Insights
What is AI visibility?
AI visibility is whether AI assistants name your business when someone asks them who to hire. It is a different question from search ranking, it is measured differently, and it can be tested.
Updated September 22, 2026 · 9 minute read
Short answer
AI visibility is whether AI assistants — ChatGPT, Gemini, Claude, Perplexity, Google's AI answers — name your business when a prospective customer asks them who to hire. It is not your Google ranking, and a business can rank well and still never be named. It is measured by asking the assistants the questions your customers ask and recording what comes back: whether you appear, in what position, alongside which competitors, and which sources the assistant drew on.
What actually changed
For twenty-five years, being found online meant one thing: appearing in a list of links, as high up as possible. The work was to rank. The measurement was a position number. Everyone in the industry — agencies, tools, clients — agreed on what the scoreboard was.
Assistants do not return a list. They return an answer. Someone asks “who should I call to take down a dead oak in Gaithersburg?” and gets a short paragraph naming two or three companies, sometimes with a reason attached to each. There is no page two. There is no position eleven. A business is either in the answer or it is not, and the ones that are not get no impression, no click, and no chance to be considered.
That is the change worth understanding. Search rewarded being present on a ranked list. Assistants reward being the thing an answer is made of. Those are related but distinctly different problems, and the second one is not measured by the tools built for the first.
Why this is not the same as ranking
The most common misconception is that AI visibility is a by-product of SEO — that if you rank, you will be recommended. It is closer to say the two overlap without being the same thing.
Where they overlap: for the assistants that search the live web before answering, being indexed and eligible for a snippet is a precondition. Google states plainly that there are no additional technical requirements for appearing in AI Overviews or AI Mode beyond ordinary Search eligibility, and that the snippet controls you already know — nosnippet, max-snippet, data-nosnippet — also limit what its AI features can use. If you are invisible to Search, you are invisible there too.
Where they diverge, and why ranking is not enough:
- An answer names a handful, a results page lists ten. Ranking fourth is a real outcome in search and often a non-outcome in an assistant. The compression is brutal, and it is not proportional to rank.
- Some answers are not retrieved at all. An assistant answering from what it absorbed during training is not consulting your site at that moment. It is recalling how your business was described across the web, if it was described at all. Ranking has no direct influence on that.
- The question is different. “Tree removal Gaithersburg” is a search query. “I have a dead oak leaning over my neighbour's fence, who do I call and what should it cost?” is what people ask an assistant. Those pull on different content.
- The answer is synthesised, not selected. A results page shows you pages. An assistant writes a sentence about your business. That sentence can be wrong — wrong service area, wrong services, a closed location — in ways a ranking cannot be.
The two ways a business ends up in an answer
Understanding this distinction is most of understanding the problem, because the two paths respond to completely different work.
- Retrieval. The assistant searches the web at the moment you ask, reads what it finds, and writes an answer from those pages — usually with citations. Perplexity works this way by design; ChatGPT, Gemini and Claude do it when a question calls for current information. Here, what matters is what is findable and readable right now, and citations make the influences visible.
- Recall. The assistant answers from the statistical impression of the web it formed during training. No live lookup, no citations. Here, what matters is how consistently and how widely your business was described across the sources in that corpus, months or years ago. This path strongly favours businesses that many independent sources have described the same way for a long time.
Why the distinction is practical, not academic
Fixing your website changes what retrieval can find, sometimes within days. It does not change what a model already recalls, and may not for a very long time. Anyone promising a fast fix for the second path is describing something they cannot control.
What AI visibility is actually made of
“Are we visible to AI?” is not answerable as posed, because visibility is not one thing. Broken into parts, each of which can be observed and counted:
| Component | The question it answers |
|---|---|
| Presence | Across the questions your customers ask, how often are you named at all? |
| Position | When you are named, are you first, or fifth in a list of six? |
| Citation | Is your own site among the sources the assistant used, or is it describing you from somebody else's page? |
| Accuracy | Is what it says about you correct — services, area, hours, specialisms? |
| Strength | Are you recommended warmly, listed neutrally, or mentioned with a caveat? |
| Share of voice | Out of all the businesses named across the whole question set, what proportion are you? |
The observable components of AI visibility
These are worth separating because they fail independently and they have different remedies. A business that is never named has a presence problem. A business named often but always fourth has a position problem. A business named accurately but never cited is being described by third parties rather than by itself, which is a source problem. Collapsing all of that into one number and calling it a score tells you nothing about what to do next.
Greenbug's AI Recommendation Score combines exactly these observations, with published weights, into a single 0–100 figure. The weights, the platforms, the prompt method and the stated limitations are all on the methodology page, linked below — a score whose method is not published is a number you are asked to take on trust.
How you measure it without fooling yourself
Assistants are not deterministic. The same question, asked twice, can produce different businesses. Model versions change underneath you. Retrieval results shift. Location and account history influence answers. This makes casual testing genuinely misleading: ask once, see your name, and conclude you are visible.
