Ask a marketing team whether they are visible in AI answers and you will usually get a screenshot: someone typed the company's category into ChatGPT last Tuesday, the brand appeared, and the matter was considered settled. The trouble is that the same question asked on Wednesday, or by a colleague two desks away, can return a different set of names entirely, which is why the screenshot proves almost nothing. AI visibility is the rate at which answer engines such as ChatGPT, Google's AI Mode, Gemini and Perplexity name your brand across the questions your buyers actually ask, measured over many runs rather than one, and the argument of this piece is that it behaves like a probability and not a position, so the only honest way to state it is as a sampled rate with the method shown.
That distinction is not pedantry, because a founder who treats AI visibility as a rank will buy the wrong tool, read the wrong number and relax at exactly the wrong moment. If a buyer asks ChatGPT to shortlist vendors in your category and your name is absent, you have lost a recommendation you will never see happen, with no rejection email and no lost-deal report to record that it occurred, so the loss is real precisely because it is invisible. Knowing your true rate, and watching it move, is the difference between managing that channel and hoping it is fine.
AI visibility is a rate, not a rank
Why AI visibility is a rate, not a rank
~70%
of repeat questions return a different answer
The answer one buyer sees need not be the one the next sees. SparkToro.
<1 in 100
chance two identical questions name the same brands
So one check is a single draw, not a position you hold. SparkToro.
62%
of AI citations never name the brand
You can feed the answer and still not survive into it. Semrush.
Start with the numbers that force the point. The same buyer question returns a materially different answer roughly 70% of the time, and two identical questions asked back to back match the same list of named brands less than one time in a hundred (SparkToro). Read plainly, that means the reply you screenshotted is one draw from a shuffled deck, closer to a coin landing heads than to a rank you have earned, so a metric built on a single look is measuring luck. Worse, 62% of the citations an engine leans on never name the brand behind the page (Semrush), so you can supply the very sentence an answer is built from and still not appear in it, which is why counting only the answers that name you undercounts the work your content is already doing.
The habit worth breaking is the reflex to screenshot, because the mental model behind it belongs to a different kind of search. A position in classic Google results was stable enough that one look told you something durable, whereas an AI answer is regenerated for every asker, so the same reflex now captures noise and files it as fact. The table below sets the old instinct against what the metric actually is.
| If you think in Google rankings | What AI visibility actually is |
|---|---|
| A fixed position you occupy | A rate that shifts from one run to the next |
| One check confirms where you stand | One check is a single draw; you need many |
| Won mainly on your own pages | Won largely in the third-party sources a model reads |
| A figure you can screenshot and file | A trend line, only as good as its sampling method |
What the metric is actually made of
Because of that variance, a usable definition has to separate what is being counted, since AI visibility is not one number but a small family of them, and confusing them is how dashboards end up flattering or frightening you for no reason. Four things are worth counting on their own before anything blends them together.
| Component | What it counts | How to read it |
|---|---|---|
| Mention rate | The share of sampled answers that name your brand at all | Your headline visibility, meaningful only across many runs |
| Citation rate | The share that link to your own pages as a source | Whether the engine trusts you enough to point at you |
| Sentiment | How you are described in the answers that do name you | Often written by third parties, so track it rather than assume it |
| Share of voice | Your mentions set against the rivals named alongside you | Tells you whether absence is your problem or the category's |
The four move independently, which is the whole reason to keep them apart: a brand can hold a high mention rate on a thin citation rate because the engine names it from other people's pages, or carry warm sentiment while quietly losing share of voice as a competitor earns more of the same conversations. A tool that collapses all four into a single visibility score is convenient, and hides exactly the movement you most need to see. That is why our companion piece on how to actually put a number on AI visibility argues for reading the components before the composite, and the mechanics of a single blended figure, with benchmarks, are unpicked in how to calculate an AI visibility score.
Want to get recommended by AI?
Check your AI search visibility, then let the Honeyb agent write, fix, and earn what gets you recommended. Free to start.
A number without a method is not a measurement
The reason the method matters as much as the number arrived, aptly, as a lesson in how easily one figure misleads. Writing in Search Engine Journal on 10 September 2026, Duane Forrester traced a single metric, Cloudflare's crawl-to-refer ratio, which compares how many pages an AI company's crawler takes against how many visitors it sends back, and found it quoted for Anthropic at 70,900 to one, then 38,000, 23,951, 11,122, 10,300, 4,580 and 2,237 to one, all within roughly thirteen months and all attributed to the same source. As he noted, "the June 2025 figure for Anthropic is 73,000 to one, with OpenAI at 1,700 to one", while Cloudflare separately "reported Google's ratio moving 19.4% week over week". The figures were not wrong so much as unusable, because each came from a different window and denominator that the headline quietly dropped.
The moral transfers directly. A vendor who tells you that you appear in "38% of answers" has told you almost nothing until you know which questions were asked, on which engines, how many times and over what period, because change any one of those and the number moves by more than the amount you were about to act on. The figure is worth exactly as much as the sampling method printed beside it, so when a dashboard shows a visibility percentage with no note on prompts, engines, run count or dates, treat it as a starting question rather than an answer.
You cannot fix it on your own website
If the number is a rate set across many answers, the next question is where those answers come from, and the uncomfortable finding is that they come mostly from other people's pages. Ahrefs, studying what correlates with appearing in AI answers, found the strongest associations were with third-party mentions and video rather than with on-page work, which means the lever most teams reach for first, another pass over their own site, is the one least likely to move the number. Reinforcing the point, Reddit alone accounts for 40.1% of all the citations AI answers draw on, the single most-cited source across the web (Semrush), so a large share of what an engine effectively knows about your category is being written in communities you do not control and may not yet read. Landed plainly, your AI visibility is largely authored off your own domain, so raising it is a matter of earning mentions where the models look rather than polishing pages they increasingly skip, a case we make at length in why brand presence in generative AI is won off your own website.
How to measure your own AI visibility
Measuring it is less like checking a rank and more like running a small poll, and the method is not complicated. Write down the ten to twenty questions a buyer would genuinely ask in your category, in their words rather than your keywords, because "best CRM for a two-person law firm" is what gets typed and "CRM software" is not. Run each question several times on each engine that matters to you, since one pass tells you almost nothing at the variance above, and record not merely whether you were named but which sources the answer leaned on, because that is where any fix will begin. Then repeat on a schedule and read the trend rather than the spot value, treating a single reading the way you would treat one day's share price. Engines differ in how much they will show you, in that Perplexity and Claude tend to expose their citations while ChatGPT and Gemini often hide theirs, so wherever the sources are visible, capture them. The full procedure, with the sampling counts spelled out, is laid out in our step-by-step AI visibility audit, and if you want to understand how the engines choose what to cite in the first place, start with what an AI answer engine is and how it cites.
The short version is the definition worth keeping: AI visibility is the measured rate at which answer engines name you across the questions your buyers ask, read as a trend and reported with its method attached, and anyone who offers you a single tidy position number is selling the one thing this metric cannot honestly be. The fastest way to see your own rate, across the main engines and with the sources shown where the engine reveals them, is to run the free AI visibility checker against your own category and treat the first result as your baseline rather than your verdict.
