrast worth seeing. GA4's Measurement Protocol requires a private `api_secret` alongside the measurement ID, and Google's own documentation says to keep it out of client-side code because exposing it lets outside parties post spam into the property. The server door has a lock on it. The browser door is the one standing open, and it has to be, because that's where the measurement happens. ### How it reached my own reports I build navigable mockups to pitch website rebuilds. Each goes up as its own Cloudflare Pages project on its own domain, and one of them started life as a copy of a layout that already carried my analytics snippet. Five of its pages shipped with the key. Four show up in reporting. Then I ran a content audit on my own site, and the audit read the analytics project as the record of what exists on connercrowe.com. It found four thin pages with low engagement, correctly noted they were weak, and produced a consolidation recommendation for them. Every step of that reasoning was sound and the premise underneath it was false. Those pages were never part of the site the audit was about. The 136 pageviews are the trivial part. The audit conclusion is the expensive part, and I found it only because I went looking for the pages in my own repository and couldn't find them. ### Where this hides on a Shopify brand Pitch mockups are my version of the problem. They aren't the common one. A brand accumulates the same foreign hostnames through ordinary work: a preview theme shared for review, a staging store, a `myshopify.com` domain still serving the live theme alongside the custom domain, a wholesale or trade portal on a subdomain, a regional storefront, a landing page a paid-media vendor built and deployed somewhere you never see. Each is a page carrying your key on a hostname that isn't the one you think you're measuring, and each contributes sessions to the same property. The `myshopify.com` case is the one worth checking first, because it costs nothing to have and it's on by default. ### The check, and the fix that follows it The check is a hostname breakdown, not a code search. Open the property, break pageviews out by hostname, and read the list. That's what surfaced this, and it's the only step that works regardless of where the offending page is deployed or who deployed it. A local code search finds nothing if the project was a direct upload rather than a repository, if it has since been deleted from disk, or if the key sits in a file type outside whatever pattern you searched. Once the hostname tells you a foreign property exists, the code search tells you which one, and the fix is to strip the snippet from the source and from the built output, redeploy, then fetch each affected page and confirm the string is gone. A stale build directory will republish what you deleted from source. ```bash # Find which local project shipped the key. grep -rl "phc_" ~/Projects --include="*.html" --include="*.js" --include="*.astro" # Then confirm every page that carried it, the root included. for p in / /page-one/ /page-two/; do echo -n "$p " curl -s "https://your-mockup.pages.dev$p" | grep -c "phc_" || true # expect 0 done ``` The reporting side doesn't clean up after itself. Events already collected stay collected, so every report I run that crosses that ninety-day window carries a hostname filter, and any figure I publish from it says so. Making that permanent in GA4 is more awkward than it sounds. Google documents three data filter types: developer traffic, internal traffic, and web hostname traffic. The hostname filter excludes domains you name rather than admitting the ones you own, so the job of knowing which hostnames are yours stays with you. ### What it costs to skip The direct cost is an audit that reasons correctly from a false inventory. Consolidation plans and redirect maps both start from a list of pages, and if that list comes from analytics rather than from the repository, it can contain pages that were never on the site. The quieter cost lands on rates. A conversion rate is a fraction with sessions on the bottom, and so is every engagement benchmark built the same way. Foreign sessions inflate that denominator with visitors who arrived somewhere else with different intent, so the resulting figure is wrong in a direction the figure itself won't reveal. Hand it to a bidding strategy or print it in a report and the error travels. This belongs in front of [the STACK Audit](/blog/google-ads-tracking-audit-guide/), the five passes I run before letting Smart Bidding optimize against numbers I haven't verified. Those passes check whether the signal is emitted, triggered, attributed and valued correctly. None of them ask whether the property is describing the site you think it is, and neither do the eight layers of [The Tracking Stack](/frameworks/tracking-stack/). This is a pre-flight on the data you audit with, ahead of the work of auditing what it says. It's the same family of check as [reading the ad destination before the campaign spends](/blog/preview-flag-ad-destination-launch-check/): a silent failure with no error state, sitting upstream of everything downstream that trusts it. ### The receipts I found and fixed this on August 11, 2026. The mockup was a direct upload to Cloudflare Pages rather than a git project, which is why a repository sweep alone would have missed it and why the hostname breakdown is the step that matters. I stripped the snippet from the project root and from the built output, redeployed, then fetched the live pages and confirmed them clean. I checked nineteen other mockup and landing-page projects on the same machine. All nineteen were clean, so the leak was confined to the one. The historical events remain in the property and I handle them with a hostname filter rather than deleting them. ### What I did not claim I haven't measured what the mispremised audit would have cost, because I found the premise before acting on the recommendation. I'm not claiming a ranking or revenue effect, and there's no before-and-after here to show. The pageview figure is approximate and I've deliberately not set it against this site's own traffic, since that comparison would imply a materiality I didn't measure. I also didn't have a stranger write into my property. Both properties were mine, deployed by me, from one machine. That a public key permits an unrelated site to write into a property is documented mechanism rather than something I observed here, and I'm reporting it as the former. What this account demonstrates is narrower and more common: a property is scoped to a key, so its page list can include pages that aren't on the site you're auditing. Finally, I haven't tested every analytics vendor. What is documented is what is written above, for PostHog and GA4. Treat the rest as unknown until you have read their documentation. ### One thing to do next Break your analytics property out by hostname and read the list. If a domain you don't recognize appears, find the page carrying your key before you trust another report built on that property. That part is free and takes minutes. What it won't tell you is which of your reports, dashboards and bidding decisions have been standing on the polluted number, or for how long. If you want me to look, [send me read-only access for the free ads audit](/audit-request/) and I'll trace it in writing within 24 hours. ### Keep going If this hit, the next two pieces in the same universe: - **[The Furniture Brand Whose Revenue Showed Up as Unassigned](/results/unassigned-revenue-dual-ga4-fix/)**. Two installs fighting over one property on Emma Grace Home. Count your installs, then count the hostnames they report from. - **[The Click Text Trigger Trap](/blog/click-text-trigger-trap/)**. The same audit turned on my own site, seven leaks in one container. Free PDF: **[The 25-page Tracking Stack](/freebies/tracking-stack.pdf)**. No email gate. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## The Preview Flag I Check Before Any Campaign Spends URL: https://connercrowe.com/blog/preview-flag-ad-destination-launch-check/ Published: 2026-08-11 | Updated: 2026-10-22 A landing page can be live, fast and finished while recording nothing. One meta tag decides it. The six-step check I run before any campaign spends. In early August 2026 a landing page went live on a client's ad destination with one meta tag set to `preview` instead of `production`. The page loaded. It looked finished. The form accepted a submission and returned a success panel. Underneath, the form never called the API, the analytics tags never loaded, and every lead would have died with the browser tab. The campaign was enabled against that URL. I caught it before a single click was paid for. ### Quick Take Landing pages built for paid traffic usually carry a switch that separates a preview build from a live one. When that switch is wrong, the page still works in every way a person can see. It renders, it takes a form, it says thank you. What it stops doing is recording. So I verify the destination before the campaign spends rather than after the first report comes back empty. Five of the six checks run from a terminal in under two minutes. The sixth needs a person. ### What the flag does The page carries a single line in its head: ```html ``` The page reads that value at runtime and gates everything measurable on it. Set to `production`, the form posts to the lead API and the GA4 and Google Ads tags load and fire. Set to anything else, the form short-circuits to a "Preview complete" panel and never reaches the network call. The tags load nothing at all. The gate is one line of JavaScript, and its bluntness is the point: ```js if (mode !== 'production') { show(LAST + 1); return; } // never reaches fetch() ``` That design is correct. A preview build should be incapable of writing a real lead into a client's CRM, and it should be incapable of polluting a conversion action with test submissions. The flag exists so a staging copy can be circulated for review without any of it counting. The failure mode is that the same property which makes preview safe makes it silent. There's no error state. Nothing turns red. A page in preview mode and a page in production mode are visually identical right through the thank-you screen. The only way to tell them apart is to read the tag or watch the network. ### The morning I found it The build was a paid-only page for a custom furniture maker taking commission work. Static assets on Cloudflare Pages, two Pages Functions behind them, leads flowing to WhatConverts. I was doing a pre-launch pass on the live URL, not the preview URL, and I pulled the raw HTML rather than looking at the rendered page. That's the only reason I saw it. `deployment-mode` read `preview`. The Search campaign pointed at that exact URL and was already enabled. Two other things surfaced in the same pass, both from the same deploy. The first: the repo root had been deployed instead of the `public/` directory, so `README.md` and `.gitignore` were being served at the site origin and both returned HTTP 200 on a live client site. The second: `wrangler` resolves the `functions/` directory relative to the working directory rather than the asset directory you hand it, so a deploy run from the wrong folder ships the static assets without the Functions. The tell is that `/api/lead` starts answering `404` where it should answer `405`. Then a fourth problem, which is the one worth remembering. The apex domain runs on Google Cloud DNS, so it's not a Cloudflare zone. No caching panel, no zone id, no `purge_cache` call. Deleting the stray `README.md` didn't evict it. The edge kept serving the deleted file with `CF-Cache-Status: HIT` under a seven-day TTL. The fix was to overwrite the path with new content so the ETag changed, which replaced the cached response immediately. ### The check that runs before the campaign This is now a script rather than a memory. It runs against the live URL, never the preview URL, and it fails loudly: 1. **Read `deployment-mode` from the served HTML.** Not the local file, not the preview origin. Fetch the URL the ad points at. 2. **Confirm both tag IDs and the conversion label are present** in the served markup, and that the bot-protection site key is there too. 3. **Confirm the lead endpoint shipped.** `GET /api/lead` should answer `405`. A `404` means the Functions didn't deploy and the form has nowhere to post. 4. **Confirm the production secrets exist on that environment.** Preview and production are separate environments, and secrets set on one don't exist on the other. POST deliberately malformed JSON to the endpoint. `invalid_json` means the secrets are wired. `lead_system_not_configured` means they're missing and the handler is failing closed. 5. **Check every anchor the ads reference.** Sitelinks and price extensions point at fragments. A renamed section turns a sitelink into a scroll to nowhere. 6. **Submit one real lead by hand.** Confirm it lands in the lead system, that the notification email arrives, and that a conversion records against the right action. No script can do this part. Bot protection is enforced server-side, which is the whole point, so a human has to be the one who proves the path works end to end. Steps one through five are automated and take seconds. Step six is the only one that proves anything, and it's the one most launches skip. ### What skipping it costs The cost isn't a bad campaign. It's an invisible one. Traffic arrives. Clicks bill. People fill out the form and see a confirmation, so nobody complains. The leads never reach the CRM and the conversions never reach the ad platform, which means Smart Bidding is optimizing against an empty signal and steering budget by nothing at all. Every day it runs, the bidding model learns from noise. The report at the end of the month shows spend and clicks and zero conversions, and the conversation that follows is about whether the offer is wrong or the keywords are wrong. That's a hard hole to climb out of, because the account now has a stretch of history that says the traffic doesn't convert, and none of it is true. I wrote about that exact symptom in [clicks with no conversions](/wasted-ad-spend/do-more-clicks-mean-more-sales-google-ads/). A misconfigured destination produces the same chart as a genuinely bad campaign. ### The receipts The flag was found on August 5, 2026 on a live Cloudflare Pages deployment with the Search campaign already enabled. The stray `README.md` at the origin returned HTTP 200 and kept serving from cache under `s-maxage=604800` after deletion, which is what forced the overwrite-instead-of-delete fix. `/api/lead` returned `404` until the second deploy shipped the Functions bundle, then `405` as expected. The verification script now checks the deployment flag, both tag IDs, the conversion label, the bot-protection key, all seven sitelink and price anchors, and the endpoint status on every relaunch. Conversions on that page fire only after the API returns a lead id, so the conversion count can never exceed the leads delivered. The measurement architecture underneath is the same one in [the Tracking Stack](/frameworks/tracking-stack/). The deploy-pipeline version of this problem, where the build itself is what drops the flag, is in [Four Ways a Site Launch Lies to You](/blog/four-ways-a-site-launch-lies-to-you/). ### One thing to do next Pull the HTML of the page your ads point at right now and read it. Not the rendered page. The source. If you want the longer version of that pass, the [Google Ads Setup Checklist](/audit/) is the twenty-five page checklist I run on accounts before I touch anything, and it's free with no email required. If you're about to launch a site or a campaign and want a second read before it spends, [send me read-only access for the free ads audit](/audit-request/). It comes back in writing within 24 hours. No call. ## The Shopify Event With No Subscriber (What I Found Auditing My Own Build) URL: https://connercrowe.com/blog/shopify-custom-event-no-subscriber-google-ads-conversion/ Published: 2026-08-03 | Updated: 2026-10-22 A Shopify theme snippet published a lead event nothing subscribed to. The conversion was Primary, counted, and recording zero. The audit and rewire. I ran a tracking audit on a Shopify storefront I had built, against the live API and the rendered page rather than against my own build notes. The primary lead conversion was in the account, marked Primary, counting. Its conversion label appeared nowhere on the storefront. The form fired an event on every successful submission. Nothing was listening. No paid traffic had run through it yet, which is the only reason this is a post about a defect instead of a post about a wasted budget. ### Quick Take Shopify lets a theme file publish a custom event with `Shopify.analytics.publish()`. It doesn't require anything to subscribe to that event. If no pixel subscribes, the call succeeds, the browser reports no error, and the event evaporates. My snippet published on a successful form post and waited for a web pixel to forward it to Google Ads. Neither of the two app pixels installed on the store had any reason to listen for a string I had invented. Check that something is listening before you trust the thing that sends. A publisher with no subscriber is not tracking. ### How the Gap Opens Shopify's Web Pixels API has two halves that are documented separately and shipped separately. The publish half lives in