Why run a GEO audit?

How it works

Why this matters

Search stopped being ten blue links. When someone asks ChatGPT which agency to hire, or asks Perplexity what a service costs, an answer engine picks a handful of sources, synthesizes them, and cites two or three. Everyone else is invisible — not ranked tenth, just absent.

Generative Engine Optimization is the work of becoming one of the sources that gets picked. It overlaps with SEO but it is not the same job. A page can rank perfectly well on Google and still never be quoted by an LLM, usually because the content is not structured into passages a machine can lift, or because the crawler that feeds that model was blocked in robots.txt two years ago and nobody noticed.

This audit measures the things that actually decide it. We fetch your homepage and a spread of interior pages from your sitemap, then check crawler access, structured data, content extractability, technical foundations and trust signals. Every finding comes from your live HTML — nothing is estimated, modeled or guessed.

What the GEO audit actually measures

The score is weighted across five categories, because they are not equally important. A blocked crawler is fatal; a missing breadcrumb is a rounding error.

• AI crawler access (22%) — whether GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and ten other agents can physically fetch your pages. This is a hard gate. If a crawler is blocked, nothing downstream matters for that engine.
• Structured data and entity clarity (20%) — whether machines can tell what your business is, what you sell and which profiles belong to you. Organization schema, sameAs links, FAQPage, Article, breadcrumbs.
• Content extractability (20%) — whether your content is shaped into passages an answer engine can lift. Word depth, question-formatted headings, heading hierarchy, and whether your copy exists in the raw HTML at all rather than being painted in by JavaScript.
• Technical foundations (16%) — sitemap, canonicals, HTTPS, server response time, HTML weight, language declaration.
• Trust and authority signals (22%) — the E-E-A-T evidence AI systems weigh before quoting a source. Named authors, visible dates, NAP data, outbound citations, reviews, a real About page.

Why AI crawler access is the first thing to check

This is the single most common own-goal we find. A site publishes a genuinely useful page, ranks for it, and then discovers it never appears in ChatGPT answers. The cause is almost always a robots.txt line added during a “block the AI scrapers” panic, or a security plugin that blocks unfamiliar user agents by default.

It is a legitimate choice to block them — some publishers have good reasons. But it should be a decision you made on purpose, with the trade-off understood, not something that happened to you. Blocking GPTBot does not stop your content being used to train models that already scraped it; it stops you being cited today.

Extractability: the part most sites get wrong

Answer engines do not retrieve pages, they retrieve passages. The unit of citation is a paragraph that answers a specific question, not a 2,000-word article that eventually gets there.

That has a concrete implication for how you write. A subheading phrased as the literal question a customer asks — “How much does local SEO cost in Sarasota?” — followed immediately by a direct answer in the first two sentences, is enormously more likely to be retrieved than the same information buried in paragraph nine under a heading that says “Pricing”.

The audit counts how many of your subheadings are phrased as real questions, and flags pages where the ratio is too low. It also catches the harder version of this problem: pages whose HTML is large but whose text content is nearly empty, which means the copy is rendered client-side. Google renders JavaScript eventually. Most AI crawlers do not render at all — to them, those pages look blank.

Trust signals AI engines weigh

Language models are trained to be cautious about sourcing. Anonymous, undated content from a site with no visible business behind it gets discounted, regardless of how good the writing is.

The fixes here are unglamorous and cheap: put a named author with a real bio on every article, show a visible last-updated date and mirror it in schema, publish a proper About page with actual people on it, put your full name, address and phone in the footer sitewide, and link out to the sources you reference. None of that requires a redesign. All of it moves the needle.

Best practices

• Fix crawler access before anything else — it is a gate, not a score.
• Add Organization or LocalBusiness schema sitewide, with a full sameAs array.
• Rewrite roughly one in four subheadings as the literal question a customer would ask.
• Answer each question in the first two sentences underneath it, then elaborate.
• Put a named author and a visible date on every piece of content.
• Make sure your main content is in the initial HTML response, not painted in by JavaScript.
• Re-run the audit after each batch of fixes — the score is reproducible, so progress is real.

Common mistakes to avoid

• Blocking AI crawlers by accident and never checking.
• Treating GEO as a separate channel from SEO. It is the same content, structured better.
• Publishing thin pages on many keywords instead of deep pages on the questions buyers ask.
• Adding schema that describes content not visible on the page — that is a policy violation.
• Assuming an llms.txt file is the whole job. It is the smallest part of it.

Frequently asked questions

More tools like this

See all 11 free tools →