Most “AI SEO” numbers don’t survive a second look. Here is exactly how we measure whether AI assistants recommend a business, what the numbers do and don’t prove, and what we will never claim.
Ask ChatGPT the same question twice and you can get two different lists of businesses back. Ask it again tomorrow and the list can change again. AI assistants don’t give one fixed answer to a question. They generate a fresh response each time, drawing on shifting context.
That means a single check is close to meaningless: you ask once, note whether your business appears, and call that the result. It tells you what the AI said in that one moment, not what it actually does most of the time.
We treat a single AI answer the way a pollster treats a single voter: informative on its own, not a result you can act on. So instead of asking once, we sample.
We use a set number of realistic questions a customer might actually ask (the kind of thing they’d type into ChatGPT or ask Perplexity), and we tell you exactly what those questions are. Nothing hidden, nothing cherry-picked after the fact.
We run each question a set number of times and count how often your business shows up in the answer versus how often it doesn’t. One run is a coin flip; a properly sized sample is a real signal.
There is no such thing as an “AI rank.” AI assistants don’t publish a rankable list the way a map or a search results page does. What we can honestly measure is how often you show up out of a disclosed number of runs, the range that number could realistically fall in (the confidence interval), and how consistent the answer was run to run.
Say we check a question 20 times and a business appears in the answer 3 times, while a well-known competitor appears 18 times out of the same 20 runs. That is a real, useful gap. It’s exactly the kind of thing an illustrative example like this shows: the method, not a specific client’s guaranteed number. Your own report comes with its own error bars, built from your own runs.
If you look at your website analytics and see a trickle of visits credited to AI assistants, treat that number as a floor, not the real total. By most industry estimates, around 70% of AI-driven visits land in your analytics as “Direct” because the AI tool strips the referrer, so the true number of people an AI sent you is almost always higher than what your analytics shows.
“Mentioned” and “cited as a source” are not the same thing. An AI answer can name your business in passing without ever linking to your site or treating you as the source of the information. Both are worth knowing, but they answer different questions, and we report them separately rather than blur them into one score.
And a one-month change that sits inside the error bars isn’t a real result yet: it’s noise. We’ll tell you when a shift is big enough to mean something and when it isn’t, even when the second answer is less exciting.
This is the selling point, not an apology. A lot of “AI SEO” reporting is round numbers with no source, no sample size, and no date attached. We’d rather show you less and have every number hold up.
Free audit: about two minutes, no credit card.