the theme. From a Liquid file you can call `Shopify.analytics.publish('my_store:event_name', event_data)`, and the platform accepts it. That call is fire and forget. It doesn't return a count of listeners, and it doesn't warn you when the count is zero. The subscribe half lives somewhere else entirely, in the admin under Customer events, where a pixel calls `analytics.subscribe('my_app:my_custom_event', callback)` and receives your payload on the `customData` field. Nothing binds the two halves at build time. They get matched by string at runtime, inside a sandbox. From the theme's side a miss looks exactly like a hit. It's a quieter cousin of the mismatch I describe in [why Shopify conversion tracking stops working](/conversion-tracking/shopify-conversion-tracking-not-working/), where a pixel emits `add_to_cart` and a Tag Manager trigger listens for `addToCart`. In that version a consumer exists and the string is wrong. In mine no consumer had been written at all. The theme half of the work looks finished either way, and both fail without an error. I should say why I reached for the sandbox at all, because that's where the defect starts. The lead form is a theme page, not checkout. It never needed the Web Pixels sandbox. I wired it that way out of habit from checkout work, where the sandbox is the only option, which is the scope I set out in [what Checkout Extensibility breaks](/conversion-tracking/shopify-checkout-extensibility-tracking/). ### The Audit Path I stopped reading my own build notes and searched the rendered storefront for the conversion label string. Zero occurrences. The label existed in Google Ads and existed nowhere in the HTML the browser receives. Then I opened the snippet. It fired on a successful post, exactly as designed, and published a neutral custom event. Then Shopify admin → Settings → Customer events. Two app pixels listed. One does session recording and the other fires its own standard ecommerce events, so neither had any reason to subscribe to a custom event name I had made up, and neither carried the conversion label it would have needed to forward it correctly if it had. The same pass turned up two more conversion actions counting nothing, one duplicating a purchase that already had a working action and one pointed at a page no visitor could reach. Both went to Secondary, on the source-emission and tag-trigger layers of [the STACK Audit](/blog/google-ads-tracking-audit-guide/) that I run on every inherited account and had not yet run on my own. The pattern across all three is the same one I found in [my own Tag Manager container in May 2026](/blog/click-text-trigger-trap/). Primary, in the Google Ads interface, describes a row in Google's database. It says nothing about whether the label exists in your HTML. My pre-launch sweep on this build covered copy and schema, and measurement wasn't in that pass, which is how a dead conversion cleared a checklist I wrote myself. ### The Rewire, and Why I Did Not Add a Tag The fix was to stop routing a theme-page conversion through the pixel sandbox and fire it on the page. Google documents this directly. Its [custom pixel limitations page](https://support.google.com/google-ads/answer/16000892) says that running Google tags inside Shopify's custom pixel feature is not a supported implementation, and names what may not work correctly from that sandbox, including enhanced conversions, phone call conversions, consent mode URL passthrough, cross-domain measurement, conversion verification and troubleshooting, and setting up conversions with a URL. Enhanced conversions matter most for a lead form, because the hashed email the form already collects is the whole reason to collect it. Google's recommendation for sitewide measurement is its own Google & YouTube app. Before writing anything I measured what the page already had. On this store the app had injected gtag into the main document and configured the Ads conversion ID with `send_page_view: false`, and the Google click identifier cookie was already being set first party. A second Google tag on top of that would have fought the app and risked double counting, which is the failure I would have created while fixing the first one. So the snippet stopped publishing into the sandbox and started firing through the queue that was already there: ```js // Before: published into the Web Pixels sandbox, where nothing subscribed. Shopify.analytics.publish('my_store:enquiry_submitted', { value: 1 }); // After: fire on the page, reusing the gtag the Google & YouTube app installed. window.dataLayer = window.dataLayer || []; window.gtag = window.gtag || function () { window.dataLayer.push(arguments); }; gtag('event', 'conversion', { send_to: 'AW-XXXXXXXXXXX/YourConversionLabel' }); ``` The `||` on that middle line is the part that matters. It reuses the app's gtag if one is present and only defines a shim when it isn't, so the store still runs one Google tag rather than two fighting over the same conversion. I verified it the way I verify any of these. Fire a real submission, watch the network, confirm the request to `googleadservices.com/pagead/conversion/` leaves the browser carrying the conversion label. A 200 on that beacon is the first evidence that any of it works. Before that, all I have is my own build notes, which is what put me here in the first place. ### What This Shape of Defect Costs An account in this state doesn't look broken. The conversion action reads Primary and counted. The pixels are listed under Customer events. The storefront serves fine. Every surface a person checks in a hurry is green. Then a budget goes live on an account in this state, and Smart Bidding optimizes toward a conversion it will never observe. Maximize conversions with no observable conversions spends the budget on whatever proxy the auction offers, and the account's own reporting agrees with the bid strategy that nothing is happening. The damage compounds, because the first read of flat conversion volume is usually a media problem, so the next month goes to keywords and creative while the measurement layer stays dead underneath. [The contractor whose only website conversion had been dead since somebody renamed a form button](/results/dead-form-conversion-fix/) is the version of this that ran for months before anyone found it. If you're the founder paying for this rather than the person building it, the question to ask whoever owns your stack is not whether tracking is set up. It's what evidence they have that the conversion left the browser, and when they last looked. A screenshot of a green conversion row is not that evidence. A network capture of the beacon is. ### What I did not claim Any recovery. No paid traffic has run through this account, so there's no lost-conversion figure, no before-and-after, and no performance result to point at. I can't tell you what the defect would have cost and I'm not going to estimate it. I also didn't claim the account is clean now. Enhanced conversions for leads isn't live end to end. Two items on my own audit list are marked unverified rather than passed, for the access reason below. What is claimed is narrow and checkable: a Shopify theme file can publish a custom event that nothing subscribes to, the platform raises no error when that happens, and the conversion will read Primary and counted in Google Ads while recording zero. The defect was in my own build, it survived my own build notes, and it took reading the page instead of the notes to find it. ### The Receipts The audit ran through the Google Ads API and against the rendered storefront rather than through the Ads interface, because the browser profile signed in at the time was the wrong account. That's a real limitation and it's why two items on the list, the auto-apply recommendation settings and the GA4 key event mappings, are marked unverified rather than passed. Fixed and confirmed: the lead conversion beacon reaches `googleadservices.com/pagead/conversion/` carrying its label. Applied and re-read after writing: two conversion actions demoted from Primary to Secondary, leaving one Primary purchase and one Primary lead. Enhanced conversions for leads is not yet live end to end, and I'm not claiming it. One correction to my own first pass, since it belongs here. I flagged an Add Payment Info action as miscategorized and withdrew it before applying anything. There's no add-payment-info value in the conversion category enum, which I confirmed by reading the enum live. The app had it right and I had it wrong. The client is unnamed by agreement. ### Keep going If this hit, the next two pieces in the same universe: - **[The Click Text Trigger Trap](/blog/click-text-trigger-trap/)**. The same class of silent failure in Tag Manager, found on my own container. A trigger that matches nothing raises no error either. - **[The Bing pixel that said YOUR_VALUE_HERE](/blog/bing-pixel-your-value-here/)**. Eleven findings inside an inherited Shopify tracking stack, including tags that were installed and firing into nothing. Open your storefront's rendered HTML and search it for your conversion label. If the string is not there, the conversion is not there, whatever the account says. The checkout-extensibility half of this, what it breaks and the order to fix it in, is at [Shopify Checkout Extensibility and tracking](/conversion-tracking/shopify-checkout-extensibility-tracking/). If a Shopify conversion in your account has gone quiet and nobody noticed, [the free ads audit](/audit-request/) is where I'd find it. Send me read-only access and it's back in writing within 24 hours. No call. Free PDF: **[The 25-page Tracking Stack](/freebies/tracking-stack.pdf)**. Layer 1 is source emission, which is the layer this defect lived on, and it covers the post-2025 Shopify checkout extensibility audit that decides which events belong in the sandbox and which do not. The layer map is on [the Tracking Stack framework page](/frameworks/tracking-stack/). No email gate. ## The Ad Manager Job Is Going Away. I Know Because It Was Mine. URL: https://connercrowe.com/blog/selling-the-system-not-the-hours/ Published: 2026-07-28 | Updated: 2026-10-03 An honest account of the system I run, what it has produced, where it breaks, and why marketing gets bought as a subscription rather than as hours. ### Quick Take Two years ago my job was logging into ad accounts and doing things by hand. Pull the search terms, add the negatives, move the bids, write the report. That job is disappearing and I would rather say it out loud than pretend it isn't happening. What replaced it, for me, is a system that carries every client I have ever worked on and gets better each time I use it. Clients don't pay me for hours anymore. They pay for access to that. Below is what it is, what it has produced, where it breaks, and the part I haven't built yet. ### The Call That Made the Point In July 2026 a furniture maker in Illinois booked a call with me. He builds custom tables, five to fifteen thousand dollars each. He had just finished ten months with an agency: twelve leads, one sale. I asked how he found me. He said ChatGPT. He had typed in a description of his business and asked for someone who would fit a company his size instead of an agency. It gave him my name. No ad. No referral. No Google search in the way that phrase meant something three years ago. I've spent about a year building content specifically so [machines would read it and recommend me](/blog/six-ai-search-audits-crawler-access/). Sixty-seven answer pages across seven libraries, starting with [the wasted ad spend library](/wasted-ad-spend/), written to be quoted by an AI rather than skimmed by a person. That call was the first time a stranger arrived through it and told me so on the record. The front door moved. The people who noticed early are getting the traffic. ### What I Used to Sell I was an ads manager. Good one. The work was real: search term reports, negative keyword sweeps, bid adjustments, monthly reporting. Every one of those tasks is now something I can hand to a machine and check. I'm not being dramatic about this. On a call I told a peer that ads managers are being replaced and that I had pivoted my whole business because of it. I meant it. If your entire value is executing a checklist inside an ad platform, the platform is coming for that job, and so is every operator who wired up a system to do it. The mistake is thinking the answer is to do the same job faster. It isn't. The job changes shape. ### What I Run I want to be specific here, because "I use AI" has become a meaningless sentence. **A memory bank.** An Obsidian vault, 232 notes as of September 2026. Every client, every project, every landing page, every mistake I've made and the rule I wrote so I would not make it twice. Plain markdown files. Nothing proprietary. I wrote about why plain files beat buying a memory product in [Your AI Agent Doesn't Need a Brain](/blog/your-ai-agent-doesnt-need-a-brain/). The notes link to each other. A landing page note for one client connects to the skill that built it, which connects to three other clients where the same approach worked. When I start work on an account, the agent reads that web first. I'm not briefing it from scratch every session. It knows how that business talks, what the owner has already rejected, and what went wrong in March. **112 skills.** Repeatable procedures written down: how to build a PMax campaign, [how to audit a Shopify storefront](/blog/how-i-audit-a-shopify-storefront/), how to verify a Google Ads change before it goes live, how to write in a specific client's voice. Each one is a file the agent loads when the task comes up. **A headless Mac Mini.** Sits under my monitor with no screen attached. It runs the recurring jobs on a timer: call scoring for a law firm every weekday evening and for a medical practice every night, weekly client reports on Monday morning, daily ad reports for one store, and an hourly heartbeat that tells me when something stopped. That's the honest inventory. Files, procedures, and a cheap computer running scheduled scripts. ### The Part Most People Get Wrong Here's where I want to correct something I've said too loosely on sales calls. When I describe this, it's easy to make it sound like an autonomous agent waking up each morning and thinking about your account. It isn't that. Most of what runs on a schedule is deterministic. Python scripts, cron jobs, a fixed pipeline. The language model gets used at specific points where judgment or writing is needed, and the output goes through a verification gate before anyone sees it. My monthly report system, for example, pulls the numbers deterministically, makes exactly one model call per client to write the narrative, and then runs a three-layer check: numbers against the cards, banned phrases, and anomaly smell tests. Anything that fails gets a REVIEW_ prefix and never reaches a client as finished work. The reason it works is not that the AI is smart. It's that the system is boring and the context is good. A model with no memory of your business writes generic garbage. The same model with two years of notes about your business writes something worth reading. Same idea I used when [an agent built my tracking stack](/blog/agent-built-my-tracking-stack/). The leverage is in the context, not the model. ### What It Has Produced Real things, with the caveats attached. In July 2026 I closed an engagement with a manufacturer who had interviewed seven agencies before picking me. Their last vendor ran ads through the vendor's own ad accounts, showed them charts of clicks, and produced one accidental purchase against a fifteen thousand dollar invoice. What closed it was not a clever deck. On the call I shared my screen and walked them through the vault: here is every note about your business, here is how it connects to other work, here is the rule I wrote after I got something wrong. Their technical guy's verdict was that it looked like I was using AI properly. That was the deal. One store I have run since last summer went from roughly $20,000 a month in orders to $49,000 within two months of the ads going live. This January through March, with Google and Meta running together on about $4,000 a month of combined ad spend, it did $87,131, then $92,365, then $116,129. I had been quoting those three months from memory on sales calls, so before writing them here I re-pulled them from the raw order records. The recall held to the dollar. And to keep this honest: the account in the next paragraph, the one whose Performance Max fell off a cliff, is the same store. The curve is real. So is the dip that followed it. Both live in the same order records. The system also catches things I would have missed. On one account in July 2026 it surfaced that Performance Max efficiency had fallen off a cliff, ROAS from roughly 14.9x down to 3.5x and cost per acquisition from about $100 to $408, while click volume went up. Traffic quality collapse, not a demand problem. Then it found the worse thing. That account's historical numbers had been inflated for months, because the Shopping campaigns were counting add-to-cart and begin-checkout events as conversions. One month showed $146,618 in "conversions" with zero purchase value behind it. Real purchases over five months were about ten. I had to go tell the client that the good numbers were not real. That's the least fun conversation in this business and it's exactly the conversation the