How Long Does AI Search Optimization Take? I Checked at 30 Days

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 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.

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 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 template. For how often AI answers name a business at all, see my AI answers field report.

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