What llms.txt Does, and Whether Your Business Needs One

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, 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.

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 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, Anthropic and Perplexity 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, 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, 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 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.

How I write one

This is the same order I use when an 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 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 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. After that, fix your facts at the source and get named on the sites the answers already cite.

Frequently asked questions

See how I check what AI assistants tell people about a business.