Free ToolsGuidesBlogMethodology AI Visibility Checker Check my site
Methodology v4 · Transparent by design

GEO score methodology: how every checkpoint is scored

The GEO score measures, from 0 to 100, how ready a page is to be found, understood and cited by AI search. This page documents all 22 checkpoints, their formulas and their data sources, so every number in a report can be checked.

The formula

GEO score = 30% Content Quality + 25% AI Readiness + 20% Technical Basis + 25% Authority & Trust. Each category is the average of its measured checkpoints (0 to 100). Checkpoints that do not apply are excluded, never counted as zero.

Last updated 15 October 2026

Content Quality

30%
ReadabilityFlesch Reading Ease for English (score 100 at 60+, falling to 40 at 30); average sentence length for other languages. Minus up to 15 points when more than 25% of sentences exceed 25 words.
Measured
ComprehensivenessMain-content word count (55%), H2/H3 subtopic sections (15%) and the share of expected subtopics covered (30%). Without AI: word count 80%, sections 20%.
Measured + AI
Transition WordsShare of sentences containing a transition word or phrase, against a 30% target. English only.
Measured
LSI KeywordsAI lists 20 concepts a complete page on the topic should cover; we check which ones actually appear in your text. Score = coverage %.
AI + measured
Answer CompletenessAI lists the six most common questions about the topic and judges whether the page answers each (yes = 1, partial = 0.5).
AI-assessed
Factual DensityNumbers, dates, prices, percentages and measurements per 100 words; 4 or more per 100 words scores 100.
Measured
CitabilitySelf-contained paragraphs of 18 to 80 words that state a fact or definition (11 points each) plus lists and tables (up to 20).
Measured

AI Readiness

25%
AI OptimizationChecklist: valid llms.txt (15), direct answer in the first paragraph (20), question-style headings (20), FAQ section or schema (15), summary section (10), lists or tables (10), snippets allowed (10).
Measured
Content StructureOne H1 (20), H2 sections (20), no skipped heading levels (15), lists (15), short paragraphs (15), tables (5), a heading at least every 350 words (10).
Measured
Image Alt-TextShare of content images with a descriptive alt attribute. Decorative and tracking images are excluded; no images means not applicable.
Measured
Platform SpecificityFor Google, Bing/Copilot, ChatGPT search, Perplexity, Claude and Apple: crawler allowed for this URL (60), snippets allowed (20), main text in the server HTML (20).
Measured
Content Type MatchingAI identifies the page type and the schema types it should declare; we compare them with the JSON-LD actually on the page.
AI + measured
Semantic ClarityTitle and H1 agreement (30), meta description length (25), language declared (15), sentence length (30); averaged with an AI clarity rating when available.
Measured + AI

Technical Basis

20%
AI Bot AccessYour robots.txt evaluated per RFC 9309 for this exact URL for 10 AI search and answer bots (OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, Googlebot, Bingbot, Applebot, DuckAssistBot). Training bots are reported, not scored.
Measured
Schema ValidatorEach JSON-LD object is parsed and checked against required (70%) and recommended (30%) properties for its type; invalid JSON scores 0; missing Organization or Person entity costs 8 points.
Measured
Sitemap & Discoverabilityrobots.txt (15), valid XML sitemap (25), sitemap declared in robots.txt (10), this URL in the sitemap (15), canonical tag (15) pointing to itself (10), indexable (10).
Measured
Page SpeedGoogle Lighthouse mobile performance score from the PageSpeed Insights API, with LCP, CLS, TBT and real-user (CrUX) data when available. If Google is unreachable, an estimate from server response time and page weight is shown and clearly labelled.
Google PageSpeed
Content FreshnessNewest valid date from schema dateModified/datePublished, article meta tags, <time> elements, sitemap lastmod and the Last-Modified header: 30 days = 100, 90 = 88, 180 = 72, 365 = 55, 730 = 35, older = 20, none = 15.
Measured

Authority & Trust

25%
Citations & SourcesExternal domains cited in the main content (up to 50), authoritative sources such as .gov, .edu, Wikipedia and official documentation (up to 35), and a references section (15).
Measured
E-E-A-T SignalsChecklist: author or byline (15), About page (12), contact details (12), privacy policy (8), HTTPS (8), visible date (10), Organization schema with logo (10), reviews or testimonials (10), credentials or experience (10), external sources (5).
Measured
Knowledge GraphOrganization or Person schema (30), name, url and logo (15), sameAs links (up to 20) including Wikipedia, Wikidata, LinkedIn, Crunchbase or GitHub (10), and a live Wikidata lookup: entity listing your site as its official website (25) or name match only (10).
Measured
Topical AuthoritySitemap size (25), other URLs on the same topic (25), contextual internal links (20) and independent websites mentioning your brand in a live web search (30).
Measured + search

How the live AI visibility test works

The AI visibility test is reported separately from the GEO score because it measures outcomes, not readiness. It follows the same retrieve-then-cite process AI search engines use:

  1. A GLM model writes realistic customer questions for your category: unbranded questions (discovery) and branded ones (reputation).
  2. Each question is searched on the live web with Z.ai Web Search.
  3. A GLM model answers each question from those numbered results and cites the ones it relies on as [n].
  4. We record whether your site was retrieved, whether it was cited and at what position, whether your brand was named, and which domains were cited instead.

AI visibility score = 50% unbranded citation rate + 20% unbranded mention rate + 10% unbranded retrieval rate + 20% branded citation rate. Results are cached for 24 hours. ChatGPT, Gemini and Perplexity are not queried, so no figures are shown for them.

Score bands

ScoreLabelMeaning
90–100ExcellentStrong across every measured checkpoint.
80–89StrongWell prepared; usually one weaker area.
70–79GoodSolid base; gains from clearer answers and authority.
50–69Needs improvementClear gaps that make other sources more likely to be cited.
0–49WeakFundamental issues such as blocked crawlers, thin content or no structured data.

What the score does not do

  • It does not reveal private rankings inside ChatGPT, Gemini or any AI product; no tool can.
  • It is not a Google ranking factor.
  • It analyses the URL you submit plus site-level files; check key pages individually.

See the definitions in the GEO glossary or run a free check.

Methodology questions

How is the overall GEO score calculated?

Each category score is the average of its measured checkpoints. The GEO score is the weighted average of the four categories: Content Quality 30%, AI Readiness 25%, Technical Basis 20% and Authority & Trust 25%. Checkpoints that do not apply, such as image alt text on a page without images, are excluded rather than scored as zero.

Which AI model is used?

AI-assessed checkpoints and the live visibility test use Z.ai GLM models with live web search. The AI only supplies topic expectations and judgements; scores are calculated by our code and every row shows its source.

Why do some checkpoints show N/A?

A checkpoint shows N/A when it cannot be measured fairly, for example readability on a page with fewer than five sentences, or AI-assessed checks when the AI service is unavailable. N/A items are left out of the averages.

Is the GEO score a Google ranking factor?

No. Google does not use third-party scores. The checkpoints reflect signals that search and AI systems are known to rely on, such as crawl access, structured data, clear answers and credible sources, so improving them supports both SEO and AI visibility.