How to Measure Your Brand's AI Visibility

AI visibility is measured as a rate, not a rank. You take a fixed set of questions your buyers ask, run each one through ChatGPT, Gemini, Perplexity, Claude and Google several times, and record how often your brand is named, where it appears, and which sources are cited. Then you repeat the same test on a schedule and compare it with your competitors. Traffic data from analytics adds a second, partial view.

The reason for all the repetition is simple: AI answers are not stable. A method that would work for classic search rankings gives misleading results here.

Why can’t you check AI visibility with one search?

Because the same question rarely gets the same answer twice. In a study by SparkToro and Gumshoe, 600 volunteers ran 12 prompts through ChatGPT, Claude and Google’s AI 2,961 times. ChatGPT and Google’s AI had less than a 1 in 100 chance of returning the same list of brands in any two runs, and the same list in the same order came up roughly once in 1,000. The authors recommend measuring “visibility % across dozens to hundreds of prompts run multiple times.” One co-author works at an AI tracking startup, which is worth knowing, but the finding matches what anyone sees when they test by hand.

Google’s AI answers move too. Ahrefs tracked AI Overviews across more than 43,000 searches and found a 70% chance an Overview changed between observations, with an average persistence of about two days, even though the core meaning stayed similar. A separate Ahrefs comparison found that AI Mode and AI Overviews shared only 13.7% of cited URLs for the same query.

So a screenshot of one good answer is not evidence of success, and one bad answer is not evidence of failure. You need a sample.

How do you build a prompt set from real buyer questions?

A prompt set is the fixed list of questions you test every time. Build it from how buyers actually talk, not from how your marketing team describes the product. Sources include sales call notes, support tickets, site search logs, the “People also ask” box in Google, and Reddit threads in your category.

Cover the main question shapes:

• Category discovery: “best CRM for a small law firm”, “top-rated running shoes for flat feet”.
• Use case: “how do I automate invoice reminders for freelancers”.
• Comparison: “[competitor] vs [competitor]”, “alternatives to [competitor]”.
• Local: “family dentist open Saturdays in Leeds”.
• Brand check: “what is [your brand]”, “is [your brand] legit”.

Aim for 20 to 50 prompts to start, weighted toward the questions closest to a purchase. BrightEdge found that commercial wording such as “deals” or “where to buy” produced 4 to 8 times more brand mentions in ChatGPT than informational questions, and that 44% of its prompts had no brand mentions at all. If your set is mostly “what is” questions, you may be measuring answers that never name any brand.

Freeze the wording once you start. Changing a prompt changes the answer, so any edit starts a new baseline. Keep the location, language and logged-in state consistent too, and note which model or mode you used.

Which metrics should you track?

Four numbers cover most of what matters. Track them per engine, because each one behaves differently.

Mention rate is the headline. If you run 40 prompts three times each in ChatGPT, that is 120 runs; if your brand appears in 18 of them, your ChatGPT mention rate is 15%. Calculate it separately for each engine and for each group of prompts, because a brand can be strong on comparison questions and absent on discovery ones.

Position matters once you are mentioned at all. Being tenth in a list of ten is very different from being named first with a reason. Because the order shuffles between runs, record the average position across runs where you appear, not the best one.

A citation is a linked source shown with the answer. Mentions and citations are not the same thing. BrightEdge reported that ChatGPT names brands about 3.2 times as often as it cites them, so a brand can be visible with few or no links. Note which domains get cited for your prompts: they tell you where the engines are reading.

Share of voice adds context. Count every brand named across all runs, then divide your count by the total. If competitors take 70% of mentions and you take 5%, you know the size of the gap.

What does a worked example look like?

Suppose a project management app for architecture firms (an illustration, not a real result) builds a set of 40 prompts and runs each one three times in ChatGPT, Gemini and Perplexity: 360 runs in all. It is named in 30 of the 120 ChatGPT runs, 12 of the Gemini runs and 6 of the Perplexity runs. Its mention rates are 25%, 10% and 5%.

