What Is an AI Visibility Score — and How to Measure Yours

“We should be visible in AI search” has become one of the most common instructions handed to marketing teams – and one of the least defined. Ask what it means numerically and the room usually goes quiet. Everyone agrees it matters; almost nobody can say what “it” is.

Search rankings never had this problem. A ranking had a position: you were third for a keyword, or eleventh, and everyone could see it. AI answers feel different – an answer either mentions you or it doesn’t, the same question can produce different answers on different days, and there is no results page to point at. So AI visibility has acquired a reputation for being unmeasurable. It isn’t. It just needs a defined measure – which is what an AI Visibility Score is.

What is an AI Visibility Score?

Three parts of that definition do the work. Per-platform means each engine is counted separately – appearing in Google AI Overviews and appearing in ChatGPT are different achievements, earned in different ways, and a single blended number would hide which is which. A tracked set of buyer-intent prompts means the score is anchored to a fixed list of questions real buyers ask, so the number means the same thing every month. And compared period over period means the score exists to show direction of travel – is your presence in AI answers growing or shrinking – not to flatter a single snapshot.

That is the whole idea. Rankings gave you a position on a page; an AI Visibility Score gives you a count of appearances in answers. Neither is a measure of revenue. Both are measures of presence – and presence is what everyone asking for “visibility in AI search” actually wants made concrete.

What the score is built from

A credible AI Visibility Score has four components:

• A tracked prompt set. A defined list of buyer-intent questions – the things a prospective client would actually ask an AI assistant – not vanity queries about your brand name. If the prompts wouldn’t precede a purchase decision, appearing in the answers is decoration.
• Per-engine appearance counts. How many tracked responses on each platform – Google AI Overviews, AI Mode, ChatGPT, Gemini, Perplexity, Copilot – included the brand. Counted engine by engine, because the engines behave differently and reward different things.
• Page coverage. How many distinct pages of the site are doing the appearing. One page carrying every appearance is a concentration risk; broad coverage means the site as a whole has become citable.
• Direction of travel. Every figure reported against the prior period, so the score reads as a trend, not a trophy.

A note on methodology, because a measure is only as honest as its source. The scores we track are built on Ahrefs Brand Radar’s AI responses index: a set of tracked prompts, with appearances observed by the tool across the platforms it monitors. That makes the numbers consistent and comparable over time – but it also means they are tool-observed appearances within a tracked index, not a census of every AI answer generated anywhere. Anyone selling you “total AI market visibility” is selling something no tool can currently see.

What a real AI Visibility Score looks like

Abstractions are easy, so here is what the measure looks like on real clients, with the tracked figures as reported.

Hennessey Law, a US law firm client, appeared in 99 tracked Google AI Overviews responses – up 41 on the prior period – across 33 pages of the site. The same tracking recorded 127 tracked Google AI Mode answers (up 67) and 59 tracked ChatGPT responses, alongside tracked appearances in Gemini (25), Copilot (30) and Perplexity (24). Read as a score, that is a firm whose presence is growing on Google’s AI surfaces fastest, already material in ChatGPT, and established – at lower counts – everywhere else the index tracks.

Mrs Tilly’s, a confectionery brand, appeared in 75 tracked AI Overviews responses – up 48 on the prior period – across 20 pages, plus 17 tracked appearances in AI Mode. A different industry, a different scale, and a different platform mix – which is precisely why the score is reported per platform rather than as one blended number.

Notice what those figures let you say that “we should be visible in AI search” never could: which engines, how often, from how many pages, and whether it is improving. That is the difference between a sentiment and a metric.

What moves the score

An AI Visibility Score is an output. The inputs – the things that make AI engines more likely to cite you – are the substance of generative engine optimisation, and they cluster into five areas:

• Entity clarity. The engines need to know, without ambiguity, who you are, what you do and where you operate – consistently, across your site and the wider web.
• Structured data. Schema markup that tells machines what each page is, rather than leaving them to infer it.
• Citable content structure. Clear questions answered directly, definitions stated crisply, claims an engine can lift and attribute. Content written to be quoted, not merely read.
• Authority signals. The credentials, coverage and links that give an engine confidence you are a source worth citing at all.
• Technical crawlability. Fast, clean, machine-legible pages that AI crawlers can fetch and parse – including build-level measures such as llms.txt and markdown endpoints, which make a site easier for AI systems to read.

None of these moves the score overnight, and no single one is decisive. But they are the levers – and because the score is tracked monthly, you can see which interventions are actually moving it.

What the score can’t tell you

A measure you can trust is one whose limits are stated plainly, so here are this one’s.

It is not a guarantee. No agency controls what an AI engine generates, and nobody can promise you a citation – the honest claim is that the inputs above improve your chances, and the score tells you whether they are working. It is not market share: a tracked-prompt index tells you how often you appear within that index, not what proportion of all AI answers in your market mention you. And it is not comparable across industries – results vary with industry, competition and content quality, which is why a law firm and a confectionery brand above show such different platform mixes, and why the only comparison that matters is your own score against your own prior period.

How to get yours

You cannot manage what you have not measured, and most organisations asking about AI search have never seen their own baseline. Establishing one is straightforward: define the buyer-intent prompts that matter in your market, measure how often you currently appear in AI answers to them – engine by engine – and report from there.

That baseline is part of our SEO & Marketing Intelligence Report, alongside the entity, content and technical analysis that explains why the number is what it is – and what would move it.

FAQ