What Is AI Visibility and How Is It Measured?

AI visibility is how often, and how favorably, AI answer tools such as ChatGPT, Gemini, Perplexity and Google's AI Overviews name or link to your business when someone asks a question you would want to be part of. It is an outcome, not a setting. You cannot switch it on in your website. You can only look at the answers people get and see whether you are in them.

The catch is that "AI visibility" is not one number. Every tool that reports it uses the same handful of words, mentions, citations, visibility, share of voice, sentiment, and quietly defines each one differently. Two dashboards can both show "visibility: 40%" and be measuring different things. That is why a score that looks clear on a screen turns out hard to act on.

This post explains what is actually being counted, using each tool's own documentation, so you can tell which numbers deserve your attention and which are noise. If you want the difference between visibility and the work you do on your own site, that is covered in readiness vs. visibility. This post stays on the visibility side.

What AI Visibility Actually Measures

Behind every AI visibility number is the same routine. A tool (or you, by hand) asks an AI a list of questions, saves what comes back, and does sums on the saved answers. The tools call those questions "prompts." Here we will just call them questions.

That routine only lets you record a few things from an answer:

• Whether your name appears in the text. This is a mention.
• Whether your website is listed as a source. This is a citation.
• Where you come in the answer. First brand named, third, and so on.
• How you are described. Positive, neutral or negative in tone.
• Who else appears. Which competitors were named alongside you or instead of you.

Every AI visibility metric is one of those five things, or a ratio built from them. There is a sixth thing that people lump in with visibility but that cannot be recorded from an answer at all: whether anyone actually clicked through to your site. We come back to that near the end.

Because the tools are asking questions and writing down answers, the numbers are samples, not readings taken from inside the AI. We covered why that makes two tools disagree in our tools comparison. The rest of this post is about the next layer down: even when two tools use the same word, they often count different things.

Mentions and Citations Are Two Different Events

These two get blurred together constantly, and they should not be.

A mention means your brand name appears in the text of the answer. A citation means one of your pages is listed as a source for the answer. You can have either without the other.

Ahrefs, which runs a large AI tracking product called Brand Radar, draws a third line that most guides skip. In its help documentation, it separates pages the AI cited from pages it labels "Found but not cited," meaning pages the AI looked at in the background while putting the answer together and then left out of the sources list. So a page can shape an answer without ever appearing in it.

The counting rules matter just as much as the definitions, and the vendors agree on them more than you might expect. Ahrefs counts a mention once per answer, however many times the name appears in it. Its own example is an answer that names Salesforce three times, which still counts as one mention. Citations are counted once per website per answer, so an answer that links to a dozen pages on the same site gives that site one citation. Otterly, another tracking tool, uses the same one-per-answer rule in its KPI definitions: if one answer names both "Nike" and "Nike Air Jordan," the count is still one.

What this means in practice: a mention count is really a count of answers that named you. It is not a running total of how often your name was written, and it does not behave like page views. Fifty mentions means fifty answers, out of however many answers were checked. Without the second number, the first one tells you very little.

"Visibility": Same Word, Different Math

The headline number on most dashboards is called some form of visibility. Here is what it actually is in four of the best-known tools, taken from their own documentation:

Tool	What it calls it	What the number actually is
Peec AI	Visibility Score	Answers that mention your brand, divided by all answers, times 100.
Otterly	Brand Coverage	Questions that mention your brand, divided by all questions tracked. Google AI Overview questions that never triggered an overview are left out.
Semrush	AI Visibility	A score from 0 to 100 for how often your brand appears in AI answers compared with your competitors.
Semrush (Prompt Tracking)	Visibility	Where your site's link sits among the cited sources for tracked ChatGPT and Google AI Mode questions. 0% means never in the top citations, 100% means first citation on every tracked question.
Ahrefs	No visibility percentage. It reports "Impressions."	The Google search volume of every tracked question where your brand shows up, added together.

The sources for each row are Peec's visibility page, Otterly's KPI page linked above, Semrush's metrics page (which lists both of its "visibility" numbers) and Ahrefs' metrics page linked above.

Read down the third column and you can see three separate questions hiding behind one word:

• "How often am I named?" Peec and Otterly answer this one, with almost identical math under different names.
• "How do I compare with my competitors?" The Semrush overview score answers this one, so the result depends on who your brand is compared against.
• "How prominent is my link?" Semrush's question-level number answers this one, and it only looks at citations, not mentions.

Ahrefs is a fourth thing again. Its impressions are weighted by how many people search for each topic on Google, which makes a mention on a popular question count for more than a mention on a rare one. That is a reasonable design, but it is an estimate of potential exposure, not a count of something that happened. Ahrefs says so itself in its methodology write-up: "Metrics are directional indicators, not exact traffic counts."

So when someone tells you their AI visibility is 40%, the honest reply is a question: 40% of what?

Share of Voice Depends on Who You Told the Tool to Track

Share of voice sounds like market share. It is not. It is your slice of a pie whose size you chose.

Peec and Otterly both use the same formula: your mentions, divided by the total mentions of every brand being tracked, times 100. Ahrefs uses a share of impressions instead of a share of mentions, so popular questions count for more. Semrush's own measurement guide says that in its enterprise product the number reflects both how many mentions a brand gets and where it appears in the answer, and that for ChatGPT the topic's search volume is factored in too.

The weak point is the same in every version: the total. Ahrefs warns about it directly in its documentation, calling share of voice a comparative metric that "can change a lot depending on how you've set up" the brands being tracked.

