What is AI visibility?

The short version

AI visibility is how often an assistant names your business when someone asks it for a recommendation. Not whether you rank. Not whether you appear in a list of links. Whether ChatGPT, Gemini, Claude or Perplexity says your name when a buyer asks it who to use.

It is measured as a share of answers, per assistant, across the questions your buyers actually ask — because assistants do not answer identically twice, and they do not agree with each other.

What is AI visibility?

Someone opens ChatGPT and types who should I use for accounting in Manchester. The assistant does not return ten blue links. It answers in two or three sentences and names three or four firms.

AI visibility is whether one of those names is yours, and how often.

That is the whole definition. The rest of this article is about why it cannot be read off your existing reports, how it is measured properly, and what a number actually tells you once you have one.

Why is it not a ranking?

A search ranking is a position in an ordered list, and the list is the same for everyone who types the same thing at the same moment. An assistant answer is neither ordered nor stable. Three differences follow.

There is no position three. There is a short list of names, and you are on it or you are not. Anyone reporting a “rank” for an assistant answer is reporting something they invented.

The same question gives different answers. Ask an assistant the same thing five times and you get overlapping but not identical lists. This is not a bug; it is how the systems work.

Each assistant answers differently. They read different sources and weigh them differently, so a business named constantly by one can be absent from another.

That last point is the one people underestimate most. When we asked four assistants the same questions about one local market, they named 65 businesses between them — and only four of those were named by all four. Forty-six appeared in a single assistant only.

An average across four would have hidden the only thing worth acting on.

How is it different from an AI Overview?

They are related and frequently confused.

Being cited in an Overview does not mean ChatGPT will name you, and being named by ChatGPT says nothing about Overviews. They read different things. Treating “AI search” as one surface produces reporting that is wrong about most of it — which is why we measure and report them separately. The Overview side is covered in Google AI Overviews optimization.

How do you measure AI visibility?

Four parts, and skipping any one of them produces a number nobody should act on.

Writing the questions

Questions are written the way a buyer asks them, not the way a search box is typed. Where do I go for an emergency dental appointment in Odesa rather than dentist Odesa. People type keywords into Google and sentences into assistants, and a question set built from a keyword export misses the phrasing entirely.

They are grouped by what the buyer wants:

• general recommendation — who should I use for X
• specific service — who does Y in this city
• price — how much does Y cost, who is affordable
• urgency — who can see me today, who is open now
• trust — who is reputable, who do people recommend
• comparison — X or Y, which is better for me

The grouping is not decoration. Visibility is rarely uniform across it: a business is often named constantly in general questions and never in price questions, and that gap is where the work is.

For a real measurement we write the set from the client’s own site — reading their pages and building questions about the services they actually sell, in their city and language.

Why each question is asked several times

One answer is an anecdote. Ask five times and you can distinguish “usually named” from “named that one time”.

We ask each question five times per assistant by default and report the share of answers that named the business. That is the difference between a measurement and a screenshot.

Why assistants are never averaged

Because a blended figure hides the assistant you are absent from — which is the one worth knowing about. In the same measurement mentioned above, shares for one business ranged from 100% of answers on one assistant down to 50% on another, for identical questions asked at the same time.

Every figure we publish is per assistant. The reasoning, with the numbers, is in Gemini and Claude visibility.

Keeping the raw answers

Any percentage should trace back to the exact text an assistant produced. If it cannot, it is not checkable — and an unverifiable number is worth less than no number, because it invites an argument you cannot win.

The full method is published here, and there is a complete sample report open on the site showing what the output looks like.

What does a low score actually tell you?

Less than people assume, and more than a single check can.

It does not tell you that your website is badly designed. In one market we measured, a single assistant answer rested on eleven pages on average — the business’s own pages alongside directories, review sites and local roundups. A low share usually means the facts a buyer asks about are missing or contradictory across those pages, starting with your own, rather than that your homepage looks wrong.

It does tell you where to look. The names that came back instead of yours are the businesses the assistant currently trusts in your category, and the pages it cited are the pages that decided it. Between those two lists, you have a map.

