# What We Learned Running AI Search on Our Own Site for 150 Days

> We ran Fame Search on fame.so from June 1, 2026. What we did, what the numbers show and do not show, and why this is not a client result.

Source: https://www.fame.so/post/ai-search-own-site-150-days  ·  Tom Hunt  ·  November 18, 2026

## Key takeaways

- On fame.so, the share of tracked AI answers citing the site went from 38% on June 1 to 76% on October 30, 2026.
- This is our own site and not a client result. Fame has no client results to report yet.
- AI assistant visits stayed small at 41 in 28 days, so referral traffic is a poor scoreboard.
- Other things changed over the same period, so the numbers show what happened after we started and do not prove cause.
- Take a baseline before you change anything, and keep the question panel fixed.

Before we sold Fame Search to anyone, we pointed it at our own site. Over about 150 days, the share of tracked AI answers citing fame.so went from 38% on June 1 to 76% on October 30, 2026. The numbers are in the [case study](https://www.fame.so/search/fame-so-case-study); this is the plainer account from the person running our [AI search agency](https://www.fame.so/search), including what the numbers do not prove.

One thing needs saying before anything else. This is our own site. It is not a client result, and Fame has no client results to report yet. A company testing its own product on its own website is evidence that the method can be run, on one domain. It does not tell you what will happen on yours.

## Why did we run it on our own site first?

Because we wanted to watch it run on a site we knew well before selling it to anyone. Fame has spent six years producing podcasts for B2B companies, and fame.so has six years of content behind it. That made it a convenient place to test and, as I will come to, an unfair one.

It also meant we were the client. Every step that we now ask a client to do, we had to do ourselves: agree the questions, read the weekly report, review each change and approve it.

## What did we actually do?

We followed the same six steps we now sell, starting on June 1.

- **We mapped the questions our buyers ask.** We agreed a set of core buyer questions, the ones that lead to a sale, drawn from our sales calls, forms, podcast and market. That set is what we tracked.
- **We measured where we stood, every week.** Each week we asked ChatGPT, Perplexity, Claude and Gemini every core question, several times, and recorded who got named and cited: our share of answers per engine, which competitors were cited in our place, and the exact pages being quoted.
- **We fixed what held the site back.** Pages AI engines could not read, missing titles and descriptions, missing structured data, slow pages, weak internal links, and an llms.txt file that tells AI crawlers what the site is about.
- **We built the pages that win the answers.** For each question we were not winning, we compared our page with the pages being cited, then improved the existing page or built a new one. The page types in the method are direct-answer articles, solution pages for a specific buyer problem, "Best X" and alternatives lists, and free tools.
- **We approved everything first.** Nothing went live without sign-off. Once a page was live we notified search engines straight away through IndexNow and tracked whether Google had indexed it. Google does not take part in IndexNow, so that check is separate.
- **We kept winning pages winning.** The method watches pages close to page one and refreshes pages every 90 days.

The rhythm was one action on the site every day: a fix, a new page or an improved one. Pages were written from our own material, meaning our site, our case studies, our proof points and our podcast transcripts, with each claim checked against a source.

## What happened to the numbers?

These are the figures from our portal at the October 30, 2026 snapshot. They are the only ones I will quote.

MeasureFigureNotesShare of tracked AI answers citing fame.so38% on June 1, 76% on October 30Same panel of questions and engines on both datesPages graded A1,416 of 1,610Scored for health and AI-readabilityGoogle impressions42,966Last 28 days at the snapshotVisits from AI assistants41ChatGPT, Perplexity and others, last 28 days

The case study page also carries a chart of daily AI citations and the number of our pages the engines drew on, running from November to October, with a line marking the day the first action shipped. I would rather you look at it than take my description of its shape.

## How did we measure it?

The 38% and the 76% come from the same panel of buyer questions, asked of the same engines, on both dates. AI answers change between runs, so each question is asked several times. Share of answers here means the percentage of those tracked answers that cited fame.so.

We have not published the size of the panel yet. A percentage without the size of the panel behind it is harder to judge, and I would say the same about anyone else's number.

Two more things about the measurement. We tracked ChatGPT, Perplexity, Claude and Gemini. The figure is citations of our site, which is not the same as being named or recommended. A page can be cited without the brand being named, and the two should be kept apart. [Our tracking method](https://www.fame.so/post/track-ai-visibility) explains how to run the same kind of panel yourself in a spreadsheet.

## Which number stayed small?

Traffic from AI assistants. Forty-one visits in 28 days is a modest number, and it is worth seeing next to the others. Over the same 28 days Google showed our pages 42,966 times.

