The GEO audit: why AI doesn’t cite you
More than forty checks on the client’s site, from robots.txt to authority, with an action plan ranked by impact.
The audit answers a simple question: what’s stopping AI from citing your client? It doesn’t guess — it checks, page by page, what answer engines need in order to reuse content.
It’s the first thing to run for a new client. If AI crawlers can’t read the site, no visibility measurement will mean anything.
What the audit checks
Technical — can crawlers read you?
robots.txt is checked crawler by crawler: OAI-SearchBot (ChatGPT), GPTBot, PerplexityBot, ClaudeBot, Google-Extended (Gemini and AI Overviews) and Bingbot. Many sites block one without knowing it, often through an anti-scraping setting or a firewall.
Plus server-side rendering (text that only exists after JavaScript runs is invisible to most crawlers), JSON-LD structured data suited to the page type, the sitemap, pages with no internal links and — if you connect Bing Webmaster Tools — actual Bing indexing, which counts twice because ChatGPT relies on Bing’s index.
In 2026, about 28% of sites serve an llms.txt, but fewer than 3% of those files get a single crawler request (Ahrefs). The audit flags it — it takes five minutes and does no harm — but it carries little weight and should never come before a content or authority fix.
Content — are your pages citable?
An accessible page can still be unusable: no direct answer up front, no sourced data, headings that don’t ask questions, no table, no FAQ, a missing or old update date. The audit scores these signals on a sample of pages and names the ones with problems.
Mentions — what does AI actually answer?
If a measurement has run: the share of answers citing your client (core business first), who AI recommends instead, the tone it uses, and the queries where Google already answers with an AI Overview. Without a measurement, the audit says “unknown” — never “zero”.
Authority — who cites you elsewhere?
The link profile, the platforms AI cites on your topics that don’t link to you (Wikipedia, Reddit, YouTube, directories…), the domains that list your competitors but not you, and the services the site claims without a page of their own.
How many pages are analyzed?
About 10% of the site, with a minimum of 8 and a maximum of 40 pages, spread across the site’s sections so a large folder (your case studies, say) doesn’t drown out your service pages.
Reading the report
One sentence: what AI answers today, and the readiness score. A decision-maker can stop there.
A strategic reading: what’s really holding you back, what order to act in, and the content to publish first.
Engine by engine and prompt by prompt: who cites you, in what position, and who’s cited instead of you.
The issues to address, ranked by impact then effort. The answer to “what do I do Monday morning?”.
Each issue with its finding and its fix, then the checks passed — proof the analysis was thorough.
A site at 90 for technical and 0 for mentions doesn’t have a technical problem: it has a content and authority problem. Follow the strategic reading rather than fixing what’s already fine.
From report to action plan
Every issue to fix becomes a task in the Action plan, with its finding and its fix. Services without a page become articles to write; missing platforms become outreach to do.