A Generative Engine Optimization checker is a tool that audits how ready a website is to be read and cited by AI answer engines — ChatGPT, Perplexity, Google AI Overviews. It works like an SEO checker, but for the signals language models rely on: crawler access, machine-readable structure, quotable content and attribution. Here's what a good one measures, and how to pass.

Worked example: the score is a weighted diagnostic

Illustrative calculation, not a scan of a real website: suppose a page earns 80% of available weighted GEO check points. Informational checks are excluded. WebAnalyzer applies round(100 × ratio^1.7), so 0.8 becomes 68. Short server-delivered pages are not failed just for having fewer than 50 words. Search permissions are checked for the actual URL; training and user-triggered permissions are listed separately. Other dimensions contribute independently to the overall weighted score.

Changing the server-delivered answer can improve observable content while leaving other issues. Compare the same URL before and after, and read the evidence for each failed check. A higher score demonstrates a change in these diagnostics, not an increase in citations.

Weighted ratio: 0.80
Curved score: round(100 × 0.80^1.7) = 68
Informational formatting observations: no score penalty

Worked example: allow search while controlling training

This illustrative robots.txt snippet allows OpenAI search crawling while declining GPTBot training crawling. It is a policy example, not a blanket recommendation to allow every bot. Check existing rules for your specific paths and test the resulting policy. ChatGPT-User represents user-requested visits and does not work identically to the search crawler.

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Group 1: AI crawler access

Start by checking which crawlers may access the submitted page. Search, user-requested retrieval and model training are separate purposes. OAI-SearchBot controls OpenAI search access; GPTBot is a training control. Google Search uses Googlebot, while Google-Extended controls some Gemini training and grounding uses. A block can affect one provider without preventing access by every other provider.

Decide which uses you want to allow, then check robots.txt, CDN restrictions and server logs. An llms.txt content map is optional; Google does not require it for Search AI features. A blocked search provider is reported separately; a training policy choice does not cap the GEO score.

Group 2: content extractability

A checker inspects the supplied HTML for text, semantic landmarks and heading structure. Compare the initial response with the rendered page when content is created by JavaScript. Google does not require a particular chunk size or paragraph length for AI inclusion.

To pass: server-render your main copy, use logical headings and readable sections. Tables and lists can help readers when the content calls for them. A fixed word count or statistics quota is not a requirement for AI citations.

Group 3: answer-shaped, quotable content

This is where GEO differs most from SEO. The checker looks for question-shaped headings, direct answers in the first sentence of each section, quotable statistics, FAQ-style coverage, comparison content, and openings that name their subject instead of starting with "it" or "we".

Make the answer useful on its own: identify its subject, state the conditions and cite evidence. These checks are editorial heuristics, not evidence that a model has read or cited your page. Measure actual citations and qualified referrals separately.

Group 4: trust and attribution

Engines cite sources they can safely attribute. Checkers score author and entity information, an about/organization page, machine-readable dates (datePublished, dateModified), outbound citations to credible sources, content freshness, and JSON-LD whose entity actually matches the page.

Use accurate authors, dates and sources, and make structured data match visible content. Update the date only after a substantive revision. Google requires no special AI schema; its ordinary Search eligibility rules still apply.

Run one now (free)

Use the GEO checker to inspect the wider scoring criteria, or the AI readiness checker for the access and HTML checklist. The initial result is the overall eight-area score. Email unlocks the GEO breakdown and full report. Three successful analyses per network are free each UTC day; a failed analysis does not use the allowance.

How a GEO Score Is Calculated: Checks, Evidence and Limits