If I had judged the whole exercise by referral visits, I would have stopped. I think that would have been the wrong call, because many buyers read an AI answer without clicking anything, and referral counts understate that. But I cannot prove the influence from analytics alone, and I am not going to pretend the 41 is bigger than it is. The measure we actually care about is a buyer who finds us through an AI answer and goes on to book a call, or who says "ChatGPT recommended you" when asked how they heard about us.

The page grades tell a similar story of unfinished work. 1,416 of 1,610 pages graded A leaves 194 that are not. The site work is not finished.

## What does this not prove?

Several things, and they matter more than the headline number.

- **It does not prove cause.** Other things changed over the same period. The chart shows what happened after we started, and it cannot show that every gain came from Fame Search.
- **It is one site.** fame.so was already established, with six years of content behind it. A newer site, or one in a more crowded category, may move more slowly or differently.
- **It does not isolate any single action.** We publish an llms.txt file, for example, and I cannot show that it contributed to a single citation, because its effect cannot be separated from everything else on the site. The same is true of most individual fixes.
- **It is our own panel.** We chose the questions. They are the questions that lead to a sale for us, which is the right set to track, but a different panel would give a different percentage.
- **It is not a forecast.** Results depend on your market, your site and who else is being cited. We do not guarantee AI mentions.

I would apply the same tests to any agency's proof, ours included. Ask whether the result is a client's or the agency's own, what the baseline and period were, and how variation between runs was handled. Those questions are in [how to compare AI search agencies](https://www.fame.so/post/compare-ai-search-agencies).

## What would we tell a B2B company starting today?

These are the general lessons I am comfortable drawing. They are opinions formed by doing the work, and the specifics behind them belong in the section above.

**Take the baseline before you touch anything.** The only reason I can write "38% to 76%" is that we measured on June 1, before the first action shipped. Without that, we would have a nice-looking number and no idea where it started.

**Fix the panel and leave it alone.** The comparison works because the questions and engines were the same on both dates. Every time you swap a question you weaken the trend line.

**Do not use referral traffic as the scoreboard.** Ours would have shown very little. Track share of answers, and add a "how did you hear about us?" question to your forms.

**Plan for the approval work.** One action a day means something to review every day. It protects accuracy, and it makes whoever approves part of the delivery process. Decide who that person is before you start.

**Write from material only you have.** Transcripts, case studies and proof points give a page things to say that a competitor's page cannot. A page built only from the open web tends to repeat what is already out there.

**Read each engine on its own.** We tracked four, and the method reports share of answers per engine for a reason: a blended figure can hide an engine where you are absent. Our headline number is a summary, and the weekly report underneath it is where the work gets decided.

**Expect to be unable to attribute it cleanly.** You will change many things at once, as we did. Be honest about that when you report upward, and let the trend over months carry the argument.

The broader method, beyond our one site, is in [the complete guide to B2B AI search](https://www.fame.so/post/b2b-ai-search-guide), and the thinking behind why we built the service is in [the question is no longer "do we rank?"](https://www.fame.so/post/ai-search-named).

## More on B2B AI search

- [B2B AI Search: The Complete Guide (2026)](https://www.fame.so/post/b2b-ai-search-guide)
- [The 13 Best AI Search Marketing Agencies for B2B (2026)](https://www.fame.so/post/best-ai-search-agencies)
- [What to Look For When Hiring an AI Search Agency (12 Questions to Ask)](https://www.fame.so/post/compare-ai-search-agencies)
- [13 Tactics to Take Your AI Search Campaigns to the Next Level](https://www.fame.so/post/ai-search-tactics)
- [Your Buyers' Shortlist Is Written Before They Visit Your Site](https://www.fame.so/post/ai-shortlist-before-site-visit)
- [Why Most AI Visibility Dashboards Measure the Wrong Thing](https://www.fame.so/post/ai-visibility-dashboards-wrong-metric)
- [Third-Party Lists Are the New Backlinks](https://www.fame.so/post/third-party-lists-new-backlinks)

**Want your company named in AI answers?** [Fame Search](https://www.fame.so/search) is our AI search agency for B2B companies: weekly tracking across four AI engines and one action on your site every day, for $2,000 a month.

## FAQ

**Is this a client result?**

No. This is our own site, which we started running Fame Search on June 1, 2026. Fame has no client results to report yet.

**How was the 38% to 76% change measured?**

Across the same panel of buyer questions and engines on both dates. AI answers change between runs, so each question is asked several times.

**Will another company get the same result?**

Not necessarily. Results depend on your market, your site and who else is being cited. We do not guarantee AI mentions.

**Which engines were tracked?**

ChatGPT, Perplexity, Claude and Gemini.
