How the Slurp Score works

Updated July 2026 - engine version 0.1

The Slurp Score is a 0-100 measure of how ready a page is to be found, read and cited by AI answer engines: ChatGPT, Google AI Overviews and AI Mode, Perplexity, and Claude. It is built from about 60 deterministic checks. Every check states why it matters, how to fix it, and links the evidence behind it. This page is the complete method - there is nothing else in the score.

Three tiers of checks

Gates. Pass or fail conditions that cap the score at 49 for the affected engine. Example: robots.txt blocks PerplexityBot. A page that no engine can fetch cannot be cited, so no amount of on-page polish should produce a high score. Gates are why a GetSlurp 90 means something.

Graded checks. Weighted checks across the 11 categories below. Pass, warn or fail, each contributing to the category score. Checks that do not apply to a page type (a homepage is not an article) are excluded rather than counted against it.

Informational. Reported, never scored. llms.txt, WebMCP readiness, and training-bot blocks live here. If the evidence says a factor does not affect AI visibility, it gets zero weight - visibly.

Per-engine scores

ChatGPT, Google, Perplexity and Claude run different crawlers against different indexes, and published citation studies show their source pools barely overlap. So GetSlurp reports one score per engine. Today they diverge when engine-specific gates fire; engine-specific weighting (Perplexity's stronger recency bias, for example) is being calibrated and will widen the differences.

The 12 categories

Category	Weight	What's checked	Key evidence
AI crawler access	15%	robots.txt rules for every AI bot (search, training and live-fetch tiers), noindex, snippet controls, sitemap, HTTP behavior	Official bot documentation from OpenAI, Anthropic, Perplexity and Google
Renderability	15%	Content present in raw HTML, app-shell detection, framework SSR verdict, text ratio, DOM size	Vercel and MERJ crawler study: no major AI crawler executes JavaScript
Answer-readiness	15%	Lead paragraphs, question headings, section length, lists and tables, summaries, definitions, content position	Citation-position studies; passage-level retrieval in Google AI Mode
Citability and evidence	12%	Statistics, quotations, outbound citations, claim-evidence pairing, keyword stuffing, readability	Princeton GEO paper (KDD 2024): measured lifts of +31.5% (statistics), +40.7% (quotations), +29.1% (citing sources)
Entity and E-E-A-T	12%	Author markup, Organization schema, sameAs profiles, trust pages, brand naming consistency	E-E-A-T visibility studies; earned-media bias findings in AI search
Freshness	8%	Machine-readable dates, recency, year signals, stale-signal contradictions	Roughly half of AI-cited content is under 13 weeks old
Structured data	8%	Valid JSON-LD, content-specific types, schema-to-page consistency, FAQ, breadcrumbs	Bing confirms schema use; evidence for direct citation impact is mixed, and we say so
Semantic HTML	6%	Single H1, heading order, landmarks, lang, link text quality	HTML specification; machine legibility for parsers and browsing agents
Meta and discoverability	5%	Title, description, canonical correctness, Open Graph, URL clarity	Wrong canonicals silently redirect citation credit
Internal linking	4%	Internal link count, breadcrumbs, anchor quality	Crawl discovery mechanics
Multimodal	3%	Alt text coverage, video transcripts, media schema	Text alternatives are the only path into text-based retrieval
AI agent readiness	info	WebMCP signals, MCP references, form operability for browsing agents	W3C WebMCP effort; ChatGPT-User and Claude-User already operate pages live

What we refuse to score

llms.txt. No engine honors it. Google declined it on the record, and an Ahrefs study of 137,000 sites found 97% of llms.txt files receive zero bot traffic. We detect it and report it, at zero weight.
WebMCP and agent standards: tracked, not yet scored. AI agents that operate sites are the next wave, and WebMCP is the standard to watch. GetSlurp already checks your agent readiness and reports it. The weight goes above zero the day engines ship support, and early adopters will already be passing.
Training-bot blocks. Blocking GPTBot or Google-Extended is a content-rights choice that does not remove you from AI answers. We report it neutrally and never penalize it.

Honest limits

Two things the research is clear about. First, content-tweak tactics are contested: the C-SEO Bench study (NeurIPS 2025) found their benefits shrink as adoption grows, which is why citability carries 12% and not 40%. Second, an on-page score measures eligibility and extractability, not guaranteed citations - the strongest predictors of actually being cited are off-site: brand search volume, earned coverage, and freshness across your whole footprint. Any tool that promises otherwise is selling something.

Sources for every claim on this page: the Princeton GEO paper (arXiv:2311.09735), C-SEO Bench (arXiv:2506.11097), the Vercel AI crawler study (vercel.com), the Ahrefs llms.txt study (ahrefs.com), and the official crawler documentation of OpenAI, Perplexity and Google. Each individual check carries its own source link in the report.