Definition
GEO Score
A GEO score rates how retrievable, understandable, and citable your site is to AI answer engines. Ours runs 16 deterministic checks across five layers — here's what a GEO score should and should not measure.
A GEO score is to AI search visibility what a Lighthouse score is to page performance: a repeatable composite that turns “is this site ready” from a vibe into a checklist. The value is not the number itself but the findings underneath it — each failed check with a priority and a fix.
The five layers we measure
- Facts (35%) — dates on content, numbers with sources, citations, methods disclosed. AI engines preferentially cite verifiable claims.
- Structure (25%) — semantic headings, FAQ schema, BLUF-style summaries, tables. Retrieval systems extract from structure.
- Semantics (15%) — entity clarity, explicit definitions, consistent naming. Machines must resolve what you are before they cite you.
- AI accessibility (15%) — robots.txt AI-crawler rules, llms.txt presence, clean HTML without client-side-only content.
- Trust (10%) — authorship, about page, contact information, consistent external identity.
Weights are self-calibrated on real audit data, not copied from academic papers — published GEO weighting schemes show weak correlation with actual citation outcomes, so we tuned ours against observed retrieval behavior instead.
What a GEO score should not do
- Not promise rankings. GEO readiness is necessary, not sufficient. Being citable does not create demand for your brand.
- Not use LLM judges for the number. Model opinions drift; deterministic checks don’t.
- Not hide the checklist. A black-box score is unverifiable. Every check should be inspectable — see ours in the GEO score docs.
Related concepts
- llms.txt — one input in the AI-accessibility layer
- AI Share of Voice — the answer-side metric this site-side score pairs with
- Topical Authority — the semantic layer’s long game
FAQ
What is a GEO score?
A GEO (Generative Engine Optimization) score rates a website's readiness to be retrieved, understood, and cited by AI answer engines. It aggregates checks across content quality, structure, semantics, AI accessibility, and trust signals into a single number with actionable findings.
What does ZensInk's GEO score measure?
16 dimensions across five weighted layers: facts layer (35% — dates, numbers, sources, citations), structure (25% — headings, schema, FAQ), semantics (15% — entity clarity, definitions), AI accessibility (15% — robots.txt AI rules, llms.txt, clean HTML), and trust (10% — authorship, about, contact). All checks are deterministic — the same site always scores the same.
Why deterministic scoring instead of AI-judged scoring?
LLM-judged scores drift between runs and can't be regression-tested. Deterministic checks give the same output every time, so a score change always means a site change. AI judgment still enters — but in interpreting findings, not generating the number.
What is a good GEO score?
Above 80 is strong, 60-80 has clear gaps, below 60 has structural problems. More useful than the number: every failed check comes with a priority (P0/P1/P2) and a specific fix, because a score without remediation is trivia.
Related terms
- llms.txt — llms.txt is a proposed standard file that tells LLMs and AI crawlers what your site is about and which pages matter most.
- Query Fan-Out — Query fan-out is how AI search engines split one question into many sub-queries before assembling an answer.
- AI Share of Voice — AI share of voice measures how often your brand appears in AI-generated answers relative to competitors across a tracked prompt set.