AI visibility audit: check the answer, the source and the claim

An AI visibility audit records whether assistants mention or cite your company for relevant buyer questions, and whether those answers are accurate. Keep the prompt and source evidence with every finding so the next review can test the same question.

By ·

A screenshot of a flattering answer is a poor audit. It does not tell a founder which question produced it, whether the answer used current sources or what the team should fix. The useful output is a short list of evidenced problems with owners.

Define the decision the audit will support

Choose the audience and the buying problem first. A SaaS marketing lead assessing discovery needs questions asked before a buyer knows the brand. A team correcting an outdated product description needs branded questions too. Keep those groups separate in the report. If the team has not yet chosen which buyer questions matter, settle that in an AEO strategy first.

State which products and markets the audit covers. Record exclusions, such as a product still in development. Otherwise the report can expand into a test of claims the company has never made.

Choose the assistants or search experiences you want to observe and record their names. Avoid presenting a result from one product as a conclusion about every AI system. The coverage statement belongs beside the findings.

Build the prompt set from real buying questions

Start with questions sales or customer-facing colleagues can recognise. Remove private customer details before using them in an external tool. Preserve the constraint that makes the question meaningful, such as a workflow requirement.

For a reporting tool, compare a broad category question with a question about who can approve a report before it is shared. The second prompt tests a specific buying concern. Do not insert your brand into every question and call the resulting mentions discovery.

Save the wording before you run the audit. If you rewrite a prompt after seeing an unfavourable answer, label it as a new test. Keeping the old result makes the report easier to challenge and more useful.

Keep an evidence log someone else can inspect

Use a row for each observed answer. This table defines the fields; it contains no claimed audit results.

FieldWhat to saveWhy it matters
Question and contextExact prompt, intended buyer and constraintsPreserves what was tested
Run conditionsDate, product or mode, search use and relevant session settingsMakes differences between runs visible
Answer evidenceSaved answer with citation destinationsLets a reviewer inspect the source
Company treatmentAbsent, mentioned, linked or misdescribedSeparates visibility from accuracy
ActionSpecific issue, owner and retest conditionTurns observation into work

Keep the evidence even when there is no mention. A file containing only favourable examples conceals the scope of the test. If you report a proportion later, show the underlying sample and how the questions were selected.

Inspect the sources before proposing a fix

Open each citation and check what it actually supports. The answer might link to your domain while describing another company, or cite a page that does not contain the stated capability. Record the precise mismatch.

OpenAI's ChatGPT search guidance, checked 16 November 2026, describes both inline citations and a Sources panel. Use the source URL as evidence; a brand name in the answer alone is a different observation. The ChatGPT SEO guide covers the crawler settings that decide whether ChatGPT search can reach a page.

Trace an inaccurate statement to a source when possible. If your own product page is ambiguous, fix that explanation. If an external profile is outdated and your team controls it, update it through the normal process. If you cannot identify the cause, say that the cause is unresolved.

Separate access problems from content problems

Ask the site owner to test the exact public URL that should answer the question. A missing citation does not prove that a crawler was blocked. Likewise, allowing access does not prove that an assistant will choose the page.

For Google, its guidance on AI features says the same SEO fundamentals apply to AI Overviews and AI Mode, with no special optimisation required. Have the site owner check the affected page against Google's ordinary search requirements, without changing deliberate restrictions on private material.

For content issues, give the reviewer the unsupported claim and the buyer question. Ask for an accurate explanation with its conditions. A vague instruction to make the page more AI-friendly is not an actionable brief. The AEO checklist gives the reviewer a page-level check to run before republishing.

Prioritise errors a buyer could act on

A wrong description of what the product does deserves attention before a missing mention on a loosely related question. Check whether an error would cause a buyer to reject a suitable product or expect something the company cannot deliver.

Assign the issue to someone who can correct the source. Product approves capability statements. Marketing owns the public explanation. The site owner diagnoses access. Record the dependency when a fix needs more than one team.

Use the marketing leadership role comparison if nobody owns the review across teams. Defining that responsibility is a hiring and management decision, separate from buying an audit tool.

Retest without claiming more than the evidence shows

Repeat the saved questions after a documented change. Record any change in test conditions alongside the new answer. A different answer is an observation; the audit alone may not establish what caused it.

Keep citation findings separate from referral traffic and enquiries. A visibility report cannot tell you whether a prospective customer reached the site unless you also inspect that evidence. Connect commercial questions to the wider acquisition-cost review.

Close the audit with the unresolved issues and their next checks. Keep the evidence log available to the people doing the work.

People also ask

Is an AI visibility score enough to judge performance? Ask how the score is calculated and which questions it covers. Keep the underlying answers and source links so the team can distinguish an accurate citation from a misleading mention.

What is the difference between a mention and a citation? A mention names the company in the answer. A citation provides a source link, which still needs to be checked for relevance and accuracy.

What should the audit deliver? An inspectable evidence log and a prioritised list of issues with owners. Record unresolved causes explicitly instead of filling the gaps with assumptions about how an assistant ranks sources.

Hiring someone to own search?

Scalerr recruits marketing leaders for tech companies, from a first Head of Marketing to a CMO. Brief us on a marketing leadership hire, or use the form below.

Request a call

Brief us on a marketing leadership hire

By sending, you agree to be contacted by the Scalerr team about your enquiry.

Scalerr's newsletter

Subscribe on LinkedIn for new editions.