Step by step
Seven steps, the same every time.
Agree the questions
At least ten questions your buyers ask, in their words, agreed with you before testing. More if you sell into several markets.
Choose the engines
ChatGPT, Perplexity, Gemini and Claude, plus Google's AI Overviews at the top of search results. We add others, such as Microsoft Copilot, if your buyers use them.
Run each question repeatedly
At least five runs per question per engine, spread over different days and times of day, in temporary or logged-out sessions.
Record what AI says
Whether you are named, where you appear, who is named instead, and which pages and sources the answer quotes.
Check what AI reads
Whether AI crawlers can reach your pages, and how AI describes your company when asked about you by name.
Rank the fixes
Changes to your site, your listings and your outside mentions, ordered by likely impact and effort.
Re-test
The same questions after the fixes go live, reported side by side with the first results, by market and by engine.
Why it's built this way
Lessons from our own audits.
AI answers vary a lot between runs and between engines. In our audits so far, the same company has been named often by one engine and never by another, for the same question. That is why we report rates per engine instead of one overall score.
The most common technical problems we find are sites that AI crawlers can't read: pages built in JavaScript that arrive empty, figures that only appear after an animation, and firewall rules that quietly block AI crawlers while search engines still get through. These are quick to fix and come first in the fix list.
The second most common problem is an out-of-date or inconsistent description. When a company changes its market or its services, some AI engines keep describing the old business for weeks. One clear description, repeated on every page and in the site's structured data, helps them catch up.
Questions
Short answers.
How do you measure AI visibility?
We ask each buyer question at least five times, on different days and at different times of day, in each AI engine, and record whether you are named. We report it as a rate, for example named in 3 of 5 runs, with a likely range, rather than a single screenshot.
Why ask the same question more than once?
AI answers change from one run to the next. A single test can show you named by luck or missing by luck. Repeated runs show how often you are really named.
Do you test in a clean session?
Yes. We use temporary or logged-out sessions where the engine allows it, so past chats and saved memory don't affect the answers, and we note how each engine was set up in the report.
What counts as being named?
Your company or product has to appear in the answer text itself. A link in the sources alone doesn't count, and neither does an answer that only mentions you because the account knows who is asking.
How do you know if a change worked?
We re-run the same questions after the changes go live and compare the rates. With small samples we say how far a result has to move before it counts as a real change.
Find out what AI says about your market.
Send us your website and the kind of customer you want more of. We'll reply with the questions we'd test.
Request an auditA fixed-price audit. No retainer needed.