AI Visibility Audit
Find out whether the answer engines can reach you, resolve you and quote you — before paying anyone to improve any of it.
Most businesses asking why they are not showing up in ChatGPT or AI Overviews are asking the wrong question first. Before visibility is a content problem it is an access problem, and before it is an access problem it is an identity problem. An audit that starts at content skips the two failures that actually explain most cases.
An AI visibility audit answers three questions in order. Can the retrieval crawlers behind the major engines fetch and render your pages at all. Can a model work out which company you are from the records it can find. And when someone asks the question your business exists to answer, whose name comes back instead of yours.
The value of doing it as a discrete piece of work is that the answers are frequently unflattering and frequently cheap to fix. A robots directive written in 2021 is not a strategy problem. Neither is a Google Business Profile listing a category the business abandoned two years ago. Both are invisible until someone looks.
At a glance
Crawler access, settled from logs
Whether GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended can reach your pages, read from server logs and robots directives rather than inferred. It is a yes or no question and it gates everything after it.
Rendering, not just fetching
A page a crawler receives as an empty shell is a page it cannot quote. Client-rendered content that a browser assembles fine is a common and silent failure at this layer.
Entity resolution
Every public record of the business checked against every other: site, Google Business Profile, directories, registries, social. Where they disagree, a model has to guess which version of you is real.
A prompt panel, baselined
A fixed set of the questions your buyers actually ask, run across ChatGPT, Claude, Gemini, Perplexity and AI Overviews, recording whether you are named, how you are described, and which competitors are named in your place.
Want results like this for your brand?
Talk to a Growth StrategistWhy the order matters more than the checklist
Audits get sold as coverage — the number of checks run, the length of the document. The useful part is the sequence, because these failures are strictly dependent and fixing them out of order wastes the budget.
If the crawlers cannot fetch you, nothing downstream is measurable. Content quality, schema, corroboration and authority are all arguments about a page no engine has read. Teams routinely spend a quarter on content while the access problem sits unexamined because it is unglamorous and takes an afternoon to check.
If the engines cannot resolve you, everything you build gets credited unpredictably. Depth accrues to whichever variant of your business a model settled on, which may be an old address, a former trading name or a directory listing you have never seen. This is the failure that most looks like bad luck and least is.
Only once those two hold does the content question become a real question. That is also the point at which a prompt panel starts measuring your work rather than measuring an obstruction.
What the prompt panel is, and what it is not
It is a frozen list of questions, written before any work ships, run on a schedule across every major engine, with the result recorded each time. The prompts come from how your buyers actually speak — sales call recordings, support tickets, the questions that come up on every first call — rather than from keyword tools, because people do not talk to an assistant the way they type into a search box.
It is not a ranking report. Individual runs vary: ask the same engine the same question twice and the answer can differ. That variance is a property of the systems, not a defect in the measurement, and any vendor presenting a single run as a position is either misunderstanding it or counting on you to.
What does not vary randomly is the trend across a stable panel over weeks. How often you were named, how you were described when you were, and who was named instead. That is the measurement, and it only works if the panel is frozen — quietly adding prompts you have started winning is the most common way this kind of reporting gets corrupted.
The baseline has to be recorded before anything ships. A panel first run after the changes are live is not a baseline, it is a screenshot.
Reading a bad result correctly
A first panel run usually shows the business named rarely or not at all, and competitors named consistently. The instinct is to read that as a content deficit and commission more content. It is usually not.
Check retrieval first. If the crawlers are blocked or the pages render empty, the absence is mechanical and the fix is technical. Check resolution second. If the engines are naming a competitor whose details agree with themselves everywhere while yours do not, the absence is an identity problem and more content makes it worse by adding another inconsistent record.
Only when access and identity are sound does a thin result mean what people assume it means: that there is nothing specific enough about the business for a model to have a reason to name it. That is a real finding and it is not solved by an agency. It is solved by the business producing something worth citing.
How this is delivered
There is no separate fixed-price AI audit product on this site, deliberately. The crawler-access checks, the entity and schema audit and the prompt panel across every major engine are all inside the full website and marketing audit, alongside the paid, social, lifecycle and analytics review — because in practice the answer to why an assistant does not name you is often somewhere else in the account.
Where a business genuinely only wants the AI visibility half, it gets scoped as a project rather than sold as a smaller version of the same document. Ask for it directly and we will price it.
What we will not do is run the audit and hand over a findings list with no sequence attached. A list of problems ordered by severity rather than by dependency is the standard audit deliverable, and it is why most audits get read once and filed.
AI Visibility Audit: common questions
How is this different from an SEO audit?
An SEO audit asks whether Google can crawl, index and rank your pages. This asks whether the answer engines can retrieve, resolve and quote your brand. They overlap at the technical layer and diverge completely after it: entity consistency across third-party records and whether a model names you in an answer are not things a traditional SEO audit looks at, and ranking position is not something this one reports.
Can I just check this myself by asking ChatGPT about my company?
Asking once tells you very little, because answers vary between runs and the model may be drawing on memorised training data rather than retrieving your site. A useful check needs the same questions asked repeatedly over time across several engines, with the results written down. You can absolutely do that yourself; the value of paying someone is the log analysis and entity reconciliation sitting underneath it, which is the part that explains the result rather than just recording it.
What if the audit finds nothing wrong?
It sometimes does at the technical layer, and that is a useful outcome: it means the absence is not mechanical and the money should go into being worth citing rather than into another technical project. We would rather deliver that finding than manufacture a backlog.
Will you tell me I need a retainer?
Only if the findings support one. A business whose problem is one robots directive and an inconsistent Google Business Profile needs a fortnight of work, not a monthly engagement, and the audit says so. The audit is a fixed-price piece of work that stands on its own and is not structured as a sales call with a document attached.
Do you re-run the panel afterwards?
Inside an engagement, weekly, against the frozen baseline. As a one-off audit you get the baseline run and the prompt list itself, so the panel is yours to re-run whether or not you work with us.
Want to be the brand the models name?