system exists to force. The second receipt is a store I launched solo. [Emma Grace Home](https://emmagracehome.com/) opened on Shopify in April 2026 and did $5,387 in sales that month. In September it did $56,182, and April through September came to $154,878.84, all from its Shopify sales report. Those are store sales, and I count them as proof of the build rather than of any ad account. [How I built it is written up here](/results/150-sku-furniture-catalog-no-photographer/). ### Where It Breaks If I only told you the wins this would be another AI hype post, and I've [written about spotting that](/blog/spotting-ai-in-your-own-writing/). The agent on that Mac Mini crash-looped 262 times in a single day because one invalid line got written into its config. I found out because it went quiet. Before that it ran the API bill to about $56 a day by dragging a 207,000 token conversation into every single call. I fixed it by resetting the session nightly, not by anything clever. It has run out of API credits mid-week and silently stopped. So I built a watchdog that checked its health every five minutes, restored the last known good config, restarted it, and emailed me what it did. I built that because the thing failed, not because I planned well. In September I retired that agent entirely. The recurring jobs now run as plain scheduled scripts, with an email that fires when one of them goes quiet. It's less clever and it breaks less. The other honest limit: the AI is worse than me at knowing what matters. It will happily produce a competent report about the wrong thing. Every client-facing output still goes through me. When I tell a client they're paying for access to me, the access is the part that keeps the machine pointed at the right problem. ### Where I Think This Goes My read, stated plainly so you can hold me to it. The single-channel specialist role goes first. Nobody is going to pay a person a monthly fee to log into Google Ads and add negative keywords, because that's now a scheduled job. Most mid-market agencies follow. Not the top tier working on enormous accounts, but the ones charging a startup two grand a month to have three junior people each touch one platform. That model is a stack of individual jobs, and the jobs are what is being automated. I think five years is the outside number. That's a claim about a pricing model, not about the people inside it. The good shops already know. Some of the sharpest operators I know work at agencies and are rebuilding the same way I did, and the ones who move will come out of this fine. The ones billing the same retainer for work that now runs on a timer are the ones with a problem. What replaces it is a subscription to a system that has context. Not a person selling hours, and not software you have to operate yourself. Someone who has built the accumulated record of what worked across many businesses, plugged you into it, and stays personally responsible for the outcome. The value compounds because every client makes the system better for the next one. Here's the part I haven't built and will not claim I have: landing pages that update themselves against live campaign performance, season, and trend, with no human in the loop. Right now I can build a page fast, informed by data about which pages converted before, and change it quickly when the numbers move. That's assisted, not autonomous. The autonomous version is where I'm pointing everything, and I will write about it when it's real rather than when it's a slide. The reason I'm doubling down anyway is that the market hasn't caught up. Almost nobody selling marketing services right now has this built. That gap closes, and the people who started early will have years of context that the late arrivals can't buy. ### What This Means If You're Buying You don't need to care about any of my tooling. Most of my clients don't, and I've stopped showing them the vault unless they ask. What you should care about is the shape of what you're buying. If you're paying an agency and your point of contact is a junior account manager reading you a dashboard, you're paying for a job that is being automated, at the price of a person doing it by hand. If you're being sold "AI-powered" anything with no explanation of where the context comes from, it's a model with no memory of your business writing generic output. You can tell because it reads generic. The question worth asking a vendor is simple. What do you know about my business that you would still know six months from now if the person on this call left? If the answer is a folder of PDFs and one account manager's memory, that's the risk. I'm not selling you a quick fix, and I will never promise you a number. You're paying for the work and for the system the work runs through. If you're still working out which shape you're buying, the comparison is at [freelancer, agency, or fractional marketer](/hiring-a-marketer/freelancer-vs-agency-vs-fractional-marketer/), and the number at [what a fractional marketer costs](/hiring-a-marketer/how-much-does-a-fractional-marketer-cost/). If you want the diagnosis before the engagement, that's what [the free ads audit](/audit-request/) is for. It's written-first, no call required, and if I think the fix is upstream of anything I would run, I will tell you that instead of selling you a retainer. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## 44 Agents Tried to Break My Case Studies. Three Succeeded. URL: https://connercrowe.com/blog/44-agents-fact-checked-my-case-studies/ Published: 2026-07-17 | Updated: 2026-10-22 I reviewed my own five case studies and passed clean. Then 44 AI agents were told to refute the copy. They found 38 problems. Three would have embarrassed me. In June 2026 I finished five case studies for this site. Real accounts, real numbers, each one traced back to the source report. Before publishing I ran my usual pass: the voice check, the fact check, the read-aloud. It came back clean. Then I pointed 44 AI agents at the same five case studies and told each one to assume the copy was wrong and prove it. Between them they raised 38 findings. Three would have embarrassed me. The lesson wasn't that AI makes a good editor, though it does. The lesson was about my own clean pass. I wrote the copy, I checked the copy, I signed off on the copy, and an adversarial reader still found a client-data leak, an overclaim, and a number I couldn't trace, all in work I had just called finished. Your own review of your own work is the least reliable review you run. ### How the 44 agents were set up The structure is the part that matters, because "ask ChatGPT to check it" doesn't do this. I split the review into six dimensions: factual accuracy, client confidentiality, voice, internal consistency, overclaiming, and links. For each dimension I ran a set of independent agents, and every one got the same instruction. Refute by default. Don't confirm the claim, try to break it. Assume a number is wrong until you can't prove it is. Assume the client can be identified until you have checked every figure. Forty-four agents in total, each hunting for what wasn't right, none able to see the others' work or lean on my confidence that the copy was done. I couldn't talk them out of a finding, because they never heard me call it finished. ### The three that landed Thirty-eight findings came back. Most were small: a link to tighten, a sentence that read templated, a stat that needed its source line. Three were not small. One case study about a consent-tracking fix said two platforms had been reporting zero conversions. The account's own history showed one of them had already been corrected before my window. I had overstated the break. Scoped it down to what was true. One case study printed a client's exact cost per click. Anonymized everywhere else, and I had still left a real dollar figure a competitor could use. Changed it to a relative multiple. One live figure on a results page, a shopping-campaign lift, I couldn't trace back to a report when the agent pushed me to. If I can't source it, it doesn't ship. Pulled it and replaced it with a number I could stand behind. None of those were lies. Each was the kind of thing that happens when you're close to your own work and reading for confirmation instead of for holes. ### Why your own pass misses them When you review your own writing, you're not checking it. You're re-experiencing writing it. You remember what you meant, so you read what you meant, not what is on the page. You remember tracing the number, so the number looks sourced even where the citation is missing. Confidence is the problem, and you have the most confidence in the work you just finished. Catching an AI tell in your own prose is the easy version of this. Catching a false number you believe is the hard version. An adversary has none of that. It didn't write the sentence, it doesn't know what you meant, and you told it to assume you got it wrong. It reads what is there. That's why refute-by-default matters more than the model behind it. A friendly reviewer confirms. An adversarial one breaks. Only the breaking finds the leak. ### I did it again in July 2026 The home-services cost-per-lead post on this site went through the same thing before it shipped. Three agents, refute by default, and they caught two problems my own pass had waved through. The draft was about to cannibalize an existing page of mine, competing with it for the same search, and I had a number backwards. Both fixed before anyone saw them. I don't publish account work here without running it now. ### What I did not claim Not that the agents wrote or fixed anything. They found, I decided and rewrote. And not that this makes the case studies perfect. It makes them checked by something other than the person most motivated to believe they were done. That's a lower bar than perfect, and a much higher one than a self-review. ### The receipts Five case studies, one 44-agent adversarial workflow across the six dimensions above, 38 findings raised and dispositioned, three rated high severity and fixed before publish. The five are live on the [results page](/results/), and each one carries a "what I did not claim" section. That habit is the same instinct, written into the copy itself: name the thing you're not saying, so the thing you're saying can be trusted. A claim that survived an adversary is worth more than one you only reviewed yourself. ### Keep going - **[Spotting AI in your own writing](/blog/spotting-ai-in-your-own-writing/)**. The manual version of one of those six dimensions, the tells you learn to catch by hand. - **[Your AI agent doesn't need a brain](/blog/your-ai-agent-doesnt-need-a-brain/)**. Why the structure around the agents matters more than the model inside them. The same reason refute-by-default works. Free PDF: **[The Voice Audit Checklist](/freebies/voice-audit-checklist.pdf)**. The by-hand version of the voice dimension, the one I still run first. No email gate. ### What's next The five case studies at the [results page](/results/) are the ones that came out the other side of all 44 agents. The leaked CPC is gone, the overclaim is scoped down, the number I couldn't trace is replaced. If you're deciding who to trust with your accounts, the section to read isn't the win. It's the "what I did not claim." That line is the difference between someone showing you their work and someone showing you only the parts that flatter it. I hold my own claims to that check, and I'll hold your ad account's numbers to it too. [Send me read-only access for the free ads audit](/audit-request/) and I'll write up which reported numbers survive a second look, 5 to 7 pages, back within 24 hours. No call. ## Cost Per Lead Is the Wrong Number for Home Services URL: https://connercrowe.com/blog/cost-per-booked-job-home-services/ Published: 2026-07-17 | Updated: 2026-10-22 Judge home-services ads on cost per booked job. How to compute it from call and CRM data, and why the cheap lead usually loses. In the July 2026 edition of my home-services benchmark, the cheapest lead cost $80. The most expensive cost $309. And the cheap one often has the worse economics. That's the problem with judging Google Ads on cost per lead. The number that looks best is frequently the one you should trust least. Cost per lead tells you what it cost to make the phone ring. It says nothing about what it cost to book a job, and the job is the thing that pays you. The number that should run a home-services account is cost per booked job. Here is how to get it, and why the cheapest lead is often the worst deal on the board. ### What the dashboard shows you, and why it varies In that July 2026 edition, across twelve home-services accounts I manage, cost per lead over ninety days ran from about $80 for a lawn-care lead to $309 for a roofing lead, with a median near $128. (The Q3 2026 refresh puts the median at $163 across nine accounts with clean tracking, with the eight trades shown running from $98 to $480. The booked-job math below works the same at either number.) The order tracks job value. Septic and lawn leads are cheap because the jobs are smaller and the search is specific. Someone typing "septic pumping near me" is close to booking. Roofing and landscaping leads cost more because the jobs are larger and more of the clicks are early-stage shopping. The spread is about four times, which is the first reason a borrowed benchmark is useless. A number you read for HVAC tells you nothing about what a fencing lead should cost. The current trade-by-trade table, now on Q3 2026 data with the July figures kept beside it, is on the [home-services benchmark page](/home-services-paid-search-benchmarks/). There's a second problem underneath the first. In a good share of the accounts I audit, the conversion the dashboard counts isn't a real lead: every phone call scored as a win, or no call tracking at all. I broke down what that looked like in the July 2026 edition of that benchmark. The Q3 2026 refresh there has the current count: 2 of 11 accounts too broken to benchmark. Assume for the rest of this that your tracking is clean. Cost per lead is still not the number to manage on. ### The number that gets you paid | Account | Cost per lead | Close rate (illustration) | Cost per booked job | | --- | --- | --- | --- | | Roofing | $309 | One in four | About $1,240 | | Septic | $84 | One in three | About $250 | A lead isn't a customer. A booked job is. To get from one to the other you need two figures Google doesn't have: your close rate and your average job value. The math is simple. Cost per booked job = cost per lead ÷ close rate. Take the roofing account. A $309 lead sounds expensive until you run it through. Close one in four of those leads and each booked job cost about $1,240 in ad spend. Against a $15,000 roof, that's a rounding error. Now the septic account at an $84 lead, near the bottom of the book. Close one in three and each booked job cost about $250. If that job is a $400 pump-out, more than half of it went to Google. If it's a $6,000 drainfield repair, it's one of the best numbers you own. Same lead price, opposite verdict, and the dashboard can't see the difference. The cheap cost per lead isn't automatically the better account. Your close rate and your job value decide it, and those two inputs can flip which trade is winning. Run your own numbers here, not mine. The close rates and job values above are illustrations. Yours are the ones that matter. ### Why this comes back to your tracking Here is the catch that makes the booked-job number hard to chase. Google's Smart Bidding optimizes toward whatever you set as the conversion. Set it to "lead" and it buys you more leads, cheap ones, whether or not they book. It can't see which leads became jobs, because that happens in your CRM days later, off the platform. So the only way to make the algorithm chase booked jobs instead of phone rings is to send the booked job back to it. That's the rebuild I run on every service account, in this order: 1. Make one qualified lead the single primary conversion. Not every call. Not every form. The one action that means a real person wants real work. 2. Track calls properly, with a minimum call duration, so a fourteen-second wrong number doesn't get counted as a lead. 3. Feed the booked job back to Google. When a lead becomes a scheduled job in the CRM, send it back as an offline conversion with its dollar value attached. Once the booked job and its value flow back, Smart Bidding stops chasing $84 leads that never book and starts chasing the ones that fill the calendar. The full architecture for that loop is the [Lead Quality Stack](/frameworks/lead-quality-stack/). ### What it costs to leave it Leave it and you keep managing on the cheap number. The account that counts every call as a win will show you a $40 lead and look like your best performer, right until someone asks how many of those calls turned into jobs. Budget flows to the lowest cost per lead, which is often the account learning the least. The answer is one query away, and it's the first thing I pull on any account. Trace a month of leads to the jobs that closed. The gap between what the dashboard reported and what the calendar shows is the whole story. ### Keep going If this hit, the next two pieces in the same universe: - **[Form fills or qualified leads? Pick qualified](/lead-quality/optimize-google-ads-form-fills-or-qualified-leads/)**. Why the account should bid on