How we measure, from the question asked to the number you act on.
This page is the whole method: which questions, which AI assistants, how often, and what each number counts. We publish all of it, because a number is worth acting on when you know exactly what it is made of.
The AI assistants we ask
We measure ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews.
Every question we ask names the city you're tracking, on all five AI assistants. Google also tailors its results to where the person searching is located, so for Google we run the search from that city as well, and read the AI Overview it returns.
Google shows an AI Overview for some searches and not others. A search without one is recorded as its own outcome and kept out of your denominator, so only searches that had an answer count toward your results.
We ask through APIs, so every week is comparable
We ask each AI assistant the same questions a customer would, through an API, so the same question asked next week is comparable to this week's. That repeatability is what makes a trend line mean something. An API answer is usually more concise than the one someone sees in the app, and cites fewer sources, so read a count here as the pattern across many answers rather than a record of one person's screen.
The same question gets different answers
Ask an AI assistant the same question twice and the answer changes. Measured across 1,622 answers collected on 18 and 19 Aug 2026: when one answer names a business, asking again names it again about four times in five on Perplexity and about half the time on ChatGPT. That is why every question is asked three times and every answer is counted. One answer on its own can mislead; three make the count reflect what each AI assistant does.
It is also why every figure on this site is a count across many answers rather than one of them, and carries its own denominator, so you can see how wide the evidence is before you act on it.
What our figures cover
Our published figures come from thirteen local industries, measured on Perplexity, Gemini and Google AI Overviews. Ten are written up on our blog, and every one has its own page.
How often we ask, and how much
Every question is asked three times, every week, because one answer on its own can mislead, and weekly is often enough to see a real change. ChatGPT and Claude are asked the four questions where someone is choosing who to call, rather than all twenty. Measuring all five AI assistants on the same schedule matters more than asking each of them every question.
We measure weekly rather than daily. Answers vary enough between identical questions that a daily chart would mostly show the AI assistant changing its mind rather than a change in your visibility.