Methodology: the public scoring rubric

Our score is not a black box. Below are all 46 checks, their weights, their confidence tiers and the public sources they rest on. Everything is versioned, so when a check or a weight changes, you can see exactly what changed and in which version.

Ruleset version: 1.0.0

Two numbers, never mixed

The AI-Readiness score (0 to 100) measures what you control: the weighted sum of categories A through E. Deterministic: the same inputs always produce the same score.

The live visibility index (“mentioned in X of N answers”) measures the outcome. We report it separately, because blending it into the score would penalize new sites for their age, not for fixable defects.

Categories and weights

Category B carries the highest weight. Content that only appears after JavaScript runs is the most common fatal defect, and the one we can prove most clearly: we set the raw HTML next to the rendered DOM and show you exactly what vanished.

CategoryWeightNature
A AI access & crawlability22deterministic
B Machine-readable content (no JS)28deterministic
C Structured data & entity clarity18deterministic
D Citability & direct answers22model-judged, against a written rubric
E Technical quality & accessibility10deterministic

All checks

A · AI access & crawlability

CheckPointsScopeConfidenceSource
A1 robots.txt reachable and parseable3siteprovendevelopers.google.com
A2 OAI-SearchBot allowed (ChatGPT search citations)3siteprovendevelopers.openai.com
A3 GPTBot policy (training exposure only)1siteprovendevelopers.openai.com
A4 ClaudeBot and Claude-SearchBot allowed2siteprovensupport.claude.com
A5 PerplexityBot allowed2siteprovendocs.perplexity.ai
A6 bingbot allowed (Copilot grounding)2siteprovenbing.com
A7 Google-Extended policy (Gemini grounding/training)1siteprovendevelopers.google.com
A8 No snippet-blocking directives on main content2pageprovendevelopers.google.com
A9 Pages reachable by AI bot user-agents3pageprovensource
A10 No noindex on audited pages1pageprovendevelopers.google.com
A11 XML sitemap present, valid and honest2siteprovensitemaps.org
A12 Canonical URL correct1pageprovendevelopers.google.com
A13 Short redirect chains1pageprovendevelopers.google.com

B · Machine-readable content (no JS)

CheckPointsScopeConfidenceSource
B1 Content parity: raw HTML vs rendered DOM8pageprovenvercel.com
B2 Substantive text in raw HTML3pageconsensusdevelopers.google.com
B3 Single H1 and logical heading hierarchy3pageprovenw3.org
B4 Semantic landmarks (main, nav, article)2pageconsensusw3.org
B5 Crawlable <a href> navigation3pageprovendevelopers.google.com
B6 Images have descriptive alt text2pageprovenw3.org
B7 Data as real tables/lists markup2pageconsensusw3.org
B8 Descriptive link text2pageprovenw3.org
B9 Sane text-to-markup ratio1pageconsensussource
B10 No content locked behind interaction2pageprovendevelopers.google.com

C · Structured data & entity clarity

CheckPointsScopeConfidenceSource
C1 JSON-LD parses without errors3pageprovendevelopers.google.com
C2 Organization/LocalBusiness schema with identity3siteprovendevelopers.google.com
C3 WebSite schema on homepage2siteprovenschema.org
C4 Page-type schema with required properties3pageprovendevelopers.google.com
C5 Structured data matches visible content2pageprovendevelopers.google.com
C6 NAP consistency (name, address, phone)3siteconsensussource
C7 html lang matches content language1pageprovenw3.org
C8 hreflang reciprocity (multilingual)1siteprovendevelopers.google.com

D · Citability & direct answers

CheckPointsScopeConfidenceSource
D1 Direct-answer block near the top4pageconsensusdevelopers.google.com
D2 Question-shaped headings3pageconsensusdevelopers.google.com
D3 Verifiable specifics over marketing prose4pagecorrelatedarxiv.org
D4 Definition blocks for key terms3pageconsensusdevelopers.google.com
D5 Stand-alone sections (RAG-friendly)3pageconsensusdevelopers.google.com
D6 Freshness signals2pageconsensusdocs.perplexity.ai
D7 E-E-A-T surface: authorship and sources3pagecorrelatedarxiv.org

Category D is judged by an AI model against a written, versioned rubric: temperature 0, with the model identity and prompt version stored with every audit for reproducibility. Each criterion earns 0–2 points plus a verbatim quote as evidence.

E · Technical quality & accessibility

CheckPointsScopeConfidenceSource
E1 HTTPS with HTTP→HTTPS redirect2siteprovendevelopers.google.com
E2 Core Web Vitals (field data)4siteprovendeveloper.chrome.com
E3 Mobile viewport meta1pageprovendevelopers.google.com
E4 Machine-readability basics (title, labels)2pageprovenw3.org
E5 No full-screen interstitial in raw HTML1pageprovendevelopers.google.com

Informational checks (0 points)

We report these for context, but they don't affect your score; because the evidence doesn't (yet) support an effect:

CheckPointsScopeConfidenceSource
A14 CCBot policy (Common Crawl, informational)0siteconsensuscommoncrawl.org
C9 Deprecated rich-result types (informational)0pageprovendevelopers.google.com
I1 llms.txt (speculative)0sitespeculativellmstxt.org
I2 IndexNow (not externally detectable)0siteconsensusindexnow.org
I3 RSS/Atom feed advertised0siteconsensussource
I4 CDN/server context (informational)0siteconsensussource

Confidence tiers

Not every “AI best practice” has the same footing in reality. Every check honestly declares what it rests on:

  • PROVEN; officially documented or directly measurable behavior (specifications, vendor documentation).
  • CORRELATED; supported by public empirical research; a correlation, not a guarantee.
  • CONSENSUS; widely accepted industry practice, without direct measurement.
  • SPECULATIVE; unproven. Worth 0 points; reported for information only.

Checks that don't apply (for example Product schema on a site with no products) are redistributed within their category; you're never penalized for what you don't have.

The methodology behind our paid work

The audit above scores what you control. When you hire us to act on it, through Implementation or Monitoring, the work is guided by SEO-Kami, the open, MIT-licensed methodology our founder maintains. It uses the same four confidence tiers you see above and the same discipline: keep what's proven apart from what's merely hyped.

We weight entity and earned-media signals

SEO-Kami's reading of the research puts earned media first: branded mentions across the web, a consistent name, address and phone, sameAs links, and verifiable authorship and E-E-A-T. It reads as the strongest lever for AI visibility, well ahead of raw backlinks, so our implementation work prioritizes those signals accordingly.

We optimize for a zero-click reality

Most searches now end without a click, and AI-Overview queries almost always do. So we treat impressions and share of AI citations as the real outcome, not just referral traffic. It is the same reason your live visibility index above is reported separately from your score.

The numbers, as context, not promises

These are CORRELATED-tier findings from public research: a correlation, not a guarantee for any one site. Branded web mentions correlate roughly 0.664 with AI-Overview visibility, versus about 0.218 for backlinks. An estimated 82% of AI citations come from earned media rather than a brand's own pages, and around 83% of AI-Overview queries end with no click. We use these numbers to prioritize the work, and we tell you honestly which fixes are proven and which are speculative.

SEO-Kami is open source under the MIT license. View SEO-Kami on GitHub

Methodology: the public scoring rubric · AI-SEO-Radar