What the audit measures
An AI visibility audit takes the questions your buyers ask, puts them to answer engines the way a buyer would, and records what comes back. For each answer it establishes three things: whether your brand is named, whether your website is cited as a source, and whether what the engine says about you is true. Everything else in the report (scores, competitor share, fix lists) is derived from those three facts across the sample.
The unit of measurement is an answer, not a ranking. An engine either names you in its reply or it does not; it either lists one of your pages among its sources or it does not. That makes the result easy to verify: every count in a good report points back to a captured answer you can read yourself.
Why a sample, not a single chat
If you have ever asked ChatGPT the same question twice and received two different shortlists, you already know why one conversation proves nothing. Answer engines are not deterministic: the same question in a fresh session can produce a different set of names, a different order and a different set of sources. Location matters too, since a question asked from Tallinn draws on different retrieval than the same question from London.
An audit deals with this by running each question more than once in fresh sessions, from your market, across several engines, and reporting the variance instead of hiding it. A brand that appears in one run and not the other is reported as exactly that. The 15 questions in our standard audit are each run 2 times on four engines, which gives a sample large enough to show patterns and small enough to read in an evening.
Which engines matter
We test four answer engines: ChatGPT, Perplexity, Google AI Overviews and Claude. They retrieve differently. Perplexity is built around retrieved sources and shows them prominently; ChatGPT with search and Claude with web search retrieve when the question calls for current information; Google AI Overviews appear above some search results and draw on Google’s own index. Google documents that AI features in Search use the same crawling and indexing as ordinary Search, which is why your crawl settings matter for them.
A useful audit tests all four and reports them separately, because a company can be well represented on one and invisible on another. It also records when an engine shows no AI answer at all for a question, since that is a finding in itself.
What a complete report contains
- Every answer, verbatim, with its sources and a timestamp. Without this the rest is assertion. The appendix should let you check any claim the summary makes.
- Mention and citation rates by engine and by question type. Commercial and comparison questions (“best”, “X vs Y”, “alternatives to”) are where buying decisions happen, so they should be reported apart from awareness questions.
- Who is named instead of you. The competitors that appear in the answers to your questions, how often, and on which question types. These are counts within the sample, not market share.
- Accuracy flags. Claims an engine makes about you that contradict the facts you supplied (prices, locations, services, founding details), graded by how much damage they would do if a buyer believed them.
- Technical readiness. Whether your robots.txt allows the AI crawlers that feed live answers, whether key pages render as text without JavaScript, structured data, sitemap, and page-level citability.
- A prioritized fix list. Specific changes with the page to edit, the change to make, an effort estimate and a suggested owner. Brand-safety corrections come first, then citation gaps on high-intent questions, then technical and content work.
How to judge an audit before buying it
Ask to see a sample report. Then check five things. Does it show the actual answers, or only charts? Does it say how many times each question was run and from where? Does it separate “named” from “cited”, which are different outcomes with different fixes? Does it check what the engines got wrong about the company, not only whether they mentioned it? And does the fix list name pages and changes, or does it say “improve your content”?
Be wary of two things. Industry benchmarks with no stated sample (“companies like yours appear in X% of answers”) are rarely verifiable. And any promise of a future mention is a promise nobody can keep. The honest claim is narrower: here is where you stand today, here is why, here is what to change, and here is how to tell whether it worked.
What happens after the audit
The fix list is where the value is realised, and most of it is ordinary work: correct a wrong fact on your own site and on the third-party pages the engines cite, open robots.txt to the crawlers that feed live answers, rewrite a key page so its first paragraph answers the question it is meant to answer, add structured data where it describes something real. Our guide on what is in your control walks through the levers in order.
Because engines change, a baseline is only useful if you repeat it. That is what monitoring is for: the same questions on the same engines each month, with the differences highlighted, so you know whether a fix moved anything.
Start with a free check
You do not need a full audit to find out whether there is a problem. Our free check reads your homepage, writes three questions your buyers would ask, puts them to ChatGPT and Perplexity and emails you the answers with your brand and your competitors marked. If you are named and cited in all of them, you may not need more. If you are absent, you will know what the full audit is for. The methodology page describes the exact procedure we use, and the sample report shows what you receive.