You measure AI visibility with four numbers: your AI share of voice against competitors, your citation presence rate on questions you should own, the accuracy and sentiment of how AI engines describe you, and the demand AI refers to you. You get them by running a fixed set of real buyer questions through ChatGPT, Perplexity, Gemini and Google AI Overviews on a regular cadence — monthly is usually enough — and scoring the answers against a consistent rubric.
The discipline that makes the numbers meaningful is consistency: the same questions, the same engines, the same scoring. Change the prompt set and you have measured noise, not progress.
Key takeaways
AI visibility is measured with four metrics: share of voice, citation presence rate, description accuracy and sentiment, and AI-referred demand.
Use a fixed set of real buyer questions across ChatGPT, Perplexity, Gemini and Google AI Overviews, re-run on a monthly cadence.
Absolute numbers matter less than trend and benchmark: your share versus competitors, this month versus last.
AI-referred demand is the business outcome — track it as a segment in your analytics, not as a vanity metric.
The same evidence-first logic applies here as everywhere else: decisions, not dashboards.
The four metrics
AI share of voice
Of the answers AI engines give to your category questions, how many mention you versus competitors.
How to track it: Run a fixed set of buyer questions across ChatGPT, Perplexity, Gemini and Google AI Overviews; count mentions as a share of all brand mentions.
Citation presence rate
How often your brand is named — with a source — in answers where you should be the answer.
How to track it: Track the questions you are qualified to own and measure the percentage where at least one AI engine names you.
Description accuracy and sentiment
Whether what AI engines say about you is true, current, and favorable.
How to track it: Compare each answer against your actual positioning; score for accuracy, completeness, and sentiment.
AI-referred demand
Traffic, referrals and conversions that arrive from AI engines.
How to track it: Segment analytics by AI referrers (chatgpt.com, perplexity.ai, gemini) and watch the trend, not the absolute number.
The method, step by step
Write down the 20–40 questions your buyers actually ask when they evaluate a category like yours — the questions a sales call hears, not the ones marketing wishes people asked.
Run every question through each AI engine, logged verbatim: which brands are mentioned, which sources are cited, and what is said about each.
Score each answer for mention (were you named?), accuracy (was it true?), completeness (did it capture your differentiators?) and sentiment.
Compute your share of voice against the competitive set, and store the raw answers so you can see drift over time.
Re-run monthly, on the same questions, and read the trend — a rising share of voice and a rising citation rate are the leading indicators; referred demand is the lagging one.
What "good" looks like
In the AI visibility audits I run for clients, most brands appear in only a small fraction of the answers they should own. That means the bar is still low, and the advantage compounds early: brands that fix their entity consistency, publish citable evidence, and earn third-party mentions now are the ones AI engines treat as defaults later. A realistic first goal is not dominance — it is being accurately described in the answers to the ten questions that matter most to your revenue.
For most teams, the fastest start is an AI Visibility Audit — the questions, the engines, the scores, and the gap to competitors — so that every improvement afterwards is visible rather than assumed.
The measurement mindset
One warning from practice: do not build a dashboard nobody acts on. Each metric earns its place by changing a decision — which pages get rewritten, which evidence gets published, which third-party sources get pursued. If a number cannot change a decision, stop tracking it. That is what responsible growth means applied to measurement: evidence that leads to decisions, not dashboards that lead to meetings.