Methodology Changelog

v4.6: Data-Driven Re-Weight

v4.6 is a weight-only update. The 48-parameter catalog is unchanged — nothing was added, removed, renamed, or rescored — but the weight each parameter carries in the composite AI-Readiness Score was recalibrated using a study of production scan data.

Source for parameter count: content/methodology/version.json (parameterCount: 48).

Why we re-weighted

In June 2026 we analysed 1,448 full reports to ask a simple question: which of our AI-readiness parameters actually line up with whether AI assistants surface a site? Two findings drove this update:

• Being built well for AI is necessary but not sufficient. In our observational sample, AI visibility is dominated by brand authority — a well-built site is easier to surface, but readiness alone does not decide it. About a quarter of sites are effectively invisible to AI assistants regardless of their readiness score.
• The signals that correlate most with visibility are cite-worthy, answer-first content backed by a machine-actionable API — not protocol files. Some of the most heavily weighted "AI-Specific Protocol" signals, including llms.txt and AI manifests, showed no measurable association with visibility in our sample.

These are correlational findings from observational data, not cause-and-effect. v4.6 recalibrates the weights so the score points at the signals that track with visibility; it does not claim that changing any single parameter will move where you are surfaced.

What changed

Weighted up — the signals that correlate with being surfaced and cited:

• Actionable API for agents (action-api) — the largest increase.
• Citation & evidence density (citation-evidence-density) — cite-worthy, sourced content.
• Entity consistency (entity-consistency) — a coherent, consistent identity across the site.
• Answer-first content signals: answer structure, answer-first compliance, TL;DR sections, and core reachability/discovery (HTTP reachability, sitemap availability).

Weighted down — signals that carried large weights but showed little association with visibility:

• llms.txt presence and AI manifests coverage — treat these as low-cost hygiene, not a visibility lever. They remain in the catalog and still score; they simply no longer carry outsized weight.
• Capability metadata, API schema quality, FAQ schema, content freshness, and author authority — reduced toward their measured leverage.

Three non-linear parameters (ai-crawler-access-control, semantic-html, ucp-manifest) and the five newest parameters (ids 49–53) kept their prior weights, pending more data.

Score compatibility

Because only the weights changed, the shift is small. Most sites see their score move by a few points on their next scan — up or down depending on where their strengths sit relative to the new weighting. The total weight budget is preserved, so scores stay on the same 0–100 scale and remain comparable across sites.

Historical reports are never re-scored. A report you saw under v4.5 keeps its original score and label; the score-history chart marks the v4.6 boundary. The public leaderboard shifts gradually as sites re-scan under the new weights.

Phase 5: New Scored Parameters + Engine Cluster Sub-Scores (v4.5)

v4.5 adds 5 new scored parameters, bringing the total to 48 active parameters (47 scored + 1 informational).

Source for parameter count: content/methodology/version.json (parameterCount: 48).

What changed

Added (5 new scored parameters):

• IndexNow Adoption (indexnow-adoption, id 49, Discovery) — detects whether the site has adopted the IndexNow push protocol (Bing/Yandex/Naver/Seznam/Yep/Amazon). Scores 100 when a valid key file is accessible at the root, 50 when a robots.txt or header hint points to IndexNow but the key file is unconfirmed, and 0 when no signal turns up. Google does not participate in IndexNow.

• Core Web Vitals / Page Experience (page-experience, id 50, Technical SEO) — measures LCP, CLS, and INP from Chrome User Experience Report (CrUX) p75 field data. Scores 100 when all three vitals are Good, 50 when at least one is Needs Improvement, and 0 when at least one is Poor. When CrUX holds no field data for the URL, it falls back to an advisory UNKNOWN and drops out of the denominator.

• Search-Bot Network Reachability (bot-network-reachability, id 51, Crawlability) — checks whether declared AI search-bot user-agents (OAI-SearchBot, Claude-SearchBot, PerplexityBot) are reachable at the WAF/CDN layer, independent of robots.txt settings. Scores 100 when all bots are reachable, 50 when reachability is mixed, and 0 when every bot is confirmed blocked. When all probes come back inconclusive, it falls back to an advisory UNKNOWN. A confirmed network block can cap the composite score for tier-A bots.

• Entity Grounding (entity-grounding, id 52, Entity & Schema) — detects whether the page's JSON-LD Organization or Person schema includes sameAs links to authoritative entity graphs (Wikipedia, Wikidata, LinkedIn). Scores 100 when two or more authoritative targets are present, 50 when exactly one is, and 0 when sameAs is missing or none of its targets are authoritative.

• Multimedia Coverage (multimedia-coverage, id 53, Content Structure) — measures image and video presence, alt-text completeness, and media-to-content alignment. Scores 100 when relevant media is present with full alt coverage and contextual alignment, 50 when media is present but its alt text is incomplete or it sits out of context, and 0 when no relevant media appears in the main content area.

Per-engine cluster sub-scores (display): The report UI now surfaces per-engine cluster weighted sub-scores (Bing, Google, Perplexity, Claude) alongside the composite AI-Readiness Score, derived from each parameter's engineRelevance.clusters field in the report-service catalog.

Crawler-tier correction: The bot-network-reachability parameter introduces a score-cap mechanism that mirrors the existing robots.txt Disallow cap for tier-A search bots — a confirmed WAF/CDN block is treated equivalently to a robots.txt disallow in the composite cap calculation.

