What is AI visibility? The plain definition, how it is measured, and why every score is an estimate

The one-sentence definition, and the words in it

Three parts carry the weight. How often is a rate: out of the answers you sampled, the share that included you. How prominently is position: the first name in a shortlist, or an afterthought in paragraph four. Name, cite or recommend are three different outcomes that get lumped together, so the next section pulls them apart.

AI visibility is an outcome, not a tactic. The tactics have their own names: answer engine optimization (AEO) is the page-level work of earning a citation inside an answer, generative engine optimization (GEO) is the brand-level work of being retrieved and represented across engines, and classic SEO is the link list underneath. AI visibility is the scoreboard those tactics are trying to move.

Mentioned, cited, recommended: the three things people mean

You can be mentioned without being cited (the engine knows you from a review site and links nothing), cited without being mentioned (your comparison page is the source for a competitor's description), and recommended by one engine while another has never heard of you. A single number hides all of that, which is why ClarAI keeps the verbatim answer beside every score.

How it is measured

Every serious measurement, ours included, is built from the same six parts. When a tool hides one of them, ask why.

1. Prompts. The buyer questions you ask the engines: awareness (what is X), consideration (best tools for Y), decision (X vs Y, is X worth it). The prompt set is the single biggest determinant of the score, so it should be visible and editable, not a black box.
2. Engines. Which assistants are asked. ClarAI reads four: ChatGPT, Gemini and Claude through the vendors' official APIs, with Claude grounded by web search, and Google AI Overviews, read through SerpAPI. It does not monitor Perplexity or Google AI Mode, and it never scrapes a logged-in ChatGPT session.
3. Runs. The same prompt asked more than once, because the same engine gives different answers to the same question minutes apart. One run is an anecdote. Several runs give you a rate, a sample size, and a band around it.
4. Presence. Whether the answer names you. Aggregated across prompts and runs, that is the mention rate.
5. Position. Where you appear when you do: first name in the list, or last. Only meaningful in the answers that mention you.
6. Citation. Which URLs the answer links to, and whether any is on your domain. Not every engine returns citations through its API; where one does not, the honest cell is "not returned", never zero.

What an AI visibility score is, and why every score is an estimate

No engine publishes an AI visibility score. OpenAI, Anthropic, Google and Perplexity do not expose how often a brand appears across their users' conversations, and Google's own guidance for generative AI features says that "No third-party tool has access to our internal ranking or AI systems" (Google Search Central, last updated 10 July 2026). Every score on the market, including ours, is therefore an outside estimate built from a sample the tool chose.

Because there is no official number, tools disagree, and the disagreement lives mostly in the prompt set, the number of runs and the formula. So here is ours, in full.

A score with no sample size beside it is not a measurement; it is a mood. The IAB's guideline "Measuring Visibility in the AI Era" (3 August 2026) treats fewer than 50 queries per measurement programme as exploratory rather than directional, asks for multiple responses per query and coverage of at least two intent types, and wants results reported per platform. Use those as your questions for any vendor, including us. We wrote up the failure modes in Your AI visibility score is lying.

AI visibility versus SEO, AEO and GEO

The terms overlap because the field is young and every vendor draws the lines where its product sits. We keep it simple: AEO and GEO are the work, AI visibility is the measurement, and no consensus definition exists for any of the three.

Brand monitoring in AI answers

"ChatGPT brand monitoring" and "ChatGPT brand tracking" are the searches people run when they want the scoreboard on a schedule. In practice it means a fixed prompt set asked on a fixed cadence, every answer stored verbatim, competitor mention rates on the same prompts, and an alert when an engine that used to name you stops. That is what ClarAI's answer monitor does, run over run.

What it cannot do, and what no tool can do honestly: see private conversations, count how many real people asked, or promise that a logged-in user with memory switched on sees the same answer as an API call. If a tool implies it can, ask which API returns that.

What Google's Search Console report shows, and does not

Since 31 August 2026 every Search Console property has a Generative AI performance report. It shows impressions, meaning how many times links to your site were shown, inside AI Overviews and AI Mode, split by page, country, device and date (Search Console Help, answer 16984139). It does not show clicks, queries or position; it is not available through the Search Analytics API, whose search types cover web, image, video, news, Discover and Google News and nothing generative; and a logging error reduced its impressions for 13 to 17 August 2026 (Search Console data anomalies page). It covers Google only.

In the same rollout Google gave property owners a "Search generative AI" control under Settings, Include or Exclude. Exclude removes a site from AI Overviews, AI Mode and generative AI features in Discover, is not a ranking signal, does not affect AI training, and leaves no trace anyone outside the property can see (Search Console Help, answer 16908024). Our reading of the report is in what the GSC generative AI report shows and hides.

How to check yours, free

Two free things, and they measure different things. The free grader reads your domain and scores AI readiness: whether the retrieval crawlers can reach you, whether snippet controls are hiding your text, whether the page structure gives an engine something to quote. That is readiness, not visibility: a grade-A site can still go unmentioned. It also shows one real answer for one prompt on one engine as a teaser; that answer is not repeatable and it is not a score.

Visibility itself, across engines and runs with the sample size attached, is the product. The free AI visibility checker page explains what free checkers can and cannot see, and the open benchmark publishes aggregate readiness figures across the domains the grader has checked, with the sample size on every figure and a 200-domain floor before anything is published at all.

Glossary: twelve terms, one line each

AI visibility

How often and how prominently AI answer engines name, cite or recommend a brand on a set of buyer prompts.

Mention rate

The share of sampled answers in which the brand name appears; the base of most scores.

Citation

A URL an answer links to as a source. It can be your page or a third-party page about you.

Own-domain citation

A citation whose URL is on your domain, as opposed to a review site or a listicle.

Position

Where the brand appears in an answer that mentions it, first or last; undefined when it is absent.

Prompt set

The buyer questions asked of the engines; the single biggest determinant of any score.

Run

One asking of one prompt on one engine. Several runs turn an anecdote into a rate.

Sample size (n)

Prompts multiplied by runs behind a number. Printed beside every ClarAI figure.

Engine

An answer engine: ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews, Google AI Mode.

AI Visibility Score

ClarAI's 0 to 100 blend of mention rate and position, formula stated on this page.

Share of voice

Your mention rate against the tracked competitors' on the same prompts; not an industry benchmark.

Query fan-out

Google's term for splitting one question into sub-queries that are searched separately (AI Mode).