Methodology

How the score works.

One score out of 100, from 80 checks across 8 weighted groups. Every check, weight, and rule on this page is read from the same source the scanner itself uses, so it cannot drift from the live product. The per-check fix docs are not repeated here: each check below links to its fix doc in the public agent-skills index.

Scoring rules

  • Pass scores full points, warn scores half points, fail scores zero.
  • Skip also scores zero, like a fail, when a check could not be passed: for example, a page that does not publish the file or data the check looks for. A skip is not a pass, and it is not neutral.
  • Not scored is different from skip: it means our own data source was unreachable, so the check is removed from both the numerator and the denominator of its group. It neither helps nor hurts the score, and the report says so.
  • Excluded means an admin preview check that is not yet promoted to public scans. Excluded checks are also removed from numerator and denominator, and they never affect a public score.
  • A group score is the rounded percentage of its scored check weights earned. A group with no scored checks does not contribute to the overall score.
  • The overall score weights each group score by its group weight, then rounds once at the end. Group weights total 100.
  • Critical failures - missing llms.txt, sitemap-llms.xml, or HTTPS, and any other check marked critical - score zero on high-weight checks, so fixing them lifts the score fastest. There is no artificial score cap.

Readiness levels

  • 80 to 100 - Level 4 · AI-Ready
  • 60 to 79 - Level 3 · Structured
  • 40 to 59 - Level 2 · Discoverable
  • 0 to 39 - Level 1 · Invisible

Groups and weights

Group weights: AI Discovery 25 · AI Pipeline Visibility 15 · Agent Protocols 8 · DNS & Identity 8 · Security 12 · Performance 10 · SEO & Content 12 · Accessibility 10. Weights total exactly 100.

AI Discovery

Can AI systems find and understand the site?

14 checks

25%

AI Pipeline Visibility

Do AI data pipelines actually have your pages?

8 checks

15%

Agent Protocols

Can agents act on the site, not just read it?

11 checks

8%

DNS & Identity

Is the domain reachable, valid, and trustworthy?

8 checks

8%

Security

Are responses hardened against common attacks?

9 checks

12%

Performance

Does it load fast on real mobile connections?

10 checks

10%

SEO & Content

Can search engines index it — and is the content real?

13 checks

12%

Accessibility

Can everyone use it, including assistive tech?

7 checks

10%

Scan profiles

Public scans use the default profile, which applies every check at its default weight. The admin Scan Lab can pick a profile that reweights groups for what the site actually is. Profiles never add or remove checks; they only change group multipliers at aggregation.

  • All checks - Every check at its default weight.
  • Content site - Blog, publisher, marketing site — discovery and content signals matter most.Multipliers: AI Discovery x1.25, AI Pipeline Visibility x1.2, SEO & Content x1.2, Performance x1.1, Agent Protocols x0.6, Accessibility x1.1
  • API / Application - API-first product or web app — agent protocols and hardening matter most.Multipliers: Agent Protocols x1.5, Security x1.25, DNS & Identity x1.1, AI Discovery x0.8, SEO & Content x0.6, Accessibility x0.7
  • Commerce - Store or booking flow — agents must be able to pay you.Multipliers: Agent Protocols x1.6, Security x1.2, DNS & Identity x1.1, Performance x1.1, SEO & Content x0.8

Projected score and weekly to-do

The projected score in a report is a full recompute, not a guess: the top failing, warning, and skipped checks are marked as passed, group scores are recomputed with the rules above, and the overall score is recomputed with the scan profile. No AI call is made for a projection. The single-fix gains shown beside each fix are estimates for fixing that check alone; they do not add up exactly, because group scores round.

The weekly to-do uses the same ranked fix list, tagged by owner. The SEO and Content group is tagged Content. Every other group is tagged Developer, except the copy and label checks listed here, which are also tagged Content: button-names, cite-comparison, cite-concrete-figures, cite-direct-open, cite-sections-headed, cite-structured, h1, heading-order, img-alt, link-text, meta-desc, thin-content, title-tag. Owner tags are a static mapping, not an AI judgement, and the same list feeds a fresh scan, a weekly scan, and a history re-run, so they can never disagree.