booked jobs, not raw form fills, and how to switch it over. - **[How I cut spend in half and doubled signups for a real estate law firm](/results/real-estate-law-google-ads-growth/)**. The same count-the-right-thing rebuild on a legal account: WhatConverts-qualified calls, a case-type rebuild and 300+ negatives. Spend down 50% and cost per qualified lead down about half in 90 days. Free PDF: **[The 25-page Lead Quality Stack](/freebies/lead-quality-stack.pdf)**. The full architecture for feeding booked jobs back to Google. [No email gate](/blog/free-pdfs-without-an-email-gate/). ### What's next If you run a home-services account, run one number this week: your cost per booked job, not your cost per lead. If you want me to run it with you, that's the [free audit](/audit-request/). I look at your conversion setup, your call tracking, and whether your booked jobs make it back to Google, then I tell you which of your numbers are real. ## I Dismissed 1,476 of Google's Ad Recommendations URL: https://connercrowe.com/blog/google-optimization-score-recommendations/ Published: 2026-07-17 Dismissing a Google Ads recommendation raises your Optimization Score, the same as accepting it. I cleared twenty accounts by dismissing 1,476. In July 2026 I dismissed 1,476 of Google's ad recommendations across twenty accounts. Every account's Optimization Score went up. Optimization Score rises whether you apply a recommendation or dismiss it. It measures whether you have cleared Google's to-do list, not whether the account makes money. I cleared the list on those twenty accounts by adding 97 keywords by hand and dismissing the other 1,476. The scores climbed toward 100, and nothing about the accounts had changed. ### What the score counts Optimization Score looks like a health grade. It's a percentage, it sits at the top of the account, and a Google rep or an agency will tell you to push it to 100%. It isn't a health grade. It's the share of Google's open recommendations you have addressed, and you address one in two ways: apply it, or dismiss it. Ignore a recommendation and it drags your score down. Dismiss that same recommendation and your score goes up. Nothing about the account changed. No performance moved. You cleared the item. So a 100% Optimization Score tells you one thing for certain: someone went through the recommendations tab. It doesn't tell you the account makes money. ### Most of the recommendations point one way Read what Google generates and the pattern shows up fast. Broaden your match types. Turn on search partners. Add display expansion. Raise your budget. Switch to a broader bid strategy. Add these eighty keywords. Almost all of them widen the net toward more of Google's inventory and more of your spend. A few are useful, like a specific budget note or a weak responsive ad flagged for a rewrite. Most are the house suggesting you bet more. That isn't a conspiracy. Google's recommendations optimize for Google's outcome, which is spend across more surfaces. Sometimes that lines up with yours. Often it does not. Optimization Score is the scoreboard for a game the house designed, and it's one move in a wider pattern I keep mapped in [the wasted-ad-spend library](/wasted-ad-spend/): the platform's own advice almost always points toward spending more. ### What I did on the twenty accounts The flag was real. A thin ad group gets starved, so "not enough relevant keywords" is worth clearing. But the fix Google wanted was a bulk-accept of its keyword list, which on these accounts meant broad, out-of-area, and off-service terms that would spend before they converted. So I cleared it the other way. On each account I added only the keywords that were locally viable and on-service, the ones a real customer in that market types. That came to 97 across the twenty. Then I dismissed the remaining 1,476, because a dismissed recommendation is a reviewed recommendation, and reviewing them is the real work. The warning cleared on all twenty, zero recommendations left open, and scores that had been sitting anywhere from the low seventies to the high nineties walked toward 100. Not because I improved anything that hour. Because I emptied the list. ### The rule I run now Treat the recommendations tab as a to-do list to review, not a score to chase. 1. Apply the few that fit your strategy. Budget notes and ad-strength fixes usually earn it. 2. Dismiss the rest deliberately. Broad match, search partners, display expansion, and "switch to a broader bid strategy" get dismissed on sight on a lead-gen account. 3. Turn off auto-apply. Leave it on and Google applies its own recommendations to your account overnight, bid-strategy changes included, with no human in the loop. I've opened accounts that quietly rebid themselves this way. Google regenerates these recommendations constantly, so the list refills within weeks and the score slips again. Clearing it is a monthly pass, not a one-time fix. Do that and your Optimization Score reads high, which is fine. Know it reads high because the list is empty, not because the account is winning. The number that says the account is winning is cost per booked sale, and it doesn't live in the recommendations tab. ### What I did not claim Not that every recommendation is wrong. Some are worth applying, and a thin ad group is a real problem. The point is narrower. Optimization Score measures whether you cleared the list, not whether the account performs, so it isn't a number to manage by, and it isn't a number to let auto-apply chase with your budget. ### Keep going - **[AI Max and your search terms](/blog/google-ads-search-term-matching-ai-max/)**. The same automation over-reach, this time in how Google decides what your keywords match. - **[How I cut spend in half and doubled signups for a real estate law firm](/results/real-estate-law-google-ads-growth/)**. Cost per qualified lead, the number to manage the account by instead of the score. - **[The common mistakes behind wasted ad spend](/wasted-ad-spend/common-mistakes-causing-wasted-digital-ad-spend/)**. Applying recommendations on autopilot is one of them, and the rest of the list costs just as much. Free PDF: **[The Google Ads Setup Checklist](/freebies/google-ads-setup-audit.pdf)**. The checkpoint-by-checkpoint version of what I verify before I trust an account. [No email gate](/blog/free-pdfs-without-an-email-gate/). ### What's next If your agency or your Google rep keeps pointing at your Optimization Score, ask them what it would be if you dismissed every open recommendation. Then ask what your cost per booked sale did last quarter. If you want someone to go through the account and tell you which recommendations are worth applying and which are the house betting your money, that's what [the free ads audit](/audit-request/) is for: read-only access, 5 to 7 pages written by me, back within 24 hours of access. ## What a Home Brand's 21x ROAS Hides URL: https://connercrowe.com/blog/what-a-home-brands-21x-roas-actually-hides/ Published: 2026-06-29 | Updated: 2026-10-22 Two home brands I run report 21x ROAS. A Shopping campaign that is 18% of the spend produces 52% of the revenue. Why the blended number lies. I pulled ninety days of Google Ads data on two home and furniture brands I run. Both report a blended return on ad spend north of 20x. One reads 21.7x, the other 20.4x. On the dashboard, both look like the ads are printing money. They are not. The blended number is hiding where the revenue comes from, and once you break it apart the story flips. ### The two accounts, broken apart Between them, the two brands spent about $13,500 on Google Ads in the ninety days to late June 2026 and tracked roughly $284,000 in revenue. That's the 21x. Here's what it's made of. Each account runs two campaign types: a Performance Max campaign that carries most of the budget, and a smaller Shopping campaign. When you split the ROAS by campaign type, the average comes apart. - **Shopping was 18% of the combined spend and produced 52% of the tracked revenue.** Its ROAS was 61x. - **Performance Max was 82% of the combined spend and produced 48% of the revenue.** Its ROAS was 12x. Per brand it's even sharper. One brand's Shopping campaign reported a 45x return. The other's reported 71x. Their Performance Max campaigns, the ones doing four-fifths of the spending, reported 17x and 8x. A single small campaign, less than a fifth of the budget, is carrying the headline number for both accounts. ### Why the Shopping campaign reads 60x A 60x or 70x ROAS isn't skill. It's intent. Shopping campaigns are very good at catching people who are already most of the way to buying. Someone who searches the brand by name, or searches the exact product they already saw on Instagram, and clicks the Shopping listing. That click was going to convert with or without the ad. The campaign takes the credit anyway. It's the same pattern as branded search dressed up as performance. The traffic is real, the sale is real, but the ad didn't create the demand. It harvested demand that already existed. Harvesting is cheap and it converts at rates prospecting never will, so its ROAS looks unreal. Because it sits inside the same account as the prospecting, it drags the blended average up and makes the whole thing look healthier than it is. ### Why the blended number is the wrong number If you judge a home brand's Google Ads on blended ROAS, you make two mistakes at once. You over-credit the ads. A 21x blended return tells the founder the account is a money machine, so the instinct is to leave it alone. But half of that revenue is demand the brand already had. Strip out the harvesting and the part of the account that's buying new customers is running at 12x, and on one of the two brands, 8x. And you starve the part that grows the business. In these two accounts Performance Max carries the prospecting, so it's where new customers come from. It isn't pure prospecting (unless you exclude them, it can still serve brand searches and remarketing), but it does most of the finding. It runs at a lower ROAS by design, because finding someone who has never heard of you costs more than closing someone who already wants you. Read the blended number and that layer looks like the weak one. Cut it, and you have just cut the only thing creating new demand, while keeping the campaign that was only ever collecting it. ### What to look at Separate the harvest from the prospecting before you judge anything. Pull the ROAS by campaign type, not the account average. Look at what Performance Max returns on its own, and decide whether that number pays back at your margin and your customer lifetime value. The Shopping or branded number is a foregone conclusion, useful for catching tracking breaks, useless for deciding where the next dollar goes. Then ask the harder question the dashboard will never answer: how much of the harvested revenue would have happened anyway. Incrementality, not attribution. A holdout test on branded and Shopping traffic tells you what the ads are adding, and it's almost always less than the platform claims. The 21x isn't a lie exactly. Every dollar in it is real. It's just an average of two completely different jobs, and averaging them together tells you nothing about whether your spend is growing the brand or just taking credit for it. If you run a home or furniture brand on Shopify, this is the first thing I check, and it's the structure I rebuild every account to. Since October 2024, Performance Max no longer gets automatic priority over Standard Shopping for the same products; Ad Rank decides, which is why either campaign can end up holding the branded traffic. The mechanics of why Performance Max and Shopping fight over the same conversions are in [why Performance Max gets credit for Shopping conversions](/blog/why-performance-max-gets-credit-for-shopping-conversions/). What a healthy ROAS looks like once margin is in the picture is in [the ROAS benchmark answer](/wasted-ad-spend/good-roas-benchmark-ecommerce-businesses/). And the way I structure the whole program for home brands is on the [home brands page](/for-home-brands/). If you want this split run on your own account, [send me read-only access for the free ads audit](/audit-request/). I'll break the ROAS out by campaign type and have it back in writing within 24 hours. No call. ## Klaviyo Flows Built on Room Scenes for a Furniture Brand URL: https://connercrowe.com/blog/email-flows-room-scenes-shopify-home-brands/ Published: 2026-06-18 | Updated: 2026-10-22 Five Klaviyo flows and 21 hero emails on seven custom room scenes for a furniture brand with a four-figure AOV. Abandoned checkout is the cheapest revenue. ### Quick Take A furniture and lighting brand on Shopify with a four-figure average order was running one default Klaviyo welcome email and nothing else. Every other lifecycle moment was empty. So the most expensive event in the funnel, a buyer abandoning a $3,000 cart, got zero follow-up. I built the full lifecycle suite: Welcome, Abandoned Checkout, Browse Abandonment, Post-Purchase, Winback. Five flows, 21 hero emails. The part that mattered was the imagery. Each hero email is built on one of 7 custom room scenes that place the brand's actual SKUs, true-to-scale, in a styled room, rendered through the same in-house lifestyle pipeline I use for the catalog. The email looks like the storefront, not a newsletter. A stock template sent to a buyer choosing a $3,000 sofa reads as spam. The inbox is a catalog surface, and most home brands waste it. ### The receipt A home furnishings brand on Shopify. Average order in four figures. Real revenue, real catalog, real buyers. The email program was one welcome email on the default Klaviyo template. No abandoned checkout flow. No browse flow. No post-purchase. No winback. That's the common state for home brands. Acquisition gets all the budget. Email gets a checkbox. The problem is the math. At a four-figure AOV, checkout abandonment is the single most expensive leak in the business. A buyer who reaches checkout has already chosen a $3,000 item. That's the most qualified person who will touch the funnel all month. With no abandoned-checkout flow, that buyer leaves and nothing follows them. I built five flows: - **Welcome.** First impression for a new subscriber who hasn't bought yet. - **Abandoned Checkout.** Recovers the buyer who chose a $3,000+ item and stalled. - **Browse Abandonment.** Catches the buyer who viewed a product and left before adding to cart. - **Post-Purchase.** Owns the gap between order and delivery, the highest-anxiety window for a high-ticket purchase. - **Winback.** Re-engages a past buyer before they go cold. Across those five flows: 21 branded hero emails, built on 7 custom room scenes. ### The inbox is a catalog surface A home brand spends months getting the storefront right. The PDP imagery, the room context, the styling that makes a $3,000 sofa feel like it belongs in a real home. Then the same brand sends an email built on a gray newsletter template with a logo at the top and a product thumbnail in a box. The buyer notices. The storefront looked like a designer's portfolio. The email looks like a coupon from a phone carrier. That gap is the leak. The email is a catalog surface. It's a placement where the buyer sees the product. Treating it like a utility instead of a storefront throws away the most-opened touchpoint you own. The fix is imagery. The hero email has to look like the catalog, because to the buyer it's the catalog. Same room context. Same styling. Same fidelity. The email should be indistinguishable from a collection page. When it is, the email stops reading as a broadcast and starts reading as the brand walking the buyer back to a specific piece. ### Why a stock template loses a premium buyer A $3,000 furniture buyer isn't an impulse shopper. They're deliberating. They compared four brands. They measured the wall. That buyer's standard for what a brand looks like is set by the storefront they just left. A default template email falls below that standard the instant it loads. Worse: the default template signals automation. It tells the buyer this email went to the whole list. A buyer making a $3,000 decision wants to feel chosen, not batched. The room-scene hero does the opposite. It shows the exact piece they were considering, staged the way they imagined it at home. That isn't a discount nudge. That's the brand finishing the sentence the buyer started at checkout. For a high-AOV brand, the abandoned-checkout flow is the highest-leverage email in the business. It catches the most qualified buyer at the most expensive moment. Sending that buyer a stock template is the most expensive template choice a home brand can make. ### How the room scenes are built The 7 room scenes aren't stock photography and not generic AI renders. Each one places the brand's actual SKUs in a styled room, true-to-scale. They run through the same in-house lifestyle pipeline I use for [catalog imagery](/results/150-sku-furniture-catalog-no-photographer/). Same controls: - **Real SKUs.** The piece in the scene is the piece that ships. Same silhouette, same finish, same hardware. - **True-to-scale.** A console reads as a console, not a toy. Proportion against the room is locked, because a buyer spending $3,000 will catch a scale error instantly. - **Brand-spec lock.