The Six Surfaces AI Engines Read

An AI-visibility audit does not measure a website as one thing. It scores six independent surfaces, each of which gates a different stage on the path from question to citation.

From Question to Citation

An answer engine does not read a website; it runs a pipeline. Between a user's question and a cited brand, the engine discovers candidate sources, fetches them, extracts passages, interprets what those passages claim, decides which claims to trust, and attributes them. A failure at any stage ends the chain. The best-written page on the web earns no citation if the retriever never fetches it, and a perfectly fetchable page earns none if nothing in it can be attributed.

Question → Retrieval → Fetch → Extraction → Synthesis → Attribution

An AI-visibility audit exists because each stage of that pipeline reads a different surface of a site. The audit's unit of analysis is not the page but the surface — six of them, each gating a different link in the chain. Measuring visibility as one thing confuses six separate failure modes; measuring six surfaces localizes the failure to the stage where it actually occurs.

The Six Surfaces

• AI Indexability measures whether retrieval can find and fetch the page at all. What does a fetch-time request actually receive — server-rendered substance, or a JavaScript shell that only becomes content in a browser? Response codes, sitemap coverage, and canonical discipline sit here. A page whose content exists only after client-side rendering is present for users and absent for a retriever that receives the shell.

2. AI Crawler Governance reads the policy surface: robots directives, crawler-specific rules, agent declaration files. The question is not only whether crawlers are permitted but whether the permission is deliberate — explicit allow and deny decisions that reflect intent — or accidental, inherited from a default configuration nobody reviewed. Ambiguous governance reads as ambiguity about the entity itself.

3. Performance & Technical Health measures fetch economics. Retrieval operates under latency budgets, and a source that responds slