AI visibility audit

An AI visibility audit is a systematic inventory of how your brand currently appears in AI answers: it measures today’s metrics and locates whether each problem sits in the entity layer, the content layer or the technical (crawl) layer.

It produces two things: a set of baseline data you can retest against later (see baseline retest), and a list of problems grouped by layer. Without that grouping, “AI can’t find us” is a complaint, not an action plan.

The three layers

• Entity layer: is the company the model is talking about actually you? Check whether your brand entity is uniquely identifiable, how high your name confusion rate is, whether knowledge graph and encyclopedia-style records exist and agree, and whether sameAs ties your profiles on different platforms to one organization.
• Content layer: is there a page that answers common questions in a form that can be lifted directly? Check for answer-first writing, valid structured data, sourced facts, and whether third-party evidence matches what you say about yourself.
• Technical layer: can crawlers get to the body text? Check that robots.txt is not blocking them by mistake, that the main content does not depend on JavaScript (see SSR vs. CSR), and that your sitemap and canonical tags are correct.

What to provide before the audit starts

• The brand’s full legal name and common short names.
• The official website and target markets.
• Main products, and any competitor information you are comfortable sharing.
• The questions customers actually ask, grouped into three stages: learning, comparing and buying.

If your brand name could be shared with another company, add industry and region qualifiers to the questions, and keep a set of category questions that do not mention your brand at all.

Separate a website check from live platform testing. A website check can find pages that don’t load, missing information or contradictory claims. Platform testing means actually asking the chosen AI platforms the questions and saving the answers. Crawling your own site alone cannot prove whether any platform recommends you.

What a reviewable result should include

Record	Why it matters
Exact question text, platform, date and test conditions	Defines the scope and makes retesting possible
Raw answers and the links actually cited	Lets anyone verify brand mentions and factual claims
Valid sample count, missing samples and failure reasons	Separates “not mentioned” from “not measured”
Problem pages, evidence, recommended action and priority	Turns diagnosis into edits someone can make

Every conclusion should state the question set, the platform list and the dates used; otherwise there is nothing to line a retest up against.

How you read the findings matters. If an answer gets your industry wrong, check your official entity information and the cited sources first. If a platform returned no answer, retest rather than blaming your content. And correlation is not causation: a rise in mention rate after you edit a page might also reflect changes in the questions, the platform version or the sources being cited.

How to schedule the retest

Keep the original question set unchanged, and put questions for new business lines in a separate group so you are not comparing totals across different samples. Confirm that website fixes have actually gone live before you retest, and record the retest date and conditions. If a platform is unreachable, mark the samples as missing; don’t generate a substitute score.

Limits

An audit describes the current state and likely causes; it cannot prove that fixing them will make things better. Technical and content problems can be confirmed and fixed. Entity-level perception, on the other hand, has to wait for models to re-crawl and update, on a timeline nobody controls.

To see what the Moments audit covers and what to prepare, go to the free GEO audit. For the background method, see generative engine optimization (GEO). The value of an audit is finding problems you can verify and next steps you can take — not a promise of rankings.