Quick Answer: AI visibility is how often an answer engine such as ChatGPT, Gemini or Google AI Overviews names, cites, links or recommends your brand. It is measured per prompt on the live answer, not as a rank and not as traffic. The four components (mentioned, cited, linked, recommended) are separate readings, and each can move without the others.
The phrase gets used loosely to mean "we showed up in ChatGPT once", so it helps to pin it down before anyone measures it. This page is the definition only. How an AI visibility score is calculated and how to measure AI visibility step by step are separate pages.
What AI visibility means
AI visibility is the degree to which AI answer engines include a brand in the answers they give to a set of relevant questions. The engines in scope are ChatGPT, Gemini, Perplexity, Claude, Grok, Microsoft Copilot and Google's AI Overviews and AI Mode. The unit of observation is one answer to one prompt on one engine on one day.
The word "visibility" is borrowed from search, where it meant how often your pages appeared in results. In an AI answer there is no list, only a synthesised paragraph with sources attached, so visibility becomes a question about that paragraph.
| Component | The question it answers | Where it is read |
|---|---|---|
| Mentioned | Does the answer text contain the brand name? | The answer body |
| Cited | Is a page on the brand's domain among the answer's sources? | The Sources panel or source list |
| Linked | Is the brand's URL clickable inside or beneath the answer? | The answer body and the link panel |
| Recommended | Does the answer put the brand forward as a choice, not merely a fact? | The answer body, read for intent |
What counts as an AI answer surface
An AI answer surface is any interface that returns a generated answer with sources rather than a ranked list of pages. Google describes AI Overviews and AI Mode as features that surface links to support a generated response, and says AI Overviews appear only when judged additive to classic Search, per its AI features documentation. ChatGPT and Perplexity are answer surfaces on every query.
Why one prompt is the unit
A brand is never visible in general; it is visible for a specific question, on a specific engine, at a specific moment. The same engine can name you for "best shower filter for hard water" and omit you for "shower filter under 3,000 rupees" an hour later.
What AI visibility is not
AI visibility is not a rank, not a traffic figure and not a single score, and confusing it with any of the three leads to the wrong fix. Each of the three is a real thing worth knowing; none of them is the same reading.
Not a ranking
There is no position one in an AI answer, because the answer is a paragraph rather than an ordered list. A brand can be the first name on Monday and absent on Tuesday. Even on Google, Search Console assigns every link in an AI Overview the same single position.
Not traffic
A mention sends no session, and most AI visibility produces no click at all. Pew found that Google users who saw an AI summary clicked a link inside it on just 1% of visits, and clicked any result on 8% of visits against 15% when no summary appeared, in its 2025 browsing study. Visibility is the presence; traffic is a downstream effect that zero-click search has already weakened.
Not one score
A single blended score hides which of the four components moved. A tool that reports "AI visibility: 42" cannot tell you whether you are named without a link or linked without a recommendation, and those two problems have different fixes. Keep the four readings separate.
What are the four components of AI visibility
The four components of AI visibility are being mentioned, cited, linked and recommended, and they form a rough ladder from weakest to strongest. A brand usually climbs it in that order, because each rung depends on something the engine can find.
Mentioned
Mentioned means the brand name appears in the answer text, whether or not anything links to it. It is the weakest component and the first to move, because a model can name a brand from training data without fetching a page. A mention with no link says the engine knows you exist but has nothing of yours to quote.
Cited
Cited means a page on your domain is listed among the answer's sources. ChatGPT shows these in a Sources panel below the response, and OpenAI's help page warns that citations can be incomplete, outdated or incorrect, in its guide to searching with ChatGPT. A citation proves the engine used your page, not that it named you.
Linked
Linked means the brand's URL is clickable inside or directly beneath the answer, which is the component that can produce a session. It is the one that GA4 can see: since 13 May 2026, sessions arriving from ChatGPT, Gemini, Copilot, Grok and similar assistants are filed under an AI Assistant channel, per Google's channel definitions. A citation buried in a collapsed panel is cited but, in practice, rarely linked.
Recommended
Recommended means the answer puts the brand forward as a choice: "consider X", "X is a strong option for", "most people in your situation pick X". It is the strongest component and the hardest to earn, because it depends on comparative evidence rather than a single fact. Whether ChatGPT chooses to cite a brand at all sits under this rung.
Why AI visibility moves
AI visibility moves because every answer is generated fresh from a retrieval step that can return different pages each time. Four mechanisms account for most of the movement, and none of them is a bug.
Retrieval and query fan-out
Both AI Overviews and AI Mode may use a "query fan-out" technique, issuing multiple related searches across subtopics to build one response, according to Google's documentation. Which pages those sub-searches return decides which brands get named, so a small change in the sub-queries changes the answer.
Non-determinism
Asking the same engine the same question twice can produce two different answers, with different sources. This is why a single reading is not evidence and why measurement uses a fixed schedule across a prompt set. A brand present on 20 of 30 prompts today and 17 tomorrow has a noisy reading, not a loss.
Model updates and freshness
A new model version can change how an engine weighs sources, and an engine that prefers recent pages will drop a citation once a fresher rival page exists. Neither is announced per brand. Visibility earned once is not held by default, which is why answer engine optimization treats refresh as part of the job.
Personalisation and location
Signed-in accounts carry history that shapes the answer, and location changes which sources an engine treats as relevant. A reading from a marketer's own account is not the reading a buyer gets. The defensible reading is signed-out, in the buyer's country, and dated.
How AI visibility is measured and improved
AI visibility is measured by asking a fixed prompt set to each engine on a fixed schedule and recording the four components per answer. The quick manual version takes about twenty minutes and is described in checking whether AI mentions your brand; the full method is in the measurement page linked above.
Improving it follows the ladder. A brand that is mentioned but never cited needs a page an engine can fetch and quote; one that is cited but not linked needs the claim placed where the engine surfaces it; one that is linked but not recommended needs comparative evidence such as reviews.
The vocabulary for all of this, from GEO to LLMO, describes the same work from different angles.