Here is what that looks like with made-up numbers. Say you check 100 answers. Your brand is named in 20 of them. Competitor A is named 50 times and Competitor B 30 times, so the total across all three brands is 100 mentions.

• Your coverage is 20 out of 100 answers, or 20%.
• Your share of voice is 20 out of 100 mentions, or 20%.

Now you add a third competitor to the tracking list. Competitor C is named 50 times in those same 100 answers. Nothing changed on the web. The AI said exactly the same things. But the total is now 150 mentions, and your share of voice falls from 20% to about 13%. Your coverage is still 20%.

That is the whole lesson. Share of voice tells you how you compare with the specific brands you chose to watch, and it moves when you change the list, not just when the answers change. It is useful inside a fixed set of competitors over time. It says nothing about your standing in your whole market.

Position and Sentiment Are the Noisiest Numbers

Position and sentiment are the two metrics that look most like a report card, and the two you should trust least.

Position

Position is meant to show how high up you appear. But "high up" means different things in different tools. Peec measures your average rank among every brand named in the answers where you appear, including brands you never told it to track. Otterly averages your position across answers in a similar way. Semrush's question-level position is about your link's place in the citation list, not your name's place in the text. A "position 2" can mean the second brand named or the second link shown.

More importantly, the order of an AI answer barely holds still. In January 2026, SparkToro and Gumshoe published a study in which 600 volunteers put 12 different questions to ChatGPT, Claude and Google's AI a combined 2,961 times. The result: there is less than a 1 in 100 chance that two responses give the same list of brands, and closer to 1 in 1,000 that two responses give the same list in the same order.

The same study found something more useful. One hospital appeared in 69 of 71 ChatGPT answers, a 97% appearance rate, yet it was the first name in only 25 of them. Its position bounced around, but its presence was steady. The author concluded that "visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric," while any ranking position is not.

That is worth holding onto. Of everything on a dashboard, the share of answers that name you, measured over many questions and many repeats, is the number the research supports most. Position is the one it argues against.

Sentiment

Sentiment tries to say how favorably the AI describes you, and there is no shared scale for it. Peec scores tone from 0 to 100, and its own documentation says most scores fall between 65 and 85, so a 72 there is ordinary, not a warning. Otterly uses a net score: positive mentions minus negative mentions, divided by all mentions, times 100, giving a range of -100 to +100. A +72 on that scale would be a very strongly positive result. Semrush reports it as the balance of positive and general sentiment. A "72" from one tool and a "72" from another are not the same result.

One limit worth knowing: the SparkToro study looked at lists and their order. Its authors said outright that they did not collect data on how the AI described each brand, so nothing in that research either supports or undermines sentiment scores. Treat them as a rough read on tone, checked by reading a few of the actual answers yourself.

What No Visibility Number Tells You

Whether anyone came to your site. Ahrefs is direct about this in its methodology page: the data "is not a substitute for audience measurement or traffic analytics," and it should be thought of as "a media-visibility audit, showing what appears in AI and search – not who saw it." Real visits come from your analytics. We walk through where AI visits show up in our guide to AI traffic tracking in GA4. Google's own number is a third, separate thing: Search Console's generative AI report counts one impression per AI answer, which is a different counting rule again from anything above.

Whether what was said about you is true. None of the definitions above measure accuracy. A mention counts the same whether the answer gives your correct address or a wrong one. Sentiment measures tone, not truth, so a glowing answer with the wrong opening hours can score well. This is the gap that a visibility number cannot see and a readiness check can.

Why the number moved. A change in your score might come from something you did, something a competitor did, a change to the list of questions, or plain randomness in the answers. The number does not tell you which.

How to Read Any AI Visibility Number

When a score is put in front of you, five questions cut through most of the confusion:

1. What is being counted, out of what? Answers, questions, mentions, or search volume? If the tool does not say, the number is not usable.
2. How many questions, and how many repeats? The SparkToro authors point to dozens to hundreds of questions run multiple times. A number built from ten questions asked once is a story, not a measurement.
3. Where did the questions come from? Ahrefs builds its list from Google's People Also Ask and its own keyword database. Other tools use the ones you type in. Either way, ask whether these are questions your real customers put to an AI.
4. Who is in the list of competitors? For share of voice or any competitor-relative score, the answer decides the number, as the example above showed.
5. Compared with what? Compare a tool only against its own past results. Never set one tool's score beside another's.

Common Mistakes

Comparing scores across tools. A Peec visibility score and a Semrush AI Visibility score are different calculations that share a word. Pick one tool, note where you start, and measure change against that starting point.

Treating share of voice as market share. It shifts when the competitor list shifts.

Chasing position. The order of an answer is the least stable thing about it, according to the SparkToro study above.

Counting a mention as a win without reading it. A mention with a wrong fact in it counts the same as a right one.

Reading impressions as visits. Estimated exposure is not traffic, and the tools that produce it say so.

Checking once. One run of one question is a single draw from answers that change every time. A single check tells you almost nothing.

Where to Start

You can do the sums yourself before paying for anything. Pick ten questions a real customer would ask an AI. Put each one to a few different AI tools, several times. For every answer, write down two yes-or-no marks: was my business named, and was my site listed as a source. The share of "yes" marks in each column is your coverage and your citation rate, the same two numbers a dashboard would compute. The step-by-step version is in our guide to improving brand visibility in AI search.

AI Schema Gen does not track these numbers. It checks the other half: whether your site gives AI accurate, well-organized material to work with in the first place, so that when you do show up, the answer is correct. If you want to see where your site stands on that side, the free check below takes about a minute.