How the Discoverability score is calculated

The five components

• Crawl access (25): robots.txt served as text/plain; no site-wide Disallow; Googlebot and Bingbot allowed; answer-engine crawlers allowed; training crawlers allowed; and, independently of robots.txt, search and AI crawler user-agents actually receive the page from your edge rather than a 403 or challenge.
• Site plumbing (20): valid HTTPS; http redirects to https; the alternate host (www or apex) redirects permanently to the canonical one; at most one redirect to the root page; root page returns 200 with no noindex; a valid sitemap; the Sitemap line in robots.txt is absolute; lastmod present; a random URL returns 404.
• Page metadata (20): title 30 to 65 characters; meta description 110 to 170; exactly one H1; an absolute canonical on the canonical host; Open Graph and Twitter tags; og:image loads at 1200×630 or larger; sampled inner pages carry the same basics.
• Structured data (20): JSON-LD present and parseable; an Organization (or Person / LocalBusiness); WebSite; a type describing what the site offers; BreadcrumbList on inner pages.
• Answer-engine readiness (15): /llms.txt; an opening paragraph that defines the brand; question-form headings; at least 300 words of real text; little hidden text; a declared language.

Severity and the cap

Failures are graded by consequence. Critical: the site cannot be crawled or indexed (site-wide Disallow, search crawlers blocked in robots.txt or at the edge, noindex on the root page, no HTTPS). Important: it will be crawled badly or read incompletely (no sitemap or a relative Sitemap line, www not redirecting, missing canonical, JSON-LD errors, AI crawlers blocked at the edge, no llms.txt). Everything else is a recommendation.

As with the Email Security score, the headline is capped at 89 while an important fix is open and at 69 while a critical one is open, and the deduction is taken from the component with the open fix so the rows still add up. Clear the fix, press Verify, and the cap disappears.

Why blocking AI crawlers costs points

The score measures whether the site will appear in answers and in future models, and a crawler that is refused cannot quote or learn from the page. Blocking is a legitimate choice; the score reports its consequence rather than a judgement. Blocking search crawlers is critical, blocking answer-engine crawlers (OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended) is important, and blocking training-only crawlers (GPTBot, CCBot, Applebot-Extended) is a recommendation worth 2 points.

What the scan fetches

About 14 to 18 requests, run concurrently and finished within a few seconds: the root page (https, http and the alternate host), robots.txt, the sitemap and up to four pages from it, llms.txt, the og:image header, one random URL, and the root page once per crawler user-agent. Bodies are capped, results are cached for 15 minutes, and a domain is scanned fresh at most 30 times an hour across all users. Opening a result link never triggers a fetch; only pressing Scan or Rescan does.