** Material, color, and finish are checked against the product spec before the scene ships. If the render drifts, it gets rejected. Seven scenes covers the lifecycle. A hero scene for Welcome. A specific-product scene for Abandoned Checkout and Browse. A delivered-in-a-home scene for Post-Purchase. A fresh-room scene for Winback. The 21 emails draw from those 7 scenes so the whole program looks like one brand, not 21 separate sends. That consistency is the asset. The buyer who sees the Welcome email and later the Abandoned Checkout email sees the same world both times. The brand never breaks character. ### When this is not the right call This doesn't fix a broken product. If the piece arrives and disappoints, no email saves it. The flows recover qualified buyers; they don't manufacture quality. It doesn't fix pricing. If the offer is wrong, a beautiful email sends a wrong offer faster. It isn't for a sub-$1M brand with a 20-SKU catalog. At low order values and low volume, a clean text email and one good product photo clears the bar. The room-scene investment pays back at high AOV and real list volume, where one recovered $3,000 cart covers a lot of build. And it breaks without a brand look. If the team can't supply the styling language the scenes should match, the renders drift and the emails stop looking like the storefront. The brand reference is the first input, not an afterthought. ### Keep going If this landed, the three pieces next to it: - **[Photographing a 100+ SKU catalog](/product-photography/product-photography-large-shopify-catalog/)**. The same imagery pipeline, applied to the PDP instead of the inbox. - **[A 150-SKU Furniture Catalog, Built Solo in Under a Month](/results/150-sku-furniture-catalog-no-photographer/)**. The pipeline on a full catalog: 150-plus products, every image rendered in-house, zero product shoots. - **[What conversion value to send to Google and Meta](/conversion-tracking/conversion-value-to-send-google-meta/)**. Why the revenue number your ads learn from has to match the one in your books. The full production exhibit, including the lifestyle pipeline shape and the fidelity controls, lives on [the home and furniture brands page](/for-home-brands/). If your home brand is sending a stock template to a $3,000 buyer, start with the [free catalog audit](/audit-request/?type=catalog). ## The $30K Shopify rebuild you probably don't need URL: https://connercrowe.com/blog/shopify-storefront-facelift-without-a-rebuild/ Published: 2026-06-18 | Updated: 2026-10-22 An agency quoted a home brand $30K and 12 weeks to rebuild its Shopify store. A facelift on the existing theme got most of the result. ### Quick Take A home brand on Shopify got a rebuild quote: $30,000 and 12 weeks. The store felt dated and "thin," and the agency proposed tearing it down and building a new theme. I ran the numbers and the diagnosis was off. The store didn't need a new theme. It needed better product cards, a readable hero, a trust layer, faster pages, and a catalog cleanup. All of that is fixable on the theme the brand already owns. I call it a storefront facelift. It got most of the result the rebuild promised, in a fraction of the time, on the existing stock Shopify theme. The lesson: a rebuild quote sells you a new theme. The conversion problem is almost never the theme. ### The receipt A Shopify boutique came to me with an agency quote in hand. $30,000. 12 weeks. A full rebuild on a new theme. The owner was tired of the store. It looked stock. It felt slow. Customers said it felt "thin," like the catalog was small. I pulled up the store and made a list of what was wrong. The product cards were the default theme cards. No quick context, weak hover, prices buried. The hero image had white text on a light photo. You couldn't read the headline on a phone. There was no trust layer. No reviews on the page, no guarantee, no proof. The pages were slow. Heavy apps and uncompressed images. And the catalog had a hidden problem I will get to in a minute. None of that is a theme problem. Every item on that list lives on top of the theme, not inside it. I shipped the fixes on a draft theme. The live store never moved while I worked. Custom product cards, a readable hero, a reviews carousel, a fit finder, internal linking, and a re-tokened palette and type so the stock theme stopped looking stock. Page-speed work on top. ### Why the theme is rarely the problem When a store feels dated, the instinct is to replace it. That instinct is wrong most of the time. A Shopify theme is a frame. What converts or fails to convert is what sits inside the frame. Cards. Hero. Trust. Speed. Catalog hygiene. You can swap the whole frame for $30,000 and still ship the same weak cards, the same unreadable hero, the same missing reviews. Plenty of expensive rebuilds do exactly that. The brand pays for a new theme and the conversion rate doesn't move. The theme is the cheapest thing to keep and the most expensive thing to replace. So I keep it. ### The facelift menu A facelift is a set of upgrades injected through theme CSS and JS. Not a page-builder app. The code lives in the theme, so there's no monthly app tax and nothing to break on the next theme update I don't control. Here's what I inject and why. **Custom product cards.** The default card is generic. I rebuild it to carry price, a second image on hover, and quick context. This is the single highest-leverage change on most stores. **Hero legibility.** A readable headline beats a pretty one. I fix contrast, add a scrim where needed, and size the type for a phone first. **A reviews carousel.** Proof on the page, not buried on a separate reviews tab nobody opens. **A fit or finder quiz.** For a catalog with real variety, a short quiz routes a confused shopper to the right product instead of leaving them to bounce. **Internal linking.** Collections and products that point at each other keep people moving and help search. **Palette and type re-tokening.** I rewrite the theme's color and font tokens so it stops reading as a template. Same theme. Different brand. **Page-speed work.** Compress images, defer what can wait, cut dead apps. None of these touch the live store while I build. They go into a draft theme and get reviewed before anything ships. ### The draft-theme and go-live discipline The live store is the brand's revenue. I don't edit it directly. Ever. Every change goes into a draft theme. The brand can preview the full facelift on a private link before a single customer sees it. Before go-live, I run a design audit with a sub-agent. A second set of eyes on the whole thing: spacing, contrast, mobile, the cards, the hero, the trust layer. It catches what I stop seeing after staring at a build for days. Then go-live is one coordinated move. The draft theme gets published in a single step, with the catalog and tracking checked in the same pass. No half-published state. No broken window where the store is part old, part new. This is the part a rebuild quote rarely spells out. The risk is not the design. The risk is the switch. ### The catalog-hygiene finding Here's the part no rebuild quote mentions. On one facelift, the owner kept saying the store felt thin. I assumed it was a design problem. It was not. I checked the catalog against what was published to the Online Store. Roughly 61 percent of the active catalog, over 1,500 products, was never published. It existed in the admin. It was invisible on the storefront. It was absent from the brand filters. The store didn't feel thin because of the design. It felt thin because more than half the catalog was hidden. A new theme wouldn't have found that. A rebuild ships the same hidden catalog onto a prettier frame and the brand still wonders why the store feels empty. A facelift surfaces problems a rebuild quote never lists. Because a facelift starts with the store you have, not the store someone wants to sell you. ### When a rebuild IS the right call I'm not against rebuilds. Some stores need one. If the theme genuinely can't carry the brand, rebuild. Some old themes are so limited that every change fights the frame. At that point you're paying me to wrestle the theme instead of improving the store. A clean theme is cheaper. If you need performance at a scale a stock theme can't hit, headless is a real answer. High traffic, complex merchandising, a content layer that has to be fast everywhere. That's a different engineering problem and a rebuild earns its cost. If you're changing platforms, that's a migration, not a facelift. Moving off a legacy platform onto Shopify is real work and worth doing well. And if there's no store yet, there's nothing to facelift. [Emma Grace Home](https://emmagracehome.com/) was a new store, so I designed the whole storefront on a Shopify theme I customized end to end, along with 150-plus products and every product image, from kickoff to launch in under a month. [That build is written up here](/results/150-sku-furniture-catalog-no-photographer/). The honest line: a rebuild is right when the foundation is the problem. It's wrong when the foundation is fine and the agency is selling you a new one anyway. For most $3M to $8M home brands with a tired store, the foundation is fine. ### Keep going If you're weighing an agency quote, two more reads in the same vein. [Why I'm still writing this blog in 2026](/blog/why-im-still-writing-this-blog-in-2026/) is the case for doing the work in the open instead of hiding it behind a pitch. [Free PDFs without an email gate](/blog/free-pdfs-without-an-email-gate/) is the same anti-friction logic applied to lead generation. If you want to see how I run an engagement end to end, that's on [my process page](/process/). The two questions that decide it: [facelift or full rebuild](/shopify-storefront/shopify-facelift-vs-full-rebuild/), and [whether a redesign will hurt SEO](/shopify-storefront/will-a-shopify-redesign-hurt-seo/). And if you have a rebuild quote in your inbox and a store you're not sure needs one: the facelift work itself, with pricing and the case studies, is on [my Shopify storefronts page](/storefronts/), and the full theme-versus-facelift-versus-rebuild decision library is [the Shopify storefront decision hub](/shopify-storefront/). If you run paid traffic to that store, send me read-only access to the ad account first for [the free ads audit](/audit-request/). I'll tell you what the tracking and the traffic say before you sign anything. It's back in writing within 24 hours. No call. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Shopify Storefront Audit: My 6-Pass Method URL: https://connercrowe.com/blog/how-i-audit-a-shopify-storefront/ Published: 2026-05-26 | Updated: 2026-10-21 My six-pass Shopify storefront audit covers technical SEO, content, mobile UX, AI discovery, competitive positioning, and site architecture. ### Quick Take A storefront audit run fast is usually a storefront audit run shallow. Someone runs a crawler, exports 300 warnings, and calls it a diagnosis. A home brand owner can feel how thin that is even when they can't name it. Speed and depth only pull against each other when the method is bad. Here's the method I use to audit a Shopify storefront in an afternoon and keep it at senior level: six domain audits in parallel, one operator judgment pass, and a verification pass because the fast tools miss things. ### The six storefront audit passes 1. **Technical SEO.** Schema, sitemaps, indexation, titles, descriptions, and internal links. 2. **On-page content and brand voice.** Whether the copy reads human, answers buying questions, and earns the price point. 3. **Conversion and mobile UX.** Where the funnel leaks, especially on phones. 4. **AI search and discoverability.** Whether the catalog is legible to ChatGPT and Google's AI surfaces. 5. **Competitive positioning.** Where the brand sits against its peer set. 6. **Site architecture.** Collections, navigation, filters, and internal discovery. I run those six passes in parallel. Then I apply two gates that don't belong to any single domain: one operator judgment pass to turn findings into a ranked diagnosis, and one rendered-site verification pass before anything reaches the owner. ### The tension worth naming There are two kinds of storefront audit a home brand usually gets. The fast one is a tool export. A crawler runs, flags 300 issues, sorts them red and yellow, and the brand gets a PDF that treats a missing alt tag and a broken canonical as the same size of problem. It's fast and it's shallow. The brand can't act on it. The slow one is a consultant who spends three weeks and bills for the calendar. It's senior and it's slow, and most of the three weeks is scheduling, not thinking. Owners assume they have to pick. They don't. The speed in a fast audit comes from running things in parallel. The shallowness comes from skipping the judgment pass. Those are two separate choices. Keep the parallelism, add the judgment pass back, and the audit is both fast and senior. ### Six domains, run at once A storefront doesn't fail in one place. It fails in six, and they're different disciplines. I audit those six domains as separate passes at the same time rather than walking the store once. Technical SEO covers schema, sitemap, indexation, and the title layer. Content and voice cover the words that have to earn trust. Conversion and mobile UX cover the route from collection to checkout. AI discovery tests whether the catalog is legible beyond a conventional search result. Competitive positioning tests the claims against the peer set. Architecture covers collections, navigation, filters, and internal discovery. The stores themselves vary more than most checklists assume. Across the four Shopify stores I manage, including [Emma Grace Home](https://emmagracehome.com/), as of October 21, 2026, catalogs run from 3 products to 5,116, and the stores hold 31 staged theme copies against one live theme each. Before any pass starts, I confirm which theme is live, because an audit of a staged copy describes a store nobody visits. A few checks inside the passes carry hard numbers: - **Speed, in the mobile UX pass.** I read Core Web Vitals against [Google's thresholds](https://web.dev/articles/vitals): Largest Contentful Paint within 2.5 seconds, Interaction to Next Paint at 200 milliseconds or less, and Cumulative Layout Shift at 0.1 or less, each at the 75th percentile of page loads. - **Redirects, in the technical pass.** Shopify [only fires a URL redirect when the old URL returns a 404](https://help.shopify.com/en/manual/online-store/menus-and-links/url-redirect). If the old page still resolves, the redirect sits in the admin and does nothing, so I test each one against the live URL. - **What the ad account calls a sale, in the conversion pass.** On 2 of 3 ecommerce accounts I measured between May 13 and August 10, 2026, add-to-cart and begin-checkout events made up 16.6% to 17.2% of the conversions the ad account counted. The rows are in my [ecommerce tracking accuracy benchmark](/benchmarks/ecommerce-tracking-accuracy/). That inflates how well the store appears to convert before anyone touches the theme. One reviewer walking the site once can't hold all six lenses at the same depth. Attention narrows to whatever that reviewer is strongest at. Six parallel passes, one per domain, fixes that. Each pass goes deep on its own discipline and returns a structured report. This is the parallelism that makes the audit fast. It's also what makes each domain deep instead of skimmed. It's the same operator-as-supervisor pattern I use in [the content engine I built for a Shopify home brand's blog](/blog/content-engine-shopify-furniture-brand/). ### Six reports is not an audit Here's where the fast-and-shallow version stops, and where the senior work starts. Six domain reports aren't an audit. They're six domain reports. An owner handed all six is back where they started: a pile of findings with no order. The audit is the judgment pass. That's the part that can't be parallelized. The judgment pass does two things. First, it collapses sixty symptoms into a handful of root causes. On one home-brand storefront I audited, the content audit flagged no reviews, the UX audit flagged no financing shown, the voice audit flagged machine-written copy, the architecture audit flagged thin trust signals. Four findings, four domains, one root cause: the storefront couldn't earn trust at a four-figure order value. You fix the root cause, not the four symptoms separately. Second, it sequences. Not everything ships at once, and order matters. SEO metadata has the longest lag, since Google has to recrawl the pages before it counts, so it ships first. The work that changes what a shopper sees the day they land ships next. Structural work ships last, because it compounds quietly once the data underneath it is clean. A list of 300 warnings has no sequence. A senior audit is mostly sequence. ### The verification pass, because the tools lie by omission The fast tools have a blind spot, and a fast audit that doesn't account for it ships wrong findings with confidence. Most automated audit tools read the raw HTML a page sends. They don't run the page's JavaScript. So anything the storefront renders client-side after load is invisible to them. On that same storefront, the first-pass automated audit reported four things missing that weren't missing. An email capture popup it called absent was installed and firing. A Shop Pay financing widget it called missing was live on every product page. An About page it called a 404 existed at a URL the tool didn't guess. A collection it called broken was a transient error. Four false negatives, one root cause: the tool saw static HTML and the storefront rendered those elements with JavaScript. The mirror of that error is the false positive, where the audit reports a page that exists but was never part of the site. That's what happens when the page list is assembled from an analytics property rather than from the storefront, because [a property is scoped to its key rather than to a site](/blog/foreign-hostname-analytics-check/) and will happily list pages from anywhere else that key is installed. This is why a fast audit needs a verification pass before the findings go to the owner. The senior move is knowing the tools have that blind spot and checking the rendered page, not the raw source, before telling a brand something is broken. Skip the pass and you hand the owner a fix list with four items that are already done. ### What a fast, shallow audit costs The shallow version does real damage because it's confidently wrong. A 300-warning crawler export gets acted on. Someone on the brand's side spends a week working the red items. Some of those items are already fixed. Some are symptoms of a root cause that the list never named, so fixing them changes nothing. The brand spends real labor and ends the week roughly where it started, and concludes that audits don't work. Audits work. Warning exports aren't audits. The difference is the judgment pass and the verification pass, and those are the two steps the fast-and-shallow version drops to stay fast. You don't have to drop them. You have to run the coverage in parallel so you can afford to spend the saved time on judgment. ### The receipts The method on this build: six domain audits run in parallel, each returning a structured report. One judgment pass that collapsed the findings into root causes and a sequenced action plan. One verification pass that caught four false negatives before they reached the owner. The diagnosis was done inside an afternoon. The findings were senior because the judgment and verification passes both ran. The honest part. An audit is a diagnosis. The fixes ship after it, and the ranking and conversion outcomes take a quarter to read. A fast audit gets you a correct, ordered, verified list of what to fix. It doesn't get you the outcome. Anyone selling you the outcome in an afternoon is selling you the shallow version. ### What to ask for When a storefront audit comes back, don't ask how many issues it found. A high number is a tooling artifact. Ask two things. What are the root causes, and in what order do I fix them. If the audit can't answer both, it was the fast, shallow kind. That's the audit I run for Shopify home brands: six domains, ordered into one diagnosis, verified, [delivered in writing within 24 hours, no call required](/audit-request/?type=catalog). For a home or furniture store I read the public store, so there's nothing to grant. If your store isn't home or furniture, [request a free ads audit instead](/audit-request/). The build work that tends to come out of it lives on [my Shopify storefronts page](/storefronts/), and [redesigning without losing sales](/shopify-storefront/redesign-shopify-store-without-losing-sales/) covers how I ship it. The store I designed for Emma Grace Home is [live](https://emmagracehome.com/) if you want to see one, and [how I built it](/results/150-sku-furniture-catalog-no-photographer/) is written up. Two findings come up often enough to have their own answers: [why a Shopify store is slow](/shopify-storefront/why-is-my-shopify-store-slow/) and [whether the store needs a facelift or a full rebuild](/shopify-storefront/shopify-facelift-vs-full-rebuild/). Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Your AI Agent Doesn't Need a Brain URL: https://connercrowe.com/blog/your-ai-agent-doesnt-need-a-brain/ Published: 2026-05-20 | Updated: 2026-10-22 An operator's read on the agent-memory hype. My research agents scored it about 60% noise and 40% real, and the real part runs on plain text files. ### Quick Take Every few months a new word shows up to tell you your AI setup is behind. Right now the word is memory. My feed is full of videos about building a "brain" for an AI agent: memory layers, knowledge graphs, context engineering. I build my own AI systems for client work, so I took the trend seriously and sent a team of research agents through all of it. The verdict came back about 60 percent noise and 40 percent real. The real part is something you can build out of plain text files. You don't need to buy a memory product. You need better notes. Here is the full read. ### The Jargon, Decoded Strip the jargon and agent memory means one thing. An AI tool that doesn't start from zero every time you open it. By default most AI agents are amnesiacs. You finish a session, close the window, and everything the tool learned about your brand and your last decision is gone. Next session you brief it from scratch. "Building a brain" is the catch-all for fixing that. The trend splits the idea into two kinds of memory. Semantic memory is facts and rules: your brand voice and your margin floor. Episodic memory is what happened: which campaign you ran last month and what it returned. Most marketing teams using AI have written down neither in a place the tool can read. Everything else is plumbing. Knowledge graphs and vector databases are storage formats for the two kinds of memory above. Useful to engineers. Mostly noise to an operator. ### Who Is Selling You This Once you see the layers, the hype sorts itself. The loudest layer is the vendors. A wave of startups now sells "memory layers" as a product. Mem0, Letta, a dozen others. They publish "State of AI Agent Memory" reports that read like research and function like brochures. Behind them are the investors who funded the round and need the category to feel inevitable. The next layer is the influencers. "Context engineer" gets called the hottest job title of the year by people who make videos for a living. That tells you nothing about whether you need one. The thin layer underneath is the real signal. Anthropic, the lab behind the model I run every day, treats context engineering as a genuine engineering discipline with named failure modes. When the people building the technology write something careful, read it. When someone selling the technology writes something urgent, discount it. The mix lands around 60 / 40. The trend is real. The volume around it is not. ### The Part the Videos Skip Here is the detail almost no video mentions. Giving an AI a memory is easy. Deciding what it should forget is hard, and it's mostly unsolved. Add memory to an agent and at first it works. Then stale facts pile up. Old decisions contradict new ones. The tool starts confidently citing something that stopped being true six weeks ago. Vendor benchmarks measure how well a system retrieves a fact. They don't measure how well it drops one that went bad. On a real workload, accuracy can slide noticeably within weeks of use. So the demo looks perfect and the production system slowly rots. Anyone selling you a memory layer is selling the easy half of the problem. ### What I Built Instead I had the same itch, because I do have a real memory gap in my own work. I run AI pipelines that produce ad creative and content for clients. They generate the work, it ships, and nothing comes back. The pipeline never learns whether the ad it made performed. That's the missing memory. Not a philosophical one. A specific, fixable one. I didn't buy a memory layer. I built a small tool that reads ad performance from Meta and Google and writes it into a plain text file. Winning angles. Creative that's fading. Concepts that are dead. One file per client. The production pipeline reads that file before it generates the next batch. The file is the brain. It's a Markdown document. I can read the whole thing in ten seconds, and so can the client. ### The Operating Rule Memory is a discipline, not a product. Your AI systems already have memory the moment you give them a file to read. A document with your brand rules is semantic memory. A running log of what each campaign returned is episodic memory. Plain files, kept current, under version control. That covers what almost every marketing operator needs. A vector database answers a problem of scale that most operators will not have for years. Reaching for one now is buying a forklift to move a single box. The rule I run on: build memory out of files a human can open and read. If you can't read your agent's memory, you can't tell when it's wrong. And it will be wrong. A memory you can audit beats a memory you have to trust. ### The Receipt The tool I built is real and it's boring. Three files. Markdown and JSON. No database, no vendor account, no monthly fee. It runs at zero marginal cost and the whole thing is legible to anyone who opens the folder. The trend will resurface next quarter in a louder form with a new word attached. The file will still be a file. Boring infrastructure you understand beats exciting infrastructure you rent. If you want to know what your ad account is actually doing, automated features included, [the free ads audit](/audit-request/) is the place to start. Send me read-only access and I write it up myself, 5 to 7 pages, back within 24 hours. No call. ### Keep going If this hit, the next two pieces in the same universe: - **[Nine GTM Tags in 90 Minutes](/blog/agent-built-my-tracking-stack/)**. The same build tooling pointed at tracking ops instead of ad creative. - **[The Workflow That Lets Me Ship a Site in 24 Hours](/blog/wispr-flow-claude-the-input-bandwidth-problem/)**. The input-layer workflow underneath all of this. Free PDF: **[The two-page Voice Audit Checklist](/freebies/voice-audit-checklist.pdf)**. No email gate. ## The Content Engine Behind a Shopify Home Brand's Blog URL: https://connercrowe.com/blog/content-engine-shopify-furniture-brand/ Published: 2026-05-19 | Updated: 2026-10-22 The nine-agent content engine I built for Emma Grace Home on Shopify, and the rule that matters most: no post publishes unless it links a real product. ### Quick Take A blog folder produces posts. An engine produces posts that are forced to earn their slot. I built a nine-agent content pipeline for [Emma Grace Home](/results/content-engine-home-brand-sixteen-posts/), a furniture brand on Shopify, that researches a topic, drafts it, runs it through two voice gates, and attaches schema, then hands a Shopify draft to a human. The rule that does the real work is the one that kills the post before it's written: if a topic can't anchor to a specific product in the catalog, the engine refuses it. That single gate is the difference between content that ranks and content that just exists. ### Why the engine exists Emma Grace Home had a blog. It had posts. The posts were generic. "What is solid wood." "The history of farmhouse furniture." Category-level education that read fine and ranked for nothing, because every other home-goods site on the internet had published the same article, and a generative tool can now write it in eight seconds. The first batch I ran through an early version of the pipeline made the same mistake. The drafts opened with contrarian category takes and never named a product the brand sold. They were articles. They weren't storefront assets. I deleted them. That failure is what defined the engine. A content engine for an e-commerce brand has one job that a media blog doesn't: every post has to move a reader toward something with a price and an add-to-cart button. If a post doesn't do that, it isn't underperforming. It's off-task. So the engine got a rule. Then it got the agents to enforce the rule. ### The pipeline Nine agents run in sequence and in parallel, each one narrow. A **keyword researcher** enriches the topic with real search terms. An **outliner** turns the topic and the keywords into a structured outline, and is required to propose an anchor-product query. Then a parallel research wave fires: a **catalog scout** searches the brand's live Shopify catalog for the products and collections the post should link, a **competitor researcher** maps how peer brands cover the topic so the post can be positioned against the gap, a **provenance researcher** pulls the material and era facts that make the writing credible, and a **hero image** agent generates the cover. Then the **writer** drafts from all of it. The draft goes to a **voice editor**, then a **schema builder** attaches JSON-LD so the post is legible to Google and to AI search. Nine narrow agents beat one general prompt for a simple reason. A single prompt asked to research, position, write, and format will do all four at a B-minus. Nine agents each do one thing, and each one is testable on its own. The pipeline ships with around thirty test files for exactly that reason. When a post comes out wrong, I can see which agent to fix. ### The two gates that make it an engine | Gate | What it checks | If it fails | | --- | --- | --- | | Anchor product | The catalog scout must return at least one real product from the live catalog. | The orchestrator stops and returns blocked_no_product_anchor. The post is not written. | | Voice | A rule-based linter checks banned patterns, then an LLM voice editor rewrites what it flags. | After three attempts the post is blocked as blocked_voice. | A generator writes whatever you ask. An engine refuses work that fails its rules. This one has two gates. The first is the anchor-product gate. The catalog scout has to return at least one real product from the live catalog. If it returns zero, the orchestrator stops and returns `blocked_no_product_anchor`. The post doesn't get written. This is the rule that blocks half the topic ideas that sound good in a brainstorm. "A guide to French country style" sounds like a post. If it can't anchor to a French country dining table the brand actually stocks, it isn't one. The second is the voice gate, and it has two layers. A rule-based linter checks every draft against a list of banned patterns: em dashes, decorative triplets, the corporate verbs, the AI-tell phrasings. Then an LLM voice editor rewrites whatever the linter flags. The draft goes back through the linter. If it still fails, the writer runs again with the failures as a hint. Three attempts, then the post is blocked as `blocked_voice` rather than shipped sounding like a machine. The voice gate is the automated version of a checklist I wrote by hand a while back, in [the post on spotting AI in your own writing](/blog/spotting-ai-in-your-own-writing/). The difference is that a checklist depends on the writer remembering to run it. A gate doesn't depend on anyone. ### What stays human The engine doesn't publish. It produces a Shopify article draft, unpublished, and stops at a review gate. A human approves, rejects, or reads the full HTML first. Approve, and it becomes a draft in Shopify. Nothing reaches the storefront without a person saying so. That's deliberate. The engine is built to remove the labor that doesn't need judgment: keyword pulls, catalog lookups, schema markup, voice cleanup, the first draft. It isn't built to remove the judgment. A home brand's blog is a brand surface. The last call stays with a person who knows the brand. This is the same shape as [the tracking stack I had an agent build under supervision](/blog/agent-built-my-tracking-stack/). The agent does the work. The operator owns the gate. The engine is faster than a writer and more consistent than a writer, and it's still not allowed to publish on its own. ### A generator without gates is worse than no blog Here is the case for the rules. A brand that buys a generic AI content subscription gets volume. Forty posts a quarter, each one a category explainer, none of them linking a product. That brand now has forty pages telling Google the site is a low-effort content farm. The blog doesn't just fail to help. It actively pulls down how Google reads the whole domain. The engine can't produce that outcome. A post with no product anchor doesn't get written. A post that sounds like a machine doesn't get shipped. The volume is lower than a generator's and every post points a reader at something the brand sells. That's the trade, and for an e-commerce brand it isn't close. ### The receipts The engine is built and tested, around thirty test files covering each agent. It runs from a backlog: one command pulls the next queued topic, drafts it end to end, and lands a reviewable Shopify draft. It has produced its first real batch, anchored to the brand's case goods and leather seating. The catalog those drafts point at is live at [emmagracehome.com](https://emmagracehome.com/) if you want to see it. A `--preflight` command verifies every API credential before the engine spends a token. The honest part. This engine drafts. It doesn't decide. Every post still passes a human review gate, and the storefront SEO foundation those posts point at shipped in May 2026, days before this post, so no ranking results were in when I wrote it. The build is real and dated. The outcome takes a quarter to read. ### What this is really about A blog folder asks what to write about next. An engine asks whether the topic earns a slot at all, and whether the brand survives the draft reaching the storefront. Those are different questions, and only the second one builds an asset. If you run a Shopify home brand and your blog is a folder, the place to start