How we measure AI visibility

We track the questions a real buyer types into an AI assistant when choosing a business like yours — “best med spa in Scottsdale for Botox?”, “which CRM is best for a small sales team?”. When you sign up we suggest a set based on your website, category and location, and you can edit, remove or add your own at any time.

For local businesses most questions include your city, because that's how people ask and AI answers change a lot from one city to the next. For software and online brands the questions focus on category, use case, audience and budget instead.

Every weekly scan asks ChatGPT, Gemini, Perplexity and Claude. We query them through their official APIs with web search turned on — so, like the apps your customers use, they look things up before answering and cite the pages they used. Perplexity is always web-connected.

The free check on our homepage is a shorter version: six questions put to ChatGPT, Perplexity and Gemini, so it finishes in under a minute.

Ask an AI assistant the same question twice and you can get two different shortlists. One answer is an anecdote, not a measurement. So each scan asks every question on every engine up to three times.

For the per-question view we take a majority vote: you count as recommended for a question on an engine when you appear in at least half of its answers, and the same rule applies to each competitor. Your score is calculated over every individual answer, so it reflects how often you really come up.

To keep each scan finishing in time, repeats are fitted to a budget of 240 answers: with up to 80 question–engine pairs each is asked up to three times, with up to 120 up to twice, and larger question sets get one sample per question and engine. Every question is still asked on every engine — large sets are already stable through sheer volume.

We look for your business name, any alternative names you add (“Smile Studio” for “Smile Studio Dental Group”) and your domain, matched as whole words so “Ace” doesn't match “place”. The same happens for every competitor you track.

Your position is where you appear among all the businesses named in that answer, in order of first mention — being named first is worth more than being named fifth. We also keep the sentence that mentions you, so you can read exactly what AI said.

One number, built from three parts:

• Mention rate — the share of answers that name you. This is most of the score, because being in the answer at all is what matters most.
• Position score — when you are named, how early: 1st place counts fully, 2nd counts 80%, down to 5th at 20%; lower than that counts nothing. Averaged over the answers that name you.
• Engine coverage — the share of engines that named you at least once. Being recommended by all four beats being a favourite of just one.

The formula is fixed and deterministic: the same answers always produce the same score. If an engine fails to answer, that answer is left out of the mention rate rather than counted as a miss.

When an engine searches the web, it tells us which pages it used. We collect those citations across every answer and show which websites AI relies on for your questions — and, more usefully, the pages it cites in answers where a competitor is named and you aren't. Those pages are usually the shortest path to getting recommended.

After each weekly scan, Pro and Agency accounts (and trials) get five concrete actions — one each for listings, content, technical, community and reviews. They're built from that week's evidence: the questions you're missing, who is named instead, and the exact pages AI cited there. Not generic SEO advice.

• AI answers vary. Repeated sampling and weekly tracking smooth this out, but no measurement of AI can be exact. Watch the trend, not a single week.
• APIs aren't identical to the apps. We query the engines through their APIs. The consumer apps can use slightly different models, settings or personalisation, so an answer you see in the ChatGPT app may differ a little from ours.
• No ranking guarantees. Nobody controls what AI recommends — not us, not anyone selling you a guarantee. We show where you stand, what's driving it and what to do next. Changes to listings, reviews and content typically take weeks to show up in answers.