All checks

Every check below links to its fix doc in the public agent-skills index at /.well-known/agent-skills/. The fix docs are the authoritative fix for each check; this page lists what is checked and what it is worth, and does not repeat them.

AI Discovery - 25%

  • llms.txt exists5 ptscriticalDeveloperFix doc
  • llms.txt has allow/deny rules3 ptsDeveloperFix doc
  • llms.txt is a reasonable size1 ptsDeveloperFix doc
  • sitemap-llms.xml exists5 ptscriticalDeveloperFix doc
  • robots.txt exists1.5 ptsDeveloperFix doc
  • robots.txt doesn't block AI crawlers1.5 ptsDeveloperFix doc
  • sitemap.xml exists2 ptsDeveloperFix doc
  • sitemap.xml is valid and non-empty1 ptsDeveloperFix doc
  • JSON-LD structured data present0.5 ptsDeveloperFix doc
  • JSON-LD blocks parse cleanly1 ptsDeveloperFix doc
  • Organization node declares name + url1 ptsDeveloperFix doc
  • Meta description present1 ptsContentFix doc
  • Canonical URL set0.5 ptsDeveloperFix doc
  • Open Graph tags present1 ptsDeveloperFix doc

AI Pipeline Visibility - 15%

  • Common Crawl has recent captures3 ptsDeveloperFix doc
  • robots.txt allows AI crawlers per-bot2 ptsDeveloperFix doc
  • AI crawlers can fetch the homepage3 ptsDeveloperFix doc
  • Speakable schema marks answer-ready content0.5 ptsDeveloperFix doc
  • FAQPage schema present1 ptsDeveloperFix doc
  • LocalBusiness schema complete2 ptsDeveloperFix doc
  • Wikidata entity exists0.5 ptsDeveloperFix doc
  • Gemini knows the business3 ptsDeveloperFix doc

Agent Protocols - 8%

  • MCP endpoint advertised2 ptsDeveloperFix doc
  • agents.json / .well-known/agents.json1.2 ptsDeveloperFix doc
  • OpenAPI spec published1.2 ptsDeveloperFix doc
  • Developer / API docs page0.8 ptsDeveloperFix doc
  • AI plugin manifest0.8 ptsDeveloperFix doc
  • Contact page reachable0.5 ptsDeveloperFix doc
  • A2A Agent Card published0.5 ptsDeveloperFix doc
  • x402 agent payments0.3 ptsDeveloperFix doc
  • Universal Commerce Protocol (UCP)0.3 ptsDeveloperFix doc
  • Agentic Commerce Protocol (ACP)0.2 ptsDeveloperFix doc
  • Machine Payment Protocol (MPP)0.2 ptsDeveloperFix doc

DNS & Identity - 8%

  • A record resolves1 ptsDeveloperFix doc
  • AAAA record resolves (IPv6)1 ptsDeveloperFix doc
  • HTTPS serves a valid certificate2 ptscriticalDeveloperFix doc
  • Certificate chain verifiable1 ptsDeveloperFix doc
  • HSTS header set1 ptsDeveloperFix doc
  • CAA record restricts issuers0.5 ptsDeveloperFix doc
  • SPF record published0.5 ptsDeveloperFix doc
  • DMARC policy published1 ptsDeveloperFix doc

Security - 12%

  • Content-Security-Policy set2.5 ptsDeveloperFix doc
  • Clickjacking protection2 ptsDeveloperFix doc
  • X-Content-Type-Options: nosniff1 ptsDeveloperFix doc
  • Referrer-Policy set1 ptsDeveloperFix doc
  • Permissions-Policy set1 ptsDeveloperFix doc
  • No server version leak1 ptsDeveloperFix doc
  • Cookies hardened1.5 ptsDeveloperFix doc
  • No mixed content1 ptsDeveloperFix doc
  • HTTP redirects to HTTPS1 ptsDeveloperFix doc