is the storefront under it: [a written audit, delivered in 24 hours, no call required](/audit-request/?type=catalog). Worked example: Emma Grace Home. Sixteen brand-voiced blog drafts for a Shopify furniture store, each anchored to a real product and checked by an automated voice linter. https://connercrowe.com/results/content-engine-home-brand-sixteen-posts/ Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## Nine GTM Tags in 90 Minutes (the build I used to do by hand) URL: https://connercrowe.com/blog/agent-built-my-tracking-stack/ Published: 2026-05-17 | Updated: 2026-10-22 Ninety minutes, nine GTM entities, one orphan cleaned up. The tracking-stack build I used to do by hand, run with API tooling on my own container. ### Quick Take Tag Manager work has been one of the moats agencies could charge a monthly retainer for. The configuration was finicky enough that founders couldn't handle it themselves and senior operators were a scarce resource. That moat is shrinking. In May 2026 I closed the tracking gaps on my own site in 90 minutes: three Data Layer Variables, three custom event triggers, three GA4 event tags wired to them, one orphaned trigger deleted, a new container version published. About ninety percent of the work ran with build tooling that has direct API access to GTM. The final publish click had to be done in the UI because of how Google handles OAuth scope verification. The senior operator who built the framework and the tooling that runs it are starting to do the same hour of work. ### The Setup Question I've been writing about [The Tracking Stack](/frameworks/tracking-stack/) for about a year. Eight layers, opinionated, documented as a reference architecture for Shopify and lead-gen sites. It's the framework I use on every client engagement. It's also what convinced me to start running the configuration with API tooling directly instead of just generating advice about it. The question I wanted to answer: can this tooling administer Google Tag Manager well enough that I'd run it on a paying client's container? Not "can it generate the right tag config." That's old news. The actual question is operational. Can it read what's already there, propose the right changes, build them, ship them, and recover when something breaks. The setup took maybe twenty minutes. The tooling runs a local server that wraps the Tag Manager API. I picked a community-maintained one rather than the hosted alternative because I'd rather not route an admin-level OAuth token through somebody else's infrastructure. That choice ended up mattering twice during the build. Once the server was installed and Google OAuth granted, the tooling had access to about a hundred Tag Manager operations: list accounts, read tags, create triggers, update variables, publish versions. From there it was a real ops session, not a demo. ### What the Agent Found Before It Built Anything First pass was the audit. List the accounts I have access to. Find the right one. Pull every tag, trigger, and variable in the container. Compare what was firing against what my site's `dataLayer` was emitting. Seven tags, five triggers, three Data Layer Variables. The audit took under a minute. The diagnosis was harder than I expected. My site fires six custom dataLayer events: `cta_click`, `calendly_open`, `lead_magnet_submit`, `contact_form_submit`, `case_study_view`, and `blog_post_view`. Three of those had GTM tags catching them. The other three were emitting into the void. `lead_magnet_submit`, in particular, fires every time someone downloads one of my free PDFs. That's the highest-intent event on the site. It was hitting the floor. The audit wrote up the gap, ranked the missing pieces, and asked which tier to ship. That conversation took two minutes. I picked the highest-priority tier: close the dataLayer gaps. New DLVs, new triggers, new GA4 tags wired to them. ### What It Built Three Data Layer Variables: `DLV - lead_magnet_name`, `DLV - form_type`, `DLV - page_path`. Each one reads the corresponding key off the dataLayer. Three custom event triggers: `CE - lead_magnet_submit`, `CE - case_study_view`, `CE - blog_post_view`. Each one fires when its named dataLayer event arrives. Three GA4 event tags: `GA4 - Lead Magnet Submit`, `GA4 - Case Study View`, `GA4 - Blog Post View`. Each one is wired to its corresponding trigger and forwards the event to the GA4 property already running on the site. One orphaned trigger deleted. There was a Scroll Depth 50% trigger sitting in the container connected to no tag. The audit flagged it and asked permission to clean it up. Everything went into the Default Workspace in a single review pass. I checked the diff, approved, and the publish step ran. The diff is not the proof. An API write can come back a success and land altered: I have sent four parameters to a Tag Manager tag and read three back, with `success: true` both times, which is [its own failure mode](/blog/fields-you-most-want-to-change-vanish/). What proves a batch landed is reading every touched object back. That's where it got interesting. ### The Three Things That Broke The first was a Windows filesystem quirk. The edit-and-restart loop hit a virtualized `AppData` folder that the writer process could touch but the running server subprocess couldn't read from. Half an hour of restart cycles before I noticed four stale Node processes, each loading the unpatched server. Killed them all. The fifth spawn loaded the patched file. The second was the server itself. Its `createVariable` function had a bug. It hardcoded the parameter shape to `{ key: 'value' }` regardless of the variable type. That works fine for Constants but breaks for Data Layer Variables, which require `{ key: 'name' }` and a `dataLayerVersion` parameter. Every DLV creation came back with `vendorTemplate.parameter.name: The value must not be empty`. Source read, bug identified, thirteen-line patch with three-branch logic for the parameter shape, permission to apply. I authorized. The patch went in, the next variable created cleanly, the build continued. I'll send the same patch back to the maintainer as a pull request when this post ships. The third was Google's own OAuth scope handling. The local server requested two scopes during the consent flow: `tagmanager.edit.containers` and `tagmanager.publish`. Both showed up in the auth URL. But when the publish step ran, Google returned `Insufficient Permission`. The edit scope was honored. The publish scope wasn't. My read is that publishing a Tag Manager version is one of the sensitive scopes Google flags for unverified OAuth apps, and an unverified app in testing mode can't actually use it even if the consent flow appeared to grant it. The build was complete. Nine new entities, all correctly wired. The only thing left was the publish button. I clicked it manually in the GTM UI. That took fifteen seconds. ### What This Means For Tracking Ops If you're a founder or operator waiting to see what fast in-house tracking ops looks like, this is the shape of it. Not the demo version. The build did the hour of work I'd otherwise be in the keyboard for. Audit, plan, build, surface the diff, approve. Three independent failure modes surfaced and got worked through. Source read, bug patched, the rest of the build continued. A permissions wall I had to step through manually, the rest of the work intact. What didn't happen automatically: crossing an OAuth verification boundary, deleting things I hadn't explicitly authorized, publishing a production version without my approval. Each refusal was correct. The safety net is doing what it should. What the next version of this looks like is fairly obvious. The bug gets patched upstream, OAuth verification gets sorted, and the same playbook runs on a real client engagement. The pieces that need a human shrink to "approve the diff" and "click publish." It since has. On [Emma Grace Home](https://emmagracehome.com/) I audited the GTM containers read-only through the same API before touching anything, and the audit found a server stack receiving zero ecommerce events and two GA4 installs fighting over one property. [The full write-up is here](/results/unassigned-revenue-dual-ga4-fix/). That's a meaningful shift for the tracking layer specifically. Tag Manager work has always been one of the moats agencies could charge for, because the configuration is finicky enough that founders couldn't handle it themselves. That moat shrinks every quarter now. The senior operator who built the framework and the tooling that runs it are starting to do the same hour of work. ### The Receipt Container `GTM-K5F244FD` moved to version 18 on May 17, 2026. The build added three Data Layer Variables, three triggers, three GA4 event tags, and removed one orphan. Version 17 was the pre-build baseline. The diff is the work. The full audit, plan, and build session is documented in this post. [The Tracking Stack framework](/frameworks/tracking-stack/) itself is unchanged. This post is the case study of operating it. If you want the same audit run on your own container, that's the [free custom audit](/audit-request/), back within 24 hours. ### One Footnote For Engineers The MCP server I used is `gtm-mcp` on npm, maintained by [pouyanafisi](https://github.com/pouyanafisi/gtm-mcp). The bug I patched is in `dist/gtm-client.js`, function `createVariable`. The fix is to branch the `parameter` array construction on `variableType` so `v` returns `[{ key: 'name' }, { key: 'dataLayerVersion' }]` instead of always returning `[{ key: 'value' }]`. I'll send the upstream PR. ### Keep going If this hit, the next three pieces in the same universe: - **[The Bing pixel that said YOUR_VALUE_HERE](/blog/bing-pixel-your-value-here/)**. The audit shape this build was modeling, eleven findings deep on an inherited Shopify stack. - **[The Workflow That Lets Me Ship a Site in 24 Hours](/blog/wispr-flow-claude-the-input-bandwidth-problem/)**. The input-layer workflow that makes fast in-house ops feasible inside a working session. - **[The Furniture Brand Whose Revenue Showed Up as Unassigned](/results/unassigned-revenue-dual-ga4-fix/)**. The client version of this playbook: a read-only container audit on Emma Grace Home that traced Unassigned revenue to three stacked failures. Free PDF: **[The 25-page Tracking Stack](/freebies/tracking-stack.pdf)**. No email gate. If you've never checked what your tags actually send against what the ad account counts, start with [the free ads audit](/audit-request/). Read-only access is enough, and I send the write-up within 24 hours. No call. Worked example: Emma Grace Home. Two GA4 installs on one property filed the store’s revenue as Unassigned, the server container received no ecommerce events, and Google Ads was optimizing toward add-to-carts. https://connercrowe.com/results/unassigned-revenue-dual-ga4-fix/ ## Why I Run Six Performance Max Campaigns Instead of One URL: https://connercrowe.com/blog/why-i-run-six-performance-max-campaigns-instead-of-one/ Published: 2026-05-17 | Updated: 2026-10-22 One Performance Max campaign is the default. The geo and value-band matrix I run instead on a mixed-catalog account with a hyperlocal pocket. ### Quick Take The default Performance Max setup is one campaign that absorbs the whole catalog. Google's interface defaults to it. Most agencies build it that way. For an account under $20K a month with one geography and one price band, the default is fine. For an account doing $50K a month or more across a mixed catalog with a hyperlocal pocket and a high-AOV tier, it isn't. The fix is a geo × value-band matrix. Two geography splits crossed with three product tiers. Six campaigns, each with its own budget, creative, audience seeds, and search themes. The signal stays clean. Smart Bidding optimizes against the right buyer in each tier instead of averaging across all of them. Here's how the matrix is structured and why each split exists. ### Why one Performance Max campaign quietly underperforms at scale Performance Max is a budget allocator wearing the trench coat of a campaign. It takes whatever budget you give it and decides which placements, audiences, products, and creative combinations get served. The optimization is opaque on purpose. That works when the inputs are homogeneous. One geography, one price band, one buyer archetype. The Smart Bidding model picks a CPA target, learns what converts, and gets better over six to ten weeks. It breaks when the inputs are heterogeneous. A $89 baby blanket and a $1,200 stroller travel through different research cycles, different competitive sets, different buying contexts. If both sit in the same campaign with the same tCPA, the model averages the signal. The blanket steals impressions from the stroller (cheaper clicks, faster conversions) while the stroller's true buyer journey gets undertrained. The dashboard still looks fine. The blended ROAS averages out. The next-90-days conversion model has no idea what to do with the high-AOV tier anymore. ### The two splits that matter The matrix I run has two dimensions. Both came from watching the same failure modes across accounts. **Split 1: Geography.** Most brands have one geography that punches above its weight. A flagship store, a regional concentration of word-of-mouth, a press hit in a specific market. If 30% of revenue comes from one state and the rest comes from the other 49, the buyer in that state has different intent (often store-visit-adjacent) and different competitive context (often less crowded). The bidding economy in that state is also distinct: less impression-share pressure, smaller pool, more responsive to creative variation. If you let national Performance Max bid on that state, you pay national CPCs to acquire local-intent buyers. You also fail to push the hyperlocal angle (local stock, ready to ship from the local warehouse, store pickup) that wins those impressions cheaply. The fix: isolate the high-density geography into its own campaign set. National Performance Max targets everywhere else. Local Performance Max targets that state. Different creative. Different budgets. Different bidding behavior. **Split 2: Value band.** A mixed catalog has at least three signal classes: | Tier | What it is | Why it gets its own campaign | |---|---|---| | Mid ($75-$500) | Accessories, smaller goods, gift-tier | Highest velocity, fastest learning loops, biggest pool of branded long-tail to capture | | Premium ($500+) | Furniture pieces, gear, big-ticket | Longer consideration window, different competitive set, needs distinct creative tone | | Bundles (multi-item, often $1K+) | Nursery sets, room-complete packages | The highest-AOV buyer cohort. Slower to convert, much higher LTV. Smart Bidding can only learn this if the signal isn't drowned out | If you put all three in one campaign with a unified tCPA, the model starves the bundle tier. The bundle tier needs a higher tCPA (because the AOV justifies it) and a slower learning expectation (because the conversion cadence is weeks, not days). One campaign with one target cannot serve all three. ### The 2 × 3 matrix in production Cross the two splits and you get six campaigns: | Campaign | Geography | Value Band | Why it exists | |---|---|---|---| | 1 | US National | $75-$500 | The volume engine. Fast learning, branded long-tail capture, the campaign that funds the others' learning periods. | | 2 | US National | $500+ | Premium goods to the national audience. Longer learning, higher tCPA, creative tone pitched to considered purchase. | | 3 | US National | Bundles | The high-LTV layer. Targets the cohort that buys a complete nursery or a complete room. Slow burn, big payoff. | | 4 | Hyperlocal (1 state) | $75-$500 | Local-intent mid-tier. Often the cheapest CPAs in the account because hyperlocal creative wins against national big-box. | | 5 | Hyperlocal | $500+ | Premium to local. Often store-visit-driving or pickup-adjacent. Creative emphasizes local availability and staff who know the product. | | 6 | Hyperlocal | Bundles | Smallest budget but often the highest blended ROAS, because local bundle buyers know the brand and the brick-and-mortar trust signal closes them. | Budget allocation runs roughly 65% national, 35% hyperlocal. Within each geography, the mid-tier carries 55-60% of the spend, premium gets 25-30%, bundles take the remaining 15%. Adjust by margin profile. ### What feeds the matrix Splitting Performance Max into six campaigns is useless if the inputs underneath are weak. The matrix depends on three pieces working in parallel. **Customer Match segments.** I seed each campaign with the customer-match audiences that match its tier. Past purchasers of mid-tier goods seed the national mid-tier campaign. Past bundle purchasers seed the bundles campaigns. Repeat-purchaser cohorts seed the premium tiers. The point isn't to target those audiences directly (Performance Max decides that). The point is to give Smart Bidding a high-quality "this is what good looks like" reference for each campaign. A mature account in this structure ends up with 20-30 segments stitched into the matrix. **Search themes, per asset group.** Google allows up to 50 search themes per asset group, so the ceiling depends on how many asset groups each campaign runs, not on the campaign itself. They should not be copy-paste across campaigns. The mid-tier national campaign gets themes around the queries actual mid-tier buyers run. The bundles campaigns get themes around "nursery set", "complete bedroom", "room package" intent. The hyperlocal campaigns get themes that include the state name plus product category. The search themes become the steering wheel that keeps each campaign pointed at its tier. **A clean product feed with custom labels.