Advisory checks: Two of the five new parameters (page-experience and bot-network-reachability) can return an UNKNOWN advisory outcome, which drops the parameter from the score denominator when field data or probe results are inconclusive. That way, sites are not penalised for missing CrUX coverage or ambiguous network probes.

Score compatibility

Because v4.5 adds 5 new scored parameters, scores computed under it are not directly comparable to scores from v4.4 or earlier. The score history chart marks the v4.5 boundary, and old reports keep their original scores.

Expect a small score adjustment on your next scan. The new criteria — IndexNow adoption, entity grounding, and multimedia coverage — are met by only a minority of sites today, so most sites see a small decrease of roughly 1–2 points the first time they re-scan under v4.5 (larger, up to about 5 points, for sites that previously scored very high). A minority are unaffected or improve. The shift reflects genuine coverage gaps that the new, stricter parameters surface — a recalibration of the methodology, not a change to your website. Historical reports are never re-scored.

Phase 4: UCP Manifest + Deprecated Check Removal (v4.4)

v4.4 is the current methodology. It runs 43 active parameters (42 scored + 1 informational) and keeps 5 parameters in deprecated status so older citations still resolve.

Source for parameter count: content/methodology/version.json (parameterCount: 43).

What changed

Removed (deprecated, no longer scored): we moved 5 API-specific checks to deprecated status after flagging them as low-signal for standard web pages:

• rate_limit_headers
• cors_headers
• real_agent_test
• content_negotiation_json
• idempotency_support

These checks stay accessible for citation preservation (EC-2), but no longer feed into the composite AI-Readiness Score.

Added: UCP (Universal Calibration Profile) manifest detection. The scanner now fetches /.well-known/ucp and validates the JSON structure. This check is conditional on commerce/API site classification.

Weight rebalance: dropping the 5 deprecated API checks and adding the UCP manifest parameter shifted the category weights, so we rebalanced them. You can find the updated weights in content/methodology/version.json and on the methodology overview page.

Score compatibility

Because of the weight rebalance, v4.4 scores aren't directly comparable to scores from v4.3 or earlier. The score history chart flags this with a methodology marker at the v4.4 boundary, and old reports keep their original scores.

Phase 2: Content Extractability (v4.2)

v4.2 introduces four new parameters and one enhancement, together worth 17 weight points. That brings the active parameter count from 35 (v4.1) to 39, and lifts total scoring weight from 199 to 216.

Parameters added

Parameter	Group	ID	Weight
Answer-first compliance	D (Content Quality)	D6 (ID 42)	5
Chunk extractability	D (Content Quality)	D7 (ID 43)	5
TL;DR / Key Takeaways section	D (Content Quality)	D8 (ID 44)	3
Dated statistics ratio	D (Content Quality)	D9 (ID 45)	4
Answer-structure (enhanced v2)	D (Content Quality)	D3 v2 (ID 31)	5 (unchanged)

Source: marketing/archive/friendly4ai-geo-scanner-v4-2-extractability-update.md — Phase 2 section.

What each check measures

Answer-first compliance (D6) checks each H2 section for whether its first sentence answers the implied question in 30–60 words, with no preamble.

Chunk extractability (D7) runs three deterministic checks on every H2 block: an explicit entity in the first sentence, no anaphoric pronouns in the first 20 words, and a noun phrase in the heading itself.

TL;DR section (D8) looks for a Key Takeaways or Summary block of 3–5 bullets — the chunk the friendly4AI Writing Guide flags as the single most-cited one per article.

Dated statistics ratio (D9) measures what share of numeric claims carry a year. Retrieval systems tend to ignore undated statistics.

Answer-structure v2 (D3) extends the existing D3 check. Its weight stays the same (5), but the logic now adds an answer-first bonus for Q&A pages.

Score compatibility

Because v4.2 raised the weight budget from 199 to 216, v4.1 and v4.2 scores for the same site are not directly comparable. The score history chart marks the v4.2 boundary with a methodology marker, and every report now carries a scoringVersionLabel field (v4.1, v4.2) so methodology changes stay visible in that chart.

Phase 1: Structural Extractability (v4.1)

v4.1 adds three structural checks worth a combined 14 weight points. That lifts the active parameter count from 32 (v3) to 35, and total scoring weight from 185 to 199.

Parameters added

Parameter	Group	ID	Weight
Paragraph length distribution	C (Content Structure)	C6 (ID 38)	5
Section length distribution	C (Content Structure)	C7 (ID 39)	5
Internal link coverage (body only)	B (SEO & Schema)	B8 (ID 47)	4

Source: marketing/archive/friendly4ai-geo-scanner-v4-2-extractability-update.md — Phase 1 section.

What each check measures

Paragraph length distribution (C6) scores the share of paragraphs that land in the 40–80 word optimal range, and penalises any paragraph over 120 words — those block AI chunking.

Section length distribution (C7) scores the average words per H2 block against a 120–180 word target, the typical span of an AI-generated answer.

Internal link coverage (B8) counts internal links found in body content only — <main>, <article>, or the largest content block — while ignoring links in <header>, <footer>, and <nav>. It expects three or more body links to distinct pages, and adds a bonus for links into glossary or definition URLs.

Score compatibility

Because v4.1 raised the weight budget from 185 to 199, v3 and v4.1 scores for the same site no longer line up directly. The score history chart marks a methodology boundary at v4.1.