It does not tell you why. Nobody outside OpenAI, Google, Anthropic and Perplexity knows how these lists are assembled. What is observable is what gets read; the rest is inference, and we label it as such.

What moves AI visibility?

In the order that tends to matter:

1. How your business is described in the sources assistants read. Being present, accurate and complete on the directories and roundups that come up in the citations.
2. Consistency across those sources. Assistants cross-check. Three different service lists in three directories is a reason to name someone else.
3. Whether your own pages answer the buying question. In the first two sentences, in the words a buyer uses — assistants extract passages rather than reading brochures.
4. Whether you have published anything checkable. Specific numbers, stated method, real examples. Assistants prefer a source that says something specific to one that says something impressive.

The expanded version, with what to do at each step, is in how to improve AI visibility.

How often does it change?

Enough that a single measurement is a snapshot rather than a position.

We treat movement under five points as noise rather than progress, because that is roughly the range answers drift in without anything changing on your side. A programme that reacts to every four-point swing will spend its budget chasing weather.

This matters commercially as well as technically. A supplier who reports a four-point rise as a win is either not measuring properly or is hoping you are not.

Who should care about this?

Businesses whose customers ask for a recommendation before choosing. Clinics, professional firms, trades, local services, and software that people evaluate before buying. If buyers in your category ask an assistant who should I use, you are already being scored by it, whether or not you are watching.

Not everyone. Some purchases are made by walking past a shop or by renewing a contract, and assistants play no part. The cheapest way to find out which side you are on is to ask your own buying questions and see whether the assistants name anyone at all.

Common mistakes

Checking once and drawing a conclusion. The single most common error. Answers vary; one ask tells you almost nothing.

Reporting an average across assistants. Produces a number that describes none of them and cannot be acted on.

Measuring traffic instead of mentions. An assistant may name you without linking to you; the buyer arrives later by typing your name, and analytics records it as direct traffic with no explanation. The event you care about happens where you have no logs.

Assuming it is an SEO problem or that SEO is now irrelevant. Neither. Assistants read the open web, so pages still matter — but the deciding pages are mostly not yours. The comparison is in GEO vs SEO.

Buying a guarantee. The lists are not published, not stable and not for sale. Anyone promising placement is guessing.

How is this different from the tools that track it?

Most AI visibility tools measure a business against a standard panel of prompts — the same questions for everyone in a category, usually in English, usually without a city. That produces a number that is comparable across customers and not especially related to how anyone actually buys.

The differences worth knowing when you compare suppliers:

Neither is wrong. A panel is cheaper and fine for watching a trend across a large brand. For a clinic in one city, a question about “dental services” in general answers a question nobody asked.

What does a measurement cost, in effort?

For you: an hour at the start, mostly confirming that the question set sounds like your buyers and correcting the ones that do not. After that, reading a report.

For whoever runs it: writing the set, running it, reading every raw answer, and deciding which of the findings are worth acting on. The last part is the one that cannot be automated, which is why a measurement that arrives as a bare dashboard is usually less useful than it looks.

A short glossary

Share of answers — the percentage of answers to a question type that named the business, across repeats. Replaces rank as the metric.

Question type — a group of buying questions with the same intent: price, urgency, trust, comparison, general recommendation, specific service.

Drift — movement caused by the system’s own variability rather than by work. Under about five points, assume drift.

Citation — a source shown with an answer. Visible on Perplexity and usually on AI Overviews; frequently absent in chat answers.

Citation — a source shown alongside an answer, where the surface shows them at all.

Where to start

Measure once, before changing anything. Without a baseline, the number you get in six weeks answers no question, because compared to what? has no answer.

• The free AI visibility check runs a short measurement on your business and emails you what the assistants said. No charge, no call.
• The method explains how the question set is written and why each question is asked several times.
• The sample report is open in full — no form — if you want to see the output before talking to anyone.
• For the wider picture of what to do with a result, start with the AI search optimization guide.