methodology · last updated July 2026
How the AI-visibility score works
The whole method is public — the pillars, the exact weights, every check and the evidence behind each one. Agencies won't show you this. We think you should be able to check our working.
In one paragraph: the free Surfaced audit fetches your website the way an AI crawler does — no JavaScript, no cookies — and scores it 0–100 across six weighted pillars: AI-crawler access (20%), server-rendered content (20%), citability (25%), structured data (15%), freshness (10%) and technical hygiene (10%). The score is deterministic: the same site always gets the same score, and every point can be traced to a named check. It measures whether AI engines can find and quote you — paid plans then track whether they actually do.
Can AI even reach you?
20% of scoreWhat we check
We parse your robots.txt and test whether the AI search crawlers — OAI-SearchBot and ChatGPT-User (ChatGPT), PerplexityBot, Claude-SearchBot, Googlebot, Bingbot and Google-Extended — are allowed to fetch your site.
Why it's weighted this way
AI engines can only cite pages their crawlers are allowed to fetch. Each blocked search crawler costs 20 points in this pillar, because each one is an entire engine your customers use that cannot quote you.
Server-rendered for AI
20% of scoreWhat we check
We fetch your homepage exactly the way an AI crawler does — no JavaScript execution — and measure how much real text arrives in the raw HTML. A page whose content only appears after scripts run scores zero here.
Why it's weighted this way
GPTBot, ClaudeBot and PerplexityBot do not run JavaScript. If your site is a JavaScript shell, AI engines see a blank page. 1,200+ characters of server-rendered text scores full marks; under 600 fails.
Are your pages answer-ready? (citability)
25% of scoreWhat we check
Seven checks: an answer-first opening (25–90 words), a self-contained passage in the 134–167-word range, sourced statistics, attributed quotes, outbound links to authoritative sources, question-shaped headings, and scannable paragraphs.
Why it's weighted this way
The Princeton/KDD-2024 GEO study measured citation lift from adding sourced statistics (~37%), citing authoritative sources (~40%) and adding quotations (~30%). SE Ranking's analysis found AI-cited passages cluster around 134–167 words, front-loaded. This is the heaviest pillar because it's where most sites lose the most.
Structured data
15% of scoreWhat we check
We parse your JSON-LD (including nested @graph) and check for entity types AI engines use to identify you: Organization or LocalBusiness (heaviest), content types like Article or Product, and BreadcrumbList.
Why it's weighted this way
Structured data tells AI engines what you are, where you are and what you offer without guessing. Note: Google retired FAQ rich results for all sites in May 2026 — we treat FAQPage as an AI-only signal now and recommend QAPage for genuine Q&A.
Freshness
10% of scoreWhat we check
We look for a visible date on the page, then fall back to the Last-Modified HTTP header. Visible date scores 100; header-only scores 45; neither scores 0.
Why it's weighted this way
AI engines favour recently updated pages, and an undated page reads as stale. A visible, maintained “last updated” date is the cheapest freshness signal there is.
Technical hygiene
10% of scoreWhat we check
Title tag (10–70 characters), meta description, canonical tag, HTTPS, and a mobile viewport.
Why it's weighted this way
Table stakes. These don't win you citations on their own, but missing them costs you eligibility everywhere.
What this score doesn't measure
The free score measures eligibility— whether AI engines can reach, read and quote your site. It does not measure whether ChatGPT actually recommends you today, how you compare with competitors, or how answers change over time. Those require querying the engines repeatedly, which is what the paid plans do (~25 to ~100 buying-intent prompts per week across ChatGPT, Google AI, Perplexity and Claude). AI answers vary between runs — that's why we track trends across weeks rather than promising any single answer, and why we guarantee the method, never a ranking.
Sources
- Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024 (Princeton et al.) — citation lift from sourced statistics (~37%), authoritative citations (~40%) and quotations (~30%).
- SE Ranking — analysis of AI-cited passage length (134–167 word cluster).
- Ahrefs — cross-engine citation overlap (~1 in 9 sites) and the 75k-brand mention study.
- Published crawler documentation from OpenAI, Anthropic, Perplexity and Google.
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.