You measure AI visibility by running a fixed panel of buyer prompts across the major AI engines on a schedule and scoring four things: mention rate, position, sentiment, and share of voice. There is no analytics account to pull from, because AI providers do not publish prompt logs; sampling is not a workaround, it is the measurement method.
The four core metrics
Each metric answers a different question, and you need all four to know what is actually happening.
- Mention rate: the share of sampled responses that name your brand at all. The base visibility number.
- Position: where you land when multiple brands are named. First-mentioned and sixth-mentioned are different commercial outcomes.
- Sentiment: how the answer frames you when you appear, from recommended leader to caveat-laden afterthought.
- Share of voice: your mentions divided by all brand mentions in the sample, which tells you whether you are gaining on competitors or just riding category growth.
Slice by prompt type and provider, or the average will lie
An overall mention rate hides the story. Brands routinely score well on recommendation prompts and score zero on alternatives prompts, or dominate Perplexity while being absent from ChatGPT, and a blended average reports both situations as "fine." Structure the panel across the four buyer prompt families (recommendation, comparison, alternatives, pricing and fit) and record results per provider. The differences between cells are where the actionable findings live, because each provider retrieves differently: Perplexity cites live sources every time, ChatGPT answers from weights plus optional search, Gemini grounds on Google's index.
Why a single number lies and a scorecard works
Composite scores have a real job: trend tracking and executive reporting. Our own scan produces a GEO score for exactly that purpose. But a composite cannot diagnose, because a 58 could mean weak coverage everywhere or total dominance in one provider and absence in the rest. The working format is a scorecard: the four metrics, broken out by prompt type and provider, with the composite on top as the headline. Read the scorecard to decide what to fix; report the composite to show whether it worked.
The sampling discipline that makes it trustworthy
Three rules keep the numbers honest. Freeze the prompt panel, because edited prompts break your trend line. Run each prompt multiple times per measurement, because AI answers vary run to run and single responses are anecdotes. Keep cadence and providers constant, weekly during active work, so week-over-week deltas mean something. This is the discipline that no Search Console for AI prompts forces on everyone. If you would rather not build the machinery yourself, a free scan runs the panel across five providers and returns the scorecard.
Get your baseline scorecard in minutes
Run a free scan to measure your mention rate, position, sentiment, and share of voice across ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews, before and after you start the work.
Frequently asked questions
What KPIs matter for AI visibility?
Four leading KPIs: mention rate, average position, sentiment, and share of voice, each split by prompt type and AI provider. Pair them with lagging business KPIs, chiefly AI referral traffic and conversions from it, so visibility gains can be tied to commercial outcomes.
How many prompts do I need to measure AI visibility?
Enough to cover the four buyer prompt families with several phrasings each; in practice panels of 20 to 50 prompts work for most brands. Depth of repetition matters as much as breadth: running each prompt several times per measurement smooths the natural variance in AI responses.
Can Google Analytics measure my AI visibility?
Only a thin slice of it. GA4 shows referral sessions from sources like chatgpt.com and perplexity.ai, which captures clicks but misses every answer where your brand was mentioned, or omitted, without a link being clicked. Prompt sampling measures the visibility itself; analytics measures one downstream effect.
Related answers