AI Visibility Audit

People now ask ChatGPT, Gemini, Perplexity and Google’s AI Overviews the questions they used to type into a search box, and the answer names a handful of brands. An AI visibility audit establishes, with evidence rather than assertion, whether yours is one of them — and if it is not, which specific sources the assistants are quoting instead.

What an AI visibility audit actually tells you

It tells you three things: how often you are named or cited when a real buyer asks a real buying question, which sources the assistants use to answer that question instead of you, and what the model believes about your brand — including anything it has got wrong. Everything else in the report follows from those three findings.

What we analyse

Prompt-set design. The audit is only as good as the questions it asks. We build the prompt set from your actual buying questions — the ones in your sales calls, your search console data and your support inbox — not from a generic template. A prompt nobody would ever type produces a finding nobody can act on.

Citation share. Across the prompt set, how often you are mentioned, how often you are cited with a link, and where in the answer you appear. Being named in the third sentence of a recommendation is worth considerably more than a footnote among nine other sources.

Source attribution. Which URLs the assistants actually pull from. This is the most useful part of the report and the least intuitive, because the answer is frequently not your website. It is often a review site, a directory, a Reddit thread, a comparison article or a trade publication — which tells you exactly where the work needs to happen.

Entity coverage. Whether the models have a coherent understanding of who you are: what you sell, where you operate, who you compete with, what you are known for. Fragmented or contradictory entity data is one of the most common reasons a well-ranking site is invisible in AI answers.

Schema and technical readiness. Whether your pages are server-rendered and readable by assistant crawlers, whether your structured data is valid, and whether your key facts are stated in extractable form rather than buried in marketing copy.

Competitor comparison. The same prompt set, run against the two or three competitors you name, so your score has a scale. A visibility score with no comparison is a number without a meaning.

The 21-prompt method

We run 21 prompts, built in three groups of seven. Seven are category questions, where the buyer knows the problem but no brands (“what should I look for in a…”). Seven are comparison questions, where they know two or three names and are choosing between them. Seven are branded questions, testing what the models say about you specifically, including whether the facts are right.

Twenty-one is a deliberate number. It is enough for a pattern to be distinguishable from noise, and few enough that we can run the whole set across every major engine — ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity, Claude and Copilot — and re-run it identically later. That last point matters more than the size of the set: an audit you cannot repeat gives you a snapshot, not a measurement. We re-run the same prompts monthly for retained clients, and set a re-test date for one-off audits.

One caveat we state on every report: these systems are non-deterministic. The same prompt can return different answers to different users on different days. That is why we run the set repeatedly and report a share across runs rather than treating a single response as fact — and why you should be sceptical of anyone quoting a precise, unrepeated number.

What you’ll receive

A scored grid: all 21 prompts against every engine, showing where you were mentioned, where you were cited, and where you were absent. A competitor comparison over the same grid. A ranked fix list, ordered by which changes are most likely to shift citation share, separating what we can change on your site from what has to be earned off it. And a re-test date, with the prompt set stored so the second run is genuinely comparable to the first.

Proof

Official London Theatre reached 85% AI search visibility and is the #1 recommended source for its core queries — the strongest AI visibility result we have measured.

The honest counterweight, and we would rather publish it than hide it: ask ChatGPT for the best SEO agencies in London and it names no agency at all — us included. It talks about how to choose one instead. That is what invisible looks like on a high-value commercial query, and it is a fair description of where most of this industry currently sits, including the agencies advertising GEO services. It is also why we run the audit on our own domain to the same standard we run it on yours.

Run the free version first

Before you buy anything, run our GEO visibility score. It queries multiple AI platforms and returns a score out of 100 with a platform-by-platform breakdown across AI Overviews, ChatGPT, Perplexity and Gemini. It takes seconds and costs nothing. Most businesses score between 20 and 40 the first time they check.

If your score comes back high, you may not need us yet. If it comes back low, the free tool tells you that you have a problem but not what is causing it — it cannot show you which competitor sources are being cited in your place, or whether the cause is entity confusion, thin source coverage or a rendering fault. That diagnosis is the paid audit. Our GEO audit tool is a useful middle step if you want more detail before committing.

What the audit costs

The audit is scoped to the brief. The variables are how many markets and languages you need covered, how many competitors you want benchmarked, and whether you want a one-off diagnosis or a tracked baseline we re-run each month. You get a fixed price before any work starts, and the free tools above cost nothing if you want to see the shape of the problem first.

If the audit leads to implementation, that work sits inside an SEO and GEO retainer from £1,100 a month ex VAT. Get in touch and we will tell you which of the three you actually need.

Frequently asked questions

How is this different from an SEO audit?

An SEO audit asks whether you rank. This asks whether you are cited. The two overlap — assistants draw heavily on pages that already rank well, so strong SEO is a large part of the answer — but they are not the same. Plenty of sites rank on page one and are never named in the AI answer above it, usually because of weak off-site source coverage or incoherent entity data.

Which AI engines do you test?

ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity, Claude and Microsoft Copilot. They behave differently: some lean on live search results, others on training data, others on a small set of trusted publishers. The per-engine breakdown usually points straight at the cause.

Why 21 prompts and not 100?

Because a prompt set has to be re-runnable to be a measurement. Twenty-one covers the category, comparison and branded intents that matter commercially, and can be run identically across seven engines, month after month. A hundred prompts run once produces a bigger document and a worse baseline.

Can you guarantee I will appear in ChatGPT answers?

No, and nobody honestly can. There is no submission process, no paid inclusion and no ranking dashboard. What we can do is establish where you stand today, identify what the engines are citing instead of you, and work on the things that demonstrably correlate with being cited — ranking, entity clarity, off-site source coverage and extractable content.

How often should the audit be re-run?

Monthly if you are actively working on visibility, quarterly if you are monitoring. These systems change their retrieval behaviour without notice, so a baseline from six months ago tells you very little about today.

Do I need to fix my website, or something else?

Usually both, and often the off-site half matters more. If the assistants are citing a comparison site, a review platform or a forum thread when answering your category question, no amount of on-site work will get you into that answer. You need to be present, accurately, in the sources they already trust.

Does structured data improve AI visibility?

It helps the models understand your entities and relationships, and valid markup is worth having. It is not a switch that makes you cited. Being server-rendered and readable matters more than any particular markup, and you can check what your pages currently expose with our schema report.

Is this worth doing if my customers are not using AI search yet?

The audit is cheap relative to being late. If your buyers are not asking assistants yet, the baseline still costs little and tells you where you stand before your competitors think to ask. If they are, you will want to know which competitor the answer currently recommends.