Performance - 10%

  • PageSpeed performance ≥ 801 ptsDeveloperFix doc
  • Low external script bloat0.5 ptsDeveloperFix doc
  • LCP ≤ 2.5s1.5 ptsDeveloperFix doc
  • CLS ≤ 0.11.5 ptsDeveloperFix doc
  • TBT ≤ 300ms1 ptsDeveloperFix doc
  • FCP ≤ 1.8s1 ptsDeveloperFix doc
  • Text compression enabled1 ptsDeveloperFix doc
  • Mobile viewport configured0.5 ptsDeveloperFix doc
  • Resource hints used1 ptsDeveloperFix doc
  • Cache headers set1 ptsDeveloperFix doc

SEO & Content - 12%

  • Title tag well-formed1.5 ptsContentFix doc
  • Single H1 heading1 ptsContentFix doc
  • Images have alt text1.5 ptsContentFix doc
  • Favicon present0.5 ptsContentFix doc
  • No broken internal links2 ptsContentFix doc
  • Logical heading order0.5 ptsContentFix doc
  • Substantial page content0.5 ptsContentFix doc
  • One canonical host (www vs apex)0.5 ptsContentFix doc
  • Content split into headed sections1 ptsContentFix doc
  • Sections open with a direct statement0.5 ptsContentFix doc
  • Sections carry concrete figures1.5 ptsContentFix doc
  • Comparison structure present0.5 ptsContentFix doc
  • Enumerations use lists or tables0.5 ptsContentFix doc

Accessibility - 10%

  • Page language declared1.5 ptsDeveloperFix doc
  • Form fields labeled2 ptsDeveloperFix doc
  • Buttons have names1.5 ptsContentFix doc
  • Landmark regions present1.5 ptsDeveloperFix doc
  • Links have discernible text1.5 ptsContentFix doc
  • Skip-to-content link1 ptsDeveloperFix doc
  • No duplicate element IDs1 ptsDeveloperFix doc

What the scan cannot check

  • Whether a named AI engine actually recommends, cites, or names the site. The scan measures readiness signals on the site, not live answers from any engine.
  • Rankings, traffic, or revenue. A high score does not promise a ranking, a citation, or traffic in any named engine.
  • Content quality or relevance, beyond the structural signals listed above. The scan checks that content is present and structured; it does not judge whether the copy persuades a reader.
  • Backlinks and authority, full accessibility conformance, legal compliance, or security beyond the headers and transport checks listed above. This is not a penetration test, a legal review, or a manual accessibility audit.
  • Pages behind a login, a paywall, or a geo-block, and pages the scan cannot reach. The public scan checks the submitted URL and the files it references; the optional site scan covers up to 10 pages, not a whole site.
  • Anything after the scan runs. Scores are point-in-time: sites change, and performance results vary between runs.

SITE SCAN - UP TO 10 PAGES, NO SCORE CHANGE

The report tab also offers an optional site scan. It crawls up to 10 pages from the same site, auto-discovered from the sitemap and homepage links, or the pages you type in. It is deterministic parsing only: no verdict, no AI cost, and it never changes the 0-100 score, which stays on the single root scan.

What it reports: a link map of internal links, orphan pages (in the sitemap, linked from no crawled page), duplicate titles and canonicals, competing topics (pages sharing the same opening words with different titles), freshness signals (sitemap lastmod, visible publish and update timestamps, stale entries over 180 days), citation-pattern presence (question headers, definition blocks, comparison tables, lists, cited figures), and ecommerce and agent-commerce presence (product schema completeness, shipping and returns pages, UCP, ACP, MCP, and other agent-protocol endpoints). Presence only: the scan does not judge content quality, and a missing shipping or returns page is only reported when a page links to one or the crawl reaches one, never as a failure for a page not crawled.

Gating: the free view shows counts and group totals only. Named pages, page URLs, and page-level detail are in the full report (Pro or Agency), and the admin console shows full detail for testing. The site scan reuses the stack lookup static-asset rule, so images, fonts, scripts, and archives are never fetched as pages, and login, admin, and account paths are excluded.