Methodology / rules v2.0 / 17 July 2026

How the website-readiness score works

The pillars, weights, observable signals and limitations are public. You can check our working and distinguish a verified result from an inference.

100

possible points

Six weighted pillars scored from one captured public page.

The method in one paragraph

The free Surfaced audit fetches one public page and scores captured evidence across discovery-crawler declarations (20%), initial-HTML readability (20%), extractability heuristics (25%), structured-data signals (15%), date transparency (10%) and technical hygiene (10%). The same inputs produce the same result under the same ruleset. Unknown crawler evidence is excluded and the score is marked provisional. This is a readiness diagnostic, not a measurement of current AI recommendations.

Six pillars, one score

Weights total 100%

Access and rendering

20%

Discovery-crawler access

What we check

We parse Allow and Disallow rules for OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot and Bingbot against the checked page path. Matching follows the most-specific user-agent group and longest rule; an equal-length Allow wins. User-triggered fetchers, training crawlers and Google-Extended are classified separately and do not affect this score.

Why this weight

A declared block can stop a discovery crawler fetching the page. The five checks are equally weighted as technical coverage, not market share. A 404 or 410 means no declared restriction; 403, 429, server errors, timeouts and invalid HTML responses remain Unknown rather than being counted as Allowed.

20%

Readable in initial HTML

What we check

We fetch the public page without executing JavaScript and measure the meaningful text in the initial HTML. An apparent empty app shell scores zero; 1,200+ characters receives full marks, 600-1,199 is partial and less than 600 is weak.

Why this weight

Rendering behaviour varies by crawler and can change. Initial HTML is the conservative, directly observable surface: it is available without waiting for a second rendering stage and usually improves speed and resilience for people as well as crawlers.

Evidence and structure

25%

Are your pages answer-ready? (citability)

What we check

Seven observable heuristics: an answer-first opening (25-90 words), a self-contained 120-180 word passage, specific figures, quote or attribution cues, outbound links to a limited set of primary/authoritative sources, question-shaped headings and scannable paragraphs.

Why this weight

The KDD 2024 GEO study reported that content changes including citations, statistics and quotations could improve its visibility metric by up to 40%, with effects varying by query and domain. Our regex checks are proxies: a number is not automatically sourced, a quote cue is not proof of authority and a passage length is not a ranking factor.

15%

Structured data

What we check

We parse JSON-LD, including nested @graph entries, and look for a business/entity type, relevant content types and BreadcrumbList. Malformed blocks are reported rather than silently treated as valid.

Why this weight

Structured data can make entity and content types explicit, but type presence alone does not prove that the markup is accurate, complete or eligible for a search feature. Visible page facts and JSON-LD must agree.

Freshness and hygiene

10%

Date transparency

What we check

We look for a visible published, reviewed or updated date. A visible date scores 100; a Last-Modified header alone scores 50; no detected date scores 25 and is shown as a warning, not a failure.

Why this weight

A maintained date helps people and machines assess time-sensitive content. It does not prove that content is accurate or recent, and a Last-Modified header may reflect a deploy rather than an editorial review.

10%

Technical hygiene

What we check

Title tag, meta-description presence/length, a parseable canonical and cross-origin warning, HTTPS, mobile viewport, plus meta robots and X-Robots-Tag noindex detection.

Why this weight

These checks reduce ambiguity and accidental exclusion. They do not prove page quality or guarantee a citation.

What this score doesn’t measure

The free score measures observable readiness signals on one page. It does not measure whether ChatGPT recommends you today, how you compare with competitors, or how answers change over time. Those require repeated provider queries, which the paid plans run across ChatGPT, Google Gemini and Perplexity. AI answers vary between runs. We track trends across weeks and guarantee the method, never a ranking.

Sources

Primary documentation and research used to define observable checks and their limits.

Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024. The study reported up to 40% improvement on its generative-engine visibility metric, with results varying by strategy and domain.
OpenAI publisher documentation for OAI-SearchBot and search inclusion.
Anthropic crawler documentation separating ClaudeBot, Claude-User and Claude-SearchBot.
Perplexity crawler documentation separating PerplexityBot and Perplexity-User.
Google crawler documentation, including Googlebot and the non-crawler Google-Extended control token.

We review the method as engine behaviour changes and date-stamp every revision. If you think a check is wrong, tell us: hello@surfaced.co.nz.

See your own score

Free, no signup, about 20 seconds.

Check my site