A measurement worth acting on needs four disciplines, none of them exotic:
- A fixed question set. Questions written the way a buyer would actually speak, frozen so that this month is comparable with last month. Prompts rewritten each period to flatter the client destroy the only thing that makes a trend meaningful.
- Repetition. Every question run more than once per platform, so one unusual answer cannot masquerade as a result.
- Several platforms. They disagree with each other, sometimes sharply. A business can be well known to one assistant and absent from another, and one platform's answer is not a measurement.
- Recorded evidence. Every response stored in full, with its date and the platform that produced it. If a score cannot be traced back to the answers that produced it, it is an assertion.
What nobody can do
No agency, tool or consultant can make an AI system recommend a business. There is no submission form, no ranking lever, no paid placement in an organic answer. What can be improved is the information those systems draw on — whether your business is described accurately, consistently and in enough independent places to be recalled and retrieved. Treat a guarantee of AI recommendations as evidence of a bad actor.
What appears to influence whether you are named
The honest framing here matters. No one outside these companies knows the mechanism, and the companies themselves describe their systems at a level of abstraction that does not settle most practical questions. What follows is what can be reasoned about from published documentation and from what is observable in the answers themselves. It is not a ranking factor list, and anyone presenting one is guessing with more confidence than the evidence supports.
- Being reachable at all. Retrieval cannot use what it cannot fetch. Crawler access, indexing and snippet eligibility are the floor, not the strategy — but a business that blocks AI crawlers has opted out of the retrieval path entirely, usually by accident.
- Being resolvable as one entity. If your name, address, phone number and service description differ across your site, your directory listings and your profiles, a system has to decide which version is you — or whether you are two businesses. Consistency is not a growth hack; it is the difference between being identifiable and being noise.
- Independent corroboration. Your own site asserts. Directories, associations, local press, review platforms and industry bodies corroborate. Recall is built from many sources agreeing, which is why a business with one excellent website and no external footprint often goes unnamed.
- Saying the thing plainly. Pages that state what the business does, where, for whom and at what kind of price are easier to quote than pages built from atmosphere. This is the one place where writing for people and writing for machines genuinely converge.
- Reviews, as content rather than as a number. Star ratings are a summary. The text of reviews describes specific services, places and problems — and that text is material an assistant can draw on.
Where to start
The first useful step is not optimisation. It is finding out where you actually stand, because almost every business that asks us this question is guessing — usually pessimistically, occasionally the opposite.
A baseline answers four things: do you appear, who appears instead, what sources are the assistants leaning on, and is what they say about you even correct. That last one produces the most surprises. Wrong service areas and services the business dropped years ago turn up more often than absence does.
From there the work is ordinary and unglamorous: fix what is wrong about your business's description of itself, earn corroboration where competitors have it and you do not, publish the things a buyer actually asks about — then measure again against the same frozen questions and find out whether any of it moved.
Common questions
Is AI visibility just SEO with a new name?
No, though they overlap. Being indexed and snippet-eligible is a precondition for the assistants that search the web before answering. But an answer names two or three businesses rather than ten, some answers involve no live search at all, and the assistant writes a description of you that can be factually wrong in ways a ranking cannot be.
Can anyone guarantee that AI will recommend my business?
No. There is no submission process and no paid placement in an organic AI answer. What can be improved is the information these systems draw on. A guarantee of AI recommendations is a reason to walk away.
How do I know if I am visible right now?
Ask the assistants the questions your customers ask, repeat each one, do it across several platforms, and record every answer. A single test proves nothing, because the same question can produce different businesses on different runs.
Does blocking AI crawlers hurt my visibility?
It removes you from the retrieval path — an assistant that cannot fetch your pages cannot answer from them. Whether to allow training crawlers is a separate judgement, and reasonable businesses decide differently. The common error is blocking everything without realising search indexing and training are controlled by different user agents.
Sources
Primary documentation from the platforms themselves, cited where this page makes a claim about how they behave. Where something is not documented publicly, this page says so rather than asserting it.
- Google Search Central — AI Features and Your Website →
Google states there are no additional technical requirements for AI Overviews and AI Mode beyond Search eligibility, and that nosnippet, data-nosnippet and max-snippet limit what AI features can use.
- OpenAI — Overview of OpenAI Crawlers →
Documents that OpenAI runs separate crawlers for model training and for search indexing, which is why the two paths into an answer have to be discussed separately.
- Perplexity — Perplexity Crawlers →
Distinguishes indexing for search results from fetches made in response to a specific user question.
More resources
- How do I get my business recommended by ChatGPT?
There is no submission form and no way to guarantee it. What you can do is make sure ChatGPT can reach your site, that your business is described consistently everywhere it appears, and that independent sources corroborate it — then measure whether any of it changed the answers.
- Why does AI recommend my competitors instead of me?
Usually not because they are better. Because more independent sources describe them, describe them consistently, and describe them in the specific terms a customer's question used.
Find out what AI currently says about your business
A baseline records which businesses are named across your buyer questions, on which platforms, and which sources the assistants relied on. Nothing on this page is a substitute for measuring your own.