A business that does not appear in ChatGPT, Claude or Perplexity answers, or is not cited by AI, is missing for a cause found by walking up a fixed chain: the crawler does not get in, or it has nothing to read, or it does not know who is speaking, or the page has no date, or the business is cited and nobody measured it. This page follows that chain branch by branch, with the GEO test check to read and the page that explains the fix.
Run the test on your home page before reading on: every branch points to one of its checks, by its exact name. A closed branch is fixed before moving to the next; the order is that of causes, not symptoms.
1. Access: does the crawler get in?
What the vendor documents. OpenAI documents the effect of a refusal: "sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers". Refusing GPTBot only concerns training, and ChatGPT-User "is not used for crawling the web in an automatic fashion" (OpenAI, bots and crawlers).
What we recommend. Read three checks of the "AI crawler access" dimension: AI answer crawlers allowed, robots.txt read and compliant, and our crawler allowed on the home page. An answer crawler blocked: our French guide to AI crawlers and robots.txt (Crawlers IA et robots.txt). A correct file but a site behind a proxy: Cloudflare and AI crawlers. To prove what each crawler receives, request by request and in the logs: vérifier l'accès de ChatGPT (French). While this branch is closed, nothing else counts.
2. Extractability: does it have anything to read?
What the vendor documents. Google documents that "server-side or pre-rendering is still a great idea" because "not all bots can run JavaScript" (Google, JavaScript SEO basics). No assistant vendor documents JavaScript execution by its fetchers.
What we recommend. Read the text readable without JavaScript and the single h1 checks in the "Content extractability" dimension. A few hundred characters on a home page, or text under 5% of the HTML, means the crawler received a shell. The fix is server rendering, per stack, in le contenu sans JavaScript pour les moteurs IA (French).
3. Identity: does it know who is speaking?
What the vendor documents. Organization structured data helps Google "disambiguate your organization in search results", with the legal name, identifiers such as vatID or iso6523Code, the address and sameAs profiles (Google, Organization structured data).
What we recommend. Read the identity entity, official sameAs profiles and legal identifier checks in the "Structured data" dimension. A missing entity, or a name shared with a software product or another company with nothing to tell them apart, and the assistant cites someone else or nobody. The graph and the clause that rules out namesakes are in Schema Organization et identité d'entité (French); the same identity, as text, in the llms.txt guide (French).
4. Freshness: does the page have a date?
What the vendor documents. For an Article, Google asks for datePublished "in ISO 8601 format", recommends providing "timezone information", and dateModified for "the date and time the article was most recently modified" (Google, Article structured data). No assistant vendor documents its use of dates.
What we recommend. Read the machine-readable date check, or its opposite, no publication or update date detectable, in the "Freshness and identity" dimension, together with the Open Graph completeness check. A page without a date is a page nobody can vouch for; the date must come from a real source, the CMS update field or the template's history, never today's date. This branch has no dedicated page: the fix is in the freshness section of the AI search optimisation guide.
5. Measurement: is it really missing?
What the vendor documents. Google states that "there are no additional requirements to appear in AI Overviews or AI Mode" and that this traffic is "included in the overall search traffic in Search Console", without a breakdown (Google, AI features in Search).
What we recommend. An absence observed on two questions asked one evening is not a measurement. Before concluding, a fixed question set, asked to the same assistants, recorded the same way, over at least two runs: the method is in measuring your visibility in AI answers. Four open branches and an absence confirmed over two runs point to a content and reputation subject, not a technical one: the questions where a competitor is cited say which pages to write.
And if everything is in order
A green GEO test, a two-run measurement confirming the absence, and nothing to fix in the four branches: that is where the GEO Audit starts. It covers the six dimensions on your priority pages, not only the home page, reads your server logs over several weeks, keeps the citation measurement before and after, and ends with the fixes implemented in your code. The complete frame, dimension by dimension, remains the AI search optimisation guide.
Frequently asked
Why does ChatGPT recommend my competitor and not me? Because, on that question, the assistant found on their site a readable, identifiable, dated page that answers it, and not on yours. The measurement protocol records those questions one by one; they are the pages to write, once the first four branches are open.
My site ranks on Google, why not on ChatGPT? Because Google renders JavaScript and has known your entity for years, while an assistant starts from zero: a robots.txt that refuses its crawlers, content absent from the raw HTML or an entity without identifiers is enough to stop it where Google gets through. Branches 1 to 3 cover exactly those three cases.