I Built a Tool to Track AI Brand Visibility — Here's What I Learned

A client asked me a question I couldn't answer with any tool I already had: "If someone asks ChatGPT for a recommendation in our space, do we even come up?" Their Google rankings were fine. Their traffic was fine. But none of that told them anything about what was happening inside an AI chat window, where a growing share of their prospects were actually starting their research.

Building a Way to Actually Measure It

So I built GEOGym to find out. The idea is simple: run the same buying-intent questions a real prospect would ask ("best tool for X," "who does Y for small businesses") across ChatGPT, Perplexity, and Gemini on a schedule, and track whether the brand gets mentioned, how it's described, and who it's mentioned alongside.

AI Answers Aren't Stable

The first thing I learned is that these answers are not stable. The same question asked a week apart can surface a different set of brands entirely. Traditional SEO rank tracking assumes a result holds roughly steady until something changes; AI answers behave more like a rolling sample, which means a single check tells you almost nothing. You need repeated measurement over time before a pattern means anything.

Where the Mentions Actually Come From

The second thing I learned is where the mentions actually come from. In every case I looked at, a brand's AI mentions traced back to a small number of sources the model seemed to trust: a handful of comparison articles, a Reddit thread with real user opinions, a review site with structured data. Traditional link-building and directory submissions had almost no visible effect. What moved the needle was getting mentioned, favorably and specifically, in the kind of content a language model would plausibly cite as a source.

That reframes the whole problem. "AI SEO" isn't really a new discipline bolted onto old SEO, it's closer to old-fashioned PR and community presence, aimed at a new kind of reader. The brands showing up inside AI answers are the ones that show up in the conversations happening around their category, not the ones with the most backlinks.

What Changed for the Client

For the client, that meant deprioritizing a few tactics that used to make sense and putting real effort into two things: getting into the comparison content their prospects were already reading, and making sure their own site had clear, quotable, factual descriptions of what they actually do; not marketing copy, but the kind of plain sentence a model could lift directly into an answer.

If you want to check this for your own brand without building a tool first, there's a free version of the first step in a follow-up post: open ChatGPT, ask it the exact question a prospect would ask, and read the answer like a stranger would. That fifteen-minute check tells you more about where you actually stand than a month of Google rank tracking will.