** Each campaign's asset group filters the product feed by custom label. `tier_mid`, `tier_premium`, `tier_bundle`. Without these labels you cannot cleanly assign products to the right campaign and the matrix collapses. This is feed work. It happens once at the Shopify layer and pays for years. ### What you need working before you build the matrix Don't build six Performance Max campaigns onto a broken foundation. The matrix needs: 1. **Server-side tracking that doesn't drop conversions.** If sGTM is misfiring on Shopify checkout extensibility, the matrix amplifies the dropout. Six campaigns all undertrained instead of one. 2. **Conversion value that excludes shipping and tax.** Smart Bidding optimizes against the signal. If the signal includes $40 of shipping per order, the model chases high-shipping orders. Audit Goals → Conversions → Purchase value before you split. 3. **Brand kept out.** A brand exclusion on every Performance Max campaign, and your brand name and misspellings as phrase negatives on a shared list. Otherwise the mid-tier national campaign eats brand long-tail and looks better than it is. Brand gets its own Search campaign only when a competitor's ad holds your brand search above the organic result. 4. **A real product feed with GTINs, custom labels, and titles that match buyer queries.** I've written about [why Performance Max gets credit for Shopping conversions](/blog/why-performance-max-gets-credit-for-shopping-conversions/). The same logic applies, more so, at this scale. The [Tracking Stack reference](/frameworks/tracking-stack/) covers the foundation layer in detail. If those eight layers aren't in place, the matrix will look great on the build day and underperform by week six. The [STACK Audit](/blog/google-ads-tracking-audit-guide/) is the five-pass walkthrough I run before any campaign architecture change at this scale. The first failure mode is almost always at the [conversion-value layer](/conversion-tracking/conversion-value-to-send-google-meta/): shipping and discounts baked into the number Smart Bidding is chasing. ### How to test the matrix on your account If you run one Performance Max campaign now and want to know whether the matrix is the right next move: 1. Pull the last 90 days. Segment by product. Look at the AOV distribution. If 60% or more of revenue comes from one tier and the rest splits across two more, the matrix will help. If your AOV is unimodal, it won't. 2. Look at geography. If one state, metro, or DMA accounts for 25% or more of your revenue, the geo split will help. 3. Look at your customer list. If you can build distinct customer-match segments by AOV tier (a "past purchaser of $500+" list with at least 1,000 members), Smart Bidding has something to learn from. Without enough audience density, the matrix is theoretical. 4. If all three boxes check, build in this order: feed plus custom labels first, then the matrix structure, then customer match seeding, then search themes per campaign. Two weeks of build. Eight to ten weeks of relearning. Don't measure performance against the prior campaign until week six. ### How this lands in the board deck A six-campaign matrix only earns the budget conversation if it produces a number the board can act on. The [board-deck quartet](/blog/board-deck-paid-acquisition-numbers/) is the format: contribution-margin ROAS, new-customer CAC, incrementality lift, payback weeks. The matrix shows up in slide two as "the strategic decision the board needs to make this quarter." Pair the architecture move with a [21-day geo holdout](/blog/21-day-geo-holdout-incrementality-test/) at the back end of week eight so the incrementality number lands in the next quarterly update. ### Keep going If this hit, the next two pieces in the same universe: - **[Why Performance Max Gets Credit for Shopping Conversions](/blog/why-performance-max-gets-credit-for-shopping-conversions/)**. The attribution misread that kills Shopping budgets prematurely. - **[The STACK Audit: the 5-layer pass I run before I touch a single bid](/blog/google-ads-tracking-audit-guide/)**. The audit that catches the foundation problems before they amplify across a six-campaign matrix. Free PDF: **[The Tracking Stack](/freebies/tracking-stack.pdf)**. Eight layers, no email gate. The foundation under every campaign architecture I run. If your account is past the point where one Performance Max campaign can carry it, [send me read-only access for the free ads audit](/audit-request/). I'll tell you in writing, within 24 hours, what the next architecture should look like. No call. ## Why I'm Still Writing This Blog in 2026 URL: https://connercrowe.com/blog/why-im-still-writing-this-blog-in-2026/ Published: 2026-05-17 | Updated: 2026-10-22 Plenty of senior operators moved off written essays. My reasons for staying with a blog, and what it does for the work. ### Quick Take Most of the senior paid-media operators I used to compare myself against moved away from writing long-form essays on their own sites. Some went to podcasts and LinkedIn. Some folded a solo newsletter into a multi-author brand. Some sold or left their agencies. Some started shipping software instead of essays. I'm staying. Here is why, and what that means for the founders who land on this site looking for a senior consultant who actually writes. ### The pattern When I looked at the operators I benchmark against in May 2026, very few were still publishing founder-written long-form on a site they own. The ones with an established audience had shifted their hours to audio, video, or short posts. The ones who had exited their agencies had little reason to keep a personal blog going. The play had shifted from "publish to build authority" to "own the listener" or "ship a product the buyer interacts with." I'm staying with the format. Here is why. ### Why most of them quit The senior-operator blog playbook stopped paying off for three concrete reasons: **1. The ROI of written essays collapsed for operators with an audience already large enough to sustain a podcast or video channel.** A weekly podcast episode reaches an audience that doesn't read blogs, and the production cost is lower per minute of attention than the same operator's writing cost per word. For operators with established names, the marginal hour is better spent recording than writing. **2. Agency exits closed the chapter.** For an operator who has sold or left the agency, the personal-brand blog as a tool to scale a one-person consultancy is irrelevant once the consultancy is sold or absorbed. **3. AI-generated content flooded the SEO surface.** Posts that used to rank for "Google Ads PMax audit" now compete against a pile of AI-written posts on the same query. The volume of AI content is so high that the senior-operator post is buried in noise even when it ranks. For an operator with a podcast or course alternative, the rational move is to abandon the SERP and own the listener instead. None of those three reasons applies to me. ### Why I'm staying I'm a one-person consultancy. Not pre-exit, not post-exit, no plans to either. The blog is the spine of the business for the exact reasons it stopped being the spine for the others: **I can't record fifty podcast episodes a year and also do the actual client work.** Operators who run a podcast alongside a business usually have a team behind it. I do not. A weekly podcast burns the calendar that should be on client accounts. A weekly blog post doesn't, because dictating an outline takes thirty minutes and the polish runs in parallel with the client work itself. **The founders I want to work with read.** The Shopify founder watching blended ROAS slip reads blog posts on a phone at 9pm. The service-business owner trying to figure out why qualified leads stopped coming in reads them on the commute. The CMO prepping a board pitch reads to sharpen a point before the meeting. The buyer I sell to is a reader. The buyer who only consumes this kind of work as a podcast is more likely somebody else's buyer. **The blog is the audit-bench reference architecture.** [The Tracking Stack](/frameworks/tracking-stack/) is an eight-layer framework. [The STACK Audit](/blog/google-ads-tracking-audit-guide/) is the five-pass methodology that verifies it. Both work because they are scannable, searchable, linkable artifacts a buyer can return to or send to a developer. A podcast can't do that work. A YouTube video can't do that work. The artifact has to be written. **The AI-flood is a moat the other way.** When the SERP fills up with AI-written posts that read identically, the human-written operator post becomes the rare signal. I run a [voice-audit checklist](/blog/spotting-ai-in-your-own-writing/) on every piece before it ships so the AI tells stay out. The result is that the posts read like one person wrote them, which is now the differentiation, not the table stakes. ### What this means for the buyer reading me If you landed on this site looking for a senior paid-media consultant, the field is smaller than you think. Many senior operators in the category are at capacity, have exited, or run an agency you would have to filter through a layer of account managers to reach. The operators still in it tend not to write, which makes them harder to evaluate before a call. What you can do here that you can't do with the others: - Read the [archive](/blog/) on conversion tracking, attribution, and the diagnostic patterns I run. Decide if my read on a problem matches yours. - Read [the meta-case-study](/results/rebuilding-my-site-with-claude/) on how this site was built so you know how I work, not just what I work on. - Read [the 150-SKU furniture catalog I launched solo](/results/150-sku-furniture-catalog-no-photographer/) for Emma Grace Home, the same way of working on a client's store. - Download [the Google Ads Setup Checklist PDF](/freebies/google-ads-setup-audit.pdf). No email gate. The only ask is on page 23, and it's the free written audit, not a call. Read 25 pages of my actual diagnostic shape and decide if you want the same read on your own account. You don't get that on a podcast feed. You don't get it on a multi-byline newsletter. You get it on a blog written by the person you would hire. ### What I'm changing in 2026 The posts you see today are not the final form. Two changes I committed to when I wrote this: **Named frameworks, not generic explainers.** [The Tracking Stack](/frameworks/tracking-stack/) is the architecture. [The STACK Audit](/blog/google-ads-tracking-audit-guide/) is the audit methodology. [The Operator Method](/process/) is the engagement shape. Future posts get tied to one of these named artifacts so the buyer who reads three posts in a row leaves with a working mental model, not three unrelated tips. **Operator Teardowns as a recurring series.** Quarterly account reviews, public, permission-granted. The structure peers use on podcasts, only written, scannable, linkable. The first one ships when the first client account clears the permission gate. ### The receipts The blog covers the tracking, lead-quality, margin, and anti-agency pillars, and the posts cross-link so a buyer reading one post lands on the next two in the same universe instead of dead-ending. I rebuilt the site itself in 24 hours using [the workflow documented here](/blog/wispr-flow-claude-the-input-bandwidth-problem/) so the writing surface scales with the way I work. The infrastructure compounds. The cadence is sustainable. The buyer who lands here knows what they're getting before they ask for anything. That's the case for staying with the format when the rest of the cohort has moved on. ### Keep going If this hit, the next three pieces in the same universe: - **[The STACK Audit: the 5-layer pass I run before I touch a single bid](/blog/google-ads-tracking-audit-guide/)**. The named diagnostic this blog is built around. - **[How I cut spend in half and doubled signups for a real estate law firm](/results/real-estate-law-google-ads-growth/)**. The shape every diagnostic post on the site tries to hit. - **[A 150-SKU Furniture Catalog, Built Solo in Under a Month](/results/150-sku-furniture-catalog-no-photographer/)**. The Emma Grace Home build: 150-plus products, every image rendered in-house, zero product shoots. Free PDF: **[The 25-page Google Ads Setup Checklist](/freebies/google-ads-setup-audit.pdf)**. The audit walkthrough behind every post in the tracking pillar. Every post here starts from an account I read. If you want yours read the same way, [send me read-only access for the free ads audit](/audit-request/). It's 5 to 7 pages, back within 24 hours. No call. Worked example: Emma Grace Home. A furniture store I launched solo on Shopify: 150+ products, every image rendered in-house, live in under a month. Monthly sales went from $5,387 in April 2026 to $56,182 in September. https://connercrowe.com/results/150-sku-furniture-catalog-no-photographer/ ## The Bing pixel that said YOUR_VALUE_HERE URL: https://connercrowe.com/blog/bing-pixel-your-value-here/ Published: 2026-05-16 | Updated: 2026-10-22 One afternoon in an inherited Shopify tracking stack: two GA4 properties, four duplicate scripts, and a Bing pixel firing placeholder text since November. ### Quick Take A six-month-old Shopify + Stape Pro stack I audited in May 2026 was firing placeholder text at Microsoft Ads for three conversion goals. The literal string `YOUR_VALUE_HERE`. Untouched since the snippet was pasted in. That was finding number four. There were seven more, including a Meta CAPI access token sitting in plaintext in GTM version history that anyone with read access could lift. None of it was on fire. The store was making money. PMax was running. Dashboards looked tidy. This is what an inherited tracking stack looks like after six months of drift. Below is the full punch list, and the structural reason agencies miss it. ### The store had been running this *stack* for six months I gave myself one afternoon to audit a six-month-old Shopify tracking stack. The store was paying an agency. They were paying for Stape Pro. They were running real PMax spend across Google, Meta, and Bing. From a distance the setup looked clean. The first thing I found, forty seconds in, was a Meta Conversions API access token hardcoded in plaintext inside a GTM tag from 2024. Still the live token. Still powering every server-side Meta purchase event in production. Sitting in version history that anyone with read access to the workspace could scroll back to and copy. The fourth thing I found was three Microsoft Ads conversion tags with the literal string `YOUR_VALUE_HERE` in the conversion-action ID fields. For six months, every conversion the store had sent to Bing carried placeholder text from the snippet documentation. Untouched, and all wrong. The PMax was running. The campaigns were "optimizing." Whatever Bing thought the brand was selling, it wasn't. This is a Shopify baby and parenting brand. Real revenue, real ad spend. The kind of operation where the tracking layer is supposed to be the boring part. They had inherited the GTM build from a previous agency relationship and had been told everything was in good shape. I gave myself an afternoon to look. Web container, server container, Shopify theme, Stape dashboard, GA4, Events Manager, Microsoft Ads. The pass any senior operator should do on a stack they didn't personally build. What was in there, in order of how angry it made me: ### Two GA4 properties were receiving production *data* The real one, the one the marketing team was reporting on, and a phantom property called `G-23179768` that nobody on the team recognized. Both collecting pageviews. Both collecting purchases. The real property was undercounting because every event was being split. The phantom was a black hole nobody was looking at. I asked who set it up. Nobody knew. It had been collecting events for months. The fix was in a place you wouldn't find unless you went looking. Shopify admin → Sales channels → Google & YouTube → Settings → Data sharing and tag management → Additional conversion measurement settings → Google tags. Six menus deep. The phantom GA4 was bound there as a "manually added" tag with a trash-can icon next to it. Click trash. Save. Phantom property stops getting fed. ### The Custom Loader was paid for and *idle* Stape Pro tier with a Custom Loader subdomain configured at `load.gtm.[brand].com`. DNS pointed at the right place. SSL valid. Hitting the URL directly returned a 200 with a real `gtm.js` payload. The custom loader had been live for ten days. Stape Analytics showed 0% adblock recovery over those ten days. The reason was simple. `theme.liquid` was still loading the container from `www.googletagmanager.com` directly. uBlock Origin and the standard blocklists caught it on every visit and that was the end of GTM for anyone with an adblocker. The entire reason to pay for Stape Pro is so the loader script comes from your own domain and adblockers stop blocking it. The store was paying for the upgrade and not using it. The fix is a find-and-replace, but in more places than you would expect. `theme.liquid` has the GTM bootstrap snippet's `j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;` line plus the `