AI Visibility Audit — Measured Weekly Across Three Assistants

An AI visibility audit measures whether AI assistants name your brand when someone asks about your category — and, more usefully, what they read instead when they do not. A frozen set of prompts goes to ChatGPT, Gemini and Google AI Mode every week. Each answer is parsed for whether you are mentioned, how early, and which sources the model leaned on. What comes back is a ranked list of gaps, not a score out of a hundred.

What the AI visibility audit measures

Five layers, reported against each other rather than as separate lists.

Presence. The percentage of tracked prompt responses that name your brand at all. This is the headline number and it is usually lower than people expect.

Position. How far into an answer your first mention appears, as a share of response length. Being named in the closing sentence is not the same as being named in the opening one, and the two get counted the same by anything that only checks presence.

Competitive set. The same two measures for competitors you name at setup, so presence is read against a field rather than in isolation. Ranking first in a category where the assistant usually names nobody is a different situation from ranking third in a crowded one, and only the comparison tells you which you are in.

Citations. Every source the assistants pulled to build their answers, with Domain Authority and Spam Score attached to each domain, and a count of how many citations were yours. This is the layer most audits skip and the one that carries the work.

Trend. The same prompt set, re-read weekly, so a change can be attributed instead of assumed.

What you get back

A readout, structured as a work queue rather than a document. Every finding names the prompt or the page, the specific change, and why it is ranked where it is. Findings that need a developer, a decision, or a piece of content that does not exist yet are separated out and named as such rather than padded into the same list.

What you will not get is a single AI visibility score. Blending three assistants that disagree with each other into one number destroys the only signal worth having, and optimising toward that number is how a brand ends up with a rising dashboard and no more mentions.

What an AI visibility audit cannot tell you

Four honest limits, because an audit that hides them is selling certainty it does not have.

The prompt set is a deliberate sample of how buyers ask about your category. It is not the whole market, and presence figures move if the frame moves — which is why the set stays frozen between readings rather than being quietly re-cut to flatter a trend.

Assistants are non-deterministic. The same prompt can return a different answer twice in a row. At low presence levels a one-week movement of a point or two is noise, and we will say so rather than reporting it as progress.

Nothing here estimates traffic or revenue from AI search. Those figures are not measurable from this data, so we do not offer them.

And a citation is not a click. Being named in an answer is worth having on its own terms; attaching a visit number to it would be invention.

How this differs from a free AI visibility checker

The free tools in this space — brand graders, one-shot visibility reports, instant AI checkers — send your brand to a model once and report the reply. For a first look that is genuinely useful, and if you have never checked, check. We run a lighter version of exactly that inside our free SEO audit, which covers whether an AI Overview triggers for your terms and who it cites.

What a single reading cannot do is separate signal from variance, and it cannot tell you what the model read instead of you. The citation layer is the difference. Knowing that a mid-authority specialist publication is cited fourteen times on your core question is a decision you can act on this month. Knowing your brand scored 31/100 is not.

Who this is for

Brands in categories where buyers research before they buy, and where an assistant now sits between the question and the website. If your customers ask short factual questions with obvious answers, the assistant will answer them without anyone's brand and there is little to win. If they ask comparative questions — which, best for, is it worth it, how does X differ from Y — the assistant is assembling an answer from sources right now, and either you are in that set or a competitor is.

What it costs

$2,500, fixed scope. That buys the full reading: 50 tracked prompts across ChatGPT, Gemini and Google AI Mode, collected weekly through the audit window, the complete citation layer with Domain Authority and Spam Score on every cited domain, up to three named competitors measured alongside you, and a findings list ranked by a person rather than sorted by a tool.

Two things move the number, and both get quoted on the call rather than guessed here: a materially larger prompt set, and additional competitors or markets. A bigger question set is a bigger measurement, and we would rather scope it with you than sell you a size that does not fit.

What is deliberately not priced on this page is the ongoing programme. Continuous tracking and the work of closing the gaps is scoped from your keyword universe, the same way every other module here is — how engagements are scoped sets out why a published tier either overcharges a small scope or under-resources a large one. The audit is a deliverable with a clean end. The programme is not, so it is not sold like one.

For context on the alternative: the tracking platforms this kind of AI visibility audit draws on run roughly $180 to $400 a month before anyone reads the output, and running the equivalent yourself is usually costed at thirty to sixty hours of internal time. The tools are not the expensive part. Deciding what the data means is.

What happens after the audit

The findings are yours regardless of what you do next. If you want the work done, the audit routes into the modules that do it: Content Gap Engine for the questions with no page behind them, Precision Page Deployment for pages being read but not credited, Authority Engine for the publications the model actually cites, and AI Search Visibility if you want the measurement to continue as an ongoing programme rather than a one-off reading.

The audit is the reading. The module is the engine. You can take the first without the second.

Frequently asked questions

What is an AI visibility audit?

An AI visibility audit measures whether AI assistants name your brand when someone asks about your category, and what they read to decide. A fixed set of prompts is put to the assistants every week, and each response is parsed for three things: whether your brand is mentioned at all, how early in the answer it appears, and which sources the model cited to build the answer. The output is a ranked list of where you are absent and what the model read instead.

How is this different from a free AI visibility checker?

Most free checkers send your brand name to a model once and report what comes back. That is a snapshot of one response on one day, and assistants are non-deterministic - the same prompt can answer differently an hour later. This audit runs a frozen prompt set on a weekly cadence, so movement means something, and it reads the citation layer, which is where the actionable work is. A one-off check tells you that you are invisible. This tells you which pages the model is reading instead of yours.

Which AI assistants does it cover?

ChatGPT, Gemini and Google AI Mode, each tracked separately. They do not agree with each other, and that disagreement is useful: a brand strong in one and absent in another usually has a source problem rather than a content problem. Reporting a single blended 'AI visibility score' across engines would hide exactly the thing worth seeing.

What do I actually get?

A readout covering brand presence across the prompt set, average position of first mention, how you place against named competitors, the trend across collections, and the full citation layer - every domain the assistants cited, with Domain Authority and Spam Score attached, and how many of those citations were yours. Findings come back as a ranked list of specific changes, not a scored badge.

How long before the numbers move?

We do not put a date on it, because nobody can. What we can say is what governs it: how much source material the assistants currently pull for your category, whether the pages they already read attribute you clearly, and whether the questions they research hardest have an answer on your site at all. The first reading tells you which of those three you are dealing with. Movement is reported against the same frozen prompt set, or it is not reported.

Do I need this if we already do SEO?

They are not the same measurement. Classic SEO asks whether you rank on a results page. This asks whether you are named inside an answer, which can be true when you do not rank and false when you rank first. In the worked example behind this page, the site was the single most-cited domain in its category and was still named in only four per cent of responses - the model was reading it and not crediting it. Ranking did not surface that; this did.