Looking closer, nearly all the ChatGPT mentions come from comparison prompts that name a competitor, and almost none from discovery prompts such as “best software for small architecture practices”. Perplexity cites three trade blogs for those discovery prompts, and none of them mention the app. That tells the team where to focus: get covered in context on the kinds of sources Perplexity is already reading, then re-measure in a month. Our guide on how AI chooses which brands to mention explains why the engines differ.

How do you run the checks across engines?

Run the full set in each engine your buyers use. ChatGPT, Gemini, Perplexity and Google’s AI features are the usual minimum; add Claude and Copilot if your audience uses them. For Google, check both AI Overviews and AI Mode, given how differently they cite.

A simple routine works:

1. Use a clean session: logged out where possible, or a fresh account with memory and personalization switched off.
2. Set the location to the market you sell in, because local and regional answers differ.
3. Run each prompt three times or more and record the results in a sheet: brand named yes or no, position, cited domains, any wrong facts.
4. Log the date and the model or mode shown.

Expect cross-engine patterns rather than identical results. Ahrefs found that brand visibility on one platform correlates with visibility on the others (0.749 between AI Overviews and ChatGPT, for example), so a brand that is strong on one engine is usually present on the rest, though not always in the same position.

Also record what the engines say. A mention with an outdated price model, a wrong founding year or a confusion with another company is a different problem from no mention. Our guide on when AI gets your brand wrong covers how to fix those.

How do you track AI referral traffic in analytics?

AI referral traffic is the visits that come to your site from links in AI answers. It is a useful second signal, though it only captures people who clicked.

• ChatGPT: OpenAI says ChatGPT adds utm_source=chatgpt.com to referral links, so those visits appear as their own source in Google Analytics and similar tools.
• Other assistants: Perplexity, Gemini, Claude and Copilot visits usually appear as referrals from their own domains, though some arrive without a referrer and show as direct traffic. Create a channel group or segment that collects them.
• Google AI features: Google says AI Overview and AI Mode clicks are included in overall Search traffic in Search Console, with no separate filter.

The volume is still small for most sites, but it tends to be valuable. Similarweb found that ChatGPT sent more than 80% of AI referrals to the top 1,000 domains it studied, and Adobe reported that AI-referred visitors to retail sites converted 31% better than other traffic in late 2025. Many people read an AI answer and never click, which is why the prompt checks remain the main measure. Our guide on whether AI visibility brings real traffic and sales goes deeper.

Which tools can do this for you?

You can start with a spreadsheet and the free AI visibility self-check, which builds a buyer prompt list for your brand and gives you a scoring grid that calculates a mention rate. That is enough to get a first baseline.

Once you track more than a few dozen prompts across several engines, paid tools save time. Several SEO platforms, including Ahrefs, Semrush and BrightEdge, now offer AI visibility tracking features, and specialist trackers such as Profound and Gumshoe focus on it. When comparing tools, ask:

• How many times is each prompt run, and is the result reported as a rate?
• Which engines, models and modes are covered, and in which countries and languages?
• Can you add your own prompts and competitors?
• Does it show cited sources, not just mentions?

If a tool reports a single “AI rank” for a prompt, ask how it accounts for answers changing between runs.

How often should you re-measure, and what counts as progress?

Monthly works for most brands, with an extra check a few weeks after any major push such as a launch or a round of new mentions. Keep the prompt set, number of runs, engines and settings the same so the numbers are comparable.

Progress looks like a rising mention rate across several checks, a better average position, more cited domains that mention you, and a larger share of voice against the same competitors. Expect some months to dip, since sources shift: Semrush tracked Reddit’s presence in ChatGPT responses falling from about 60% to about 10% within weeks in 2025. Read trends, not single checks. Results are measured, not promised, and anyone promising a fixed position by a fixed date is promising something no one controls.

For context on what good looks like, in SparkToro’s follow-up headphone test the leading brands appeared in about 55 to 77% of answers. Even category leaders are not named every time.

What to do next

• Write 20 to 50 buyer prompts and freeze the wording.
• Run each one at least three times in each engine and record mention rate, position, citations and share of voice.
• Get a free AI visibility check if you want us to run the baseline for you, or read how to get your brand mentioned by AI to plan what comes after it.
• If your mention rate is already high, the next metric to watch is position, which is what The Blue Ocean GPT works on in ChatGPT answers.