How We Score Your Site
GEO Audit scores your site out of 100 by running 14 checks across three categories: technical (35%), content (35%), and authority (30%). Each check is scored 0–10 based on how well your site meets the criteria AI search engines use to discover, understand, and cite content. Sites scoring 80+ are rated Excellent; below 40 is Not GEO-ready.
Every check is grounded in published research — including the Princeton/Georgia Tech GEO study, OpenAI and Anthropic crawler documentation, and Zyppy's AI citation analysis. Below is the full breakdown of what we check, why it matters, and how each score is calculated.
Scoring Breakdown
7 checks — crawlability, structured data, and infrastructure signals
4 checks — depth, structure, accessibility, and answer quality
3 checks — entity identity, discoverability, and trust
Each check scores 0–10. Scores are weighted within their category, then categories combine to produce a final score out of 100. Grades: Excellent (80+), Good (60–79), Needs Work (40–59), Not GEO-ready (<40).
Technical
7 checks — 35% of your total score
robots.txt AI Rules
What we check: Whether your robots.txt explicitly addresses AI crawlers like GPTBot, ClaudeBot, and PerplexityBot.
Why it matters: AI search engines respect robots.txt directives. If you don't mention AI crawlers, you're leaving access to default behaviour — which may change. Explicitly allowing crawlers signals intent and ensures your content is indexed for AI-generated answers. OpenAI's GPTBot and Anthropic's ClaudeBot both document robots.txt compliance.
How we score: 0 if no robots.txt exists. 3 if it exists but doesn't mention AI bots. 4 if it blocks AI crawlers. 5–10 based on the ratio of AI crawlers explicitly allowed.
llms.txt
What we check: Whether your site serves an llms.txt file at the root domain.
Why it matters: llms.txt is an emerging standard that gives AI systems a structured summary of your site — what it does, what content matters, and where to find it. It's the equivalent of a README for language models, helping them understand your site without crawling every page. The specification is defined at llmstxt.org.
How we score: 0 if not found. Base score of 5 for existence, +1 for title, +1 for description, up to +3 for linked pages. Maximum 10.
Schema Coverage
What we check: The percentage of pages with structured data (JSON-LD) and the diversity of schema types used.
Why it matters: Structured data is how machines understand your content. The Princeton/Georgia Tech GEO study found that adding structured markup significantly increases citation likelihood in AI-generated responses. Google's Structured Data documentation confirms that rich results depend on proper schema implementation.
How we score: Coverage ratio drives 0–7 of the score. Schema type diversity adds up to 3 bonus points. Maximum 10.
Sitemap
What we check: Whether your site has a discoverable sitemap referenced in robots.txt.
Why it matters: Sitemaps help AI crawlers discover your full content inventory efficiently. OpenAI's GPTBot documentation references sitemap discovery as a primary mechanism for finding pages to index. Without a sitemap, crawlers may miss important content.
How we score: 0 if no sitemap found. 5 if sitemap exists but has no URLs. 8 if sitemap exists with URLs referenced in robots.txt.
Meta Descriptions
What we check: The proportion of pages with substantive meta descriptions (50+ characters) and titles (10+ characters).
Why it matters: Meta descriptions are one of the first signals AI retrieval systems use when deciding which content to cite. In RAG (Retrieval-Augmented Generation) pipelines, the description often serves as the snippet that determines relevance scoring before the full page is ever read.
How we score: Description coverage contributes 60% of the score, title coverage 40%. Pages with missing or very short descriptions/titles reduce the score proportionally.
HTTPS
What we check: Whether your site enforces HTTPS.
Why it matters: HTTPS has been a confirmed Google ranking factor since 2014, and AI search engines inherit these trust signals. A site without HTTPS is penalised in both traditional and AI search. It's a baseline infrastructure requirement.
How we score: 3 if not HTTPS. 7 if HTTPS is enforced.
OG & Twitter Completeness
What we check: The completeness of Open Graph (og:title, og:description, og:image, og:url) and Twitter Card tags across all pages.
Why it matters: Open Graph and Twitter Card metadata are used by AI systems as structured signals about page content. They provide clean, pre-formatted summaries that retrieval systems can use without parsing the full page. The Open Graph protocol (ogp.me) and Twitter Card documentation define these standards.
How we score: Average completeness across all pages (5 fields checked per page). 10 means every page has all OG + Twitter tags.
Content
4 checks — 35% of your total score
Content Depth
What we check: Median word count across pages and the ratio of thin pages (under 100 words).
