How to measure your brand's visibility in AI search
By Ben Frampton · · 5 min read
Short answer: measure AI visibility by asking AI assistants the questions your buyers ask, many times, and counting how often your brand is mentioned, recommended and cited compared with competitors. Report each as a rate over valid answers, per assistant and pooled, with the sample size and a margin of error beside it.
What exactly are you measuring?
In classic search you track a position: you're third for a keyword. AI answers have no positions. Each answer is a short piece of writing that may or may not name you. So AI visibility is a set of rates: out of all the answers to a set of questions, in what share did a given thing happen?
The four core metrics
| Metric | Question it answers | How to calculate |
|---|---|---|
| Mention rate | How often does the assistant name us at all? | Answers that mention the brand ÷ valid answers |
| Recommendation rate | How often does it suggest us as a good choice? | Answers that recommend the brand ÷ valid answers |
| Citation rate | How often does it link to our site as a source? | Answers citing one of our URLs ÷ valid answers |
| Share of voice | How much of the conversation do we get against competitors? | Our mentions ÷ mentions of all tracked brands |
A few definitions matter:
- Mentioned vs recommended. "Brand X is popular but expensive" is a mention. "For a small team, I'd suggest Brand X" is a recommendation. They often differ a lot, and recommendation is closer to what drives sales.
- Valid answers. An answer that failed to load, timed out or came back empty is excluded from the denominator. Counting it as "not mentioned" makes you look worse than you are.
- Citations. Only some assistants show sources, and some only for some answers. Compare citation rates within an assistant, not across them.
Which assistants should you measure?
Measure the ones your buyers use, and measure them separately. As of October 2026 the main ones are:
- ChatGPT
- Google AI Overviews (the summary above some Google results)
- Google AI Mode (Google's conversational search)
- Gemini
- Perplexity
- Microsoft Copilot
- Claude
They use different models, different search indexes and different rules for citing sources. A brand can be strong on Perplexity and nearly absent on ChatGPT. A pooled number across all of them is useful as a headline, but always look at the breakdown beside it.
What questions should you ask?
The questions you track decide what the numbers mean. Group them by intent (what the buyer is trying to do) and keep branded and non-branded questions separate:
- Non-branded: "What's the best payroll software for a 20-person company?" This is where new customers discover you.
- Branded: "Is Brand X good for payroll?" This tells you how you are described once someone already knows your name.
Mixing the two inflates your numbers, because assistants nearly always mention a brand that is named in the question. We cover how to build a fair question set in how to choose which prompts to track.
How often should you collect answers?
Often enough to have a useful sample in your reporting window. As a rough guide:
- Collect answers to two or three phrasings of each question, on each assistant, several times a week.
- Report over a rolling 28-day window, which smooths out daily swings.
- Show "insufficient data" rather than a number when you have fewer than about 30 valid answers, or when the margin of error is wider than ±15 points.
Why one AI answer tells you almost nothing explains where those thresholds come from.
Common mistakes
- Reporting a percentage with no sample size. "42% visibility" from 7 answers is mostly noise.
- Choosing flattering prompts. If the person tracking the numbers also picks the questions, they can drift towards ones you already win. Agree the question set first and keep it visible to everyone.
- Counting failures as misses. See valid answers above.
- Comparing assistants as if they were one. Pool them deliberately, not by accident.
- Stopping at the number. A rate tells you where you stand. It doesn't tell you why, or what to fix. For that, look at which sources the assistants cite and whether you have a page that answers the question at all.
What should you do with the numbers?
For each question where you are weak, work out the first thing that's missing:
- No page answers the question. Write one. See how to write pages AI assistants can quote.
- The page exists but doesn't rank for the searches the assistant runs. Work on ranking for those specific queries.
- Your page is cited, but competitors are the ones recommended. Earn mentions on the third-party sources the assistants rely on. See the sources AI assistants cite.
Frequently asked questions
Can I measure AI visibility with Google Search Console?
Partly. Google includes AI Overviews and AI Mode traffic in the Search Console Performance report, but doesn't separate it from regular results there. It tells you about clicks from Google, not how often you are mentioned or recommended in the answer text.
How is share of voice different from mention rate?
Mention rate compares you with all answers. Share of voice compares you with your competitors. You can have a 20% mention rate and a 50% share of voice if competitors are rarely mentioned either.
Can I track AI visibility manually?
For a first look, yes: ask ten buyer questions in three assistants and write down who is mentioned. For anything you want to track over time, the number of answers needed quickly makes manual tracking impractical.