How to measure AI visibility
By the Why Won't AI team · Published · Last updated
As of November 2026, AI visibility is the share of answers to a fixed set of buyer questions that name or cite your product, on each AI assistant that matters to you. Measure it with the same questions, the same assistants and the same schedule every time, and show what each number is counted out of.
Decide what the number is for
A visibility number must support a decision: which page to write next, which assistant to work on, or whether a change worked. If no decision depends on the number, it is a report, not a measurement.
Count four things separately
| Measure | What it counts | Counted out of |
|---|---|---|
| Mention | The answer names your product | Answers that came back |
| Citation | The answer links to a page on your site | Answers that came back |
| Position | Where your product is in the list of products named | Answers that name you |
| Accuracy | What the answer says about you is correct | Answers that name you |
Do not merge these into one score. An answer can cite your page and not name you, or name you with the wrong price. One score hides which problem you have.
Keep the question set fixed
- Use questions that real buyers ask, in their words.
- Keep questions that name your brand in a separate group from questions that do not.
- Do not edit a question after the baseline. Add a new question in its place and keep the old one in your records.
- Ask on the same assistants, in the same country, on the same day of the week.
Count failures as failures
Some requests fail. The assistant times out, or Google shows no AI Overview for that search. Record these as no answer. They are not answers that left you out, so they must not lower your rate.
Treat repeat runs with care
AI answers change from run to run. Several runs of one question tell you how stable that answer is. They are not several buyers. Report results for each question and each assistant, not only as one total.
An example from our own baseline
On November 3, 2026 we asked 15 buyer questions in our own category on ChatGPT, Gemini, Google AI Overviews, Claude and Perplexity, once each. 74 of the 75 answers came back; one Perplexity request failed. Our product was named in 0 of the 74 answers. Otterly.ai was named in 22, Peec AI in 20, Ahrefs in 15 and Profound in 14. That zero is our starting point, and we measure every change against it.
Connect visibility to visits with care
Visibility is not traffic. Look at visits from AI assistants in your analytics and at clicks in Search Console next to your visibility numbers. Google counts clicks from AI Overviews and AI Mode inside your normal web search traffic in Search Console, so you cannot see them as a separate line there.
Questions people ask
How often should I measure?
Weekly is enough to see a trend for most sites. Measure on the same day each week.
Can I measure by hand?
Yes, for a few questions. Save the full answer and its links each time, in the same way.
Is one score enough?
No. Keep mentions, citations, position and accuracy apart, and report each assistant on its own.
What this does not tell you
Which questions your own buyers ask, how many runs make a number stable in your category, and which page to change first.
Sources
- Google Search Central: AI features and your website: how AI Overviews and AI Mode choose pages, and how Search Console counts that traffic