Why it matters: The Princeton/Georgia Tech GEO study (Aggarwal et al., 2023) found that content depth is one of the strongest predictors of citation in AI-generated responses. Thin, shallow pages are unlikely to be cited because they don't provide enough substance for AI systems to extract and reference.
How we score: Median word count sets the base: 800+ words = 9, 500+ = 7, 300+ = 5, 100+ = 3, below = 1. Thin page ratio penalises up to 4 points.
Heading Quality
What we check: Whether each page has exactly one H1 and uses a proper heading hierarchy (no skipped levels).
Why it matters: AI systems parse heading structure to understand content hierarchy and extract key topics. The Princeton GEO study identifies heading optimisation as one of nine strategies that improve visibility in generative engines. Skipped heading levels (H1 to H3) confuse both AI parsers and screen readers.
How we score: H1 correctness contributes 60% and heading hierarchy contributes 40%. Each is measured as a ratio across all pages.
Image Accessibility
What we check: Alt text coverage and quality across all images on crawled pages.
Why it matters: Images without alt text are invisible to AI systems that process pages as text. WCAG 2.1 requires alt text for accessibility, and the same principle applies to AI indexing — if an image isn't described, its content is lost. Weak alt text (e.g. "image1.jpg") is penalised because it provides no semantic value.
How we score: Ratio of images with good alt text, with a 50% penalty applied for weak alt text. 7 if no images exist (not penalised for image-free pages).
Answer-First Content
What we check: Whether content pages (300+ words) lead with substantive answers rather than preamble.
Why it matters: AI search engines extract answers from the beginning of content. Research by Zyppy/Cyrus Shepard on AI citations shows that pages leading with direct answers are significantly more likely to be cited. The Princeton GEO study confirms that content structure — particularly putting key information first — is a major factor in AI visibility.
How we score: Ratio of content pages that lead with substantive answers, scaled 0–10. Only pages with 300+ words are evaluated.
Authority
3 checks — 30% of your total score
Entity Clarity
What we check: Whether your site has Organization schema, sameAs links to external profiles, and consistent author information.
Why it matters: AI systems build knowledge graphs to understand who is behind content. Google's Knowledge Graph uses Organization schema and sameAs properties to connect your site to a verified entity. Without these signals, AI systems can't confidently attribute your content or assess its authority.
How we score: Organization schema (+3), sameAs links (+1–3 based on count), author info (+2), consistent author across pages (+2). Maximum 10.
Agent Discoverability
What we check: The presence of discovery mechanisms — RSS feeds, sitemaps, OpenAPI specs, and llms.txt.
Why it matters: AI agents need multiple ways to discover and navigate your content. RSS feeds enable real-time content monitoring, sitemaps provide a full page inventory, OpenAPI specs expose programmatic interfaces, and llms.txt provides a curated content guide. Each mechanism serves a different type of AI system.
How we score: RSS feed (+3), sitemap (+3), OpenAPI spec (+2), llms.txt (+2). Maximum 10.
Trust Signals
What we check: Consistent author identity, links to known platforms (LinkedIn, GitHub, Twitter/X), and schema type diversity.
Why it matters: Google's E-E-A-T framework (Experience, Expertise, Authoritativeness, Trustworthiness) influences how AI systems weight content. Consistent author identity across pages, verifiable links to established platforms, and diverse schema markup all contribute to a stronger trust profile that AI systems use when deciding what to cite.
How we score: Consistent author identity (+1–4), links to known platforms (+1–3), schema type diversity (+1–3). Maximum 10.
Sources & Research
- GEO: Generative Engine Optimization — Aggarwal et al. (2023), Princeton University & Georgia Tech. The foundational study on optimising content for AI-generated responses.
- Google Structured Data Documentation — Guidelines for implementing schema markup that Google (and AI systems) can parse.
- OpenAI Crawlers Documentation — How GPTBot and OAI-SearchBot discover and crawl web content, including robots.txt and sitemap compliance.
- Anthropic ClaudeBot Documentation — ClaudeBot's crawling behaviour, robots.txt compliance, and how to manage access.
- llmstxt.org Specification — The emerging standard for providing AI-readable site summaries.
- W3C WCAG 2.1 — Web Content Accessibility Guidelines, including image alt text requirements.
- Open Graph Protocol — The standard for structured social and sharing metadata.
- Twitter/X Card Markup — How to implement Twitter Card meta tags for rich previews.
- Zyppy — How AI Overviews Shift Traffic — Cyrus Shepard's research on how AI overviews impact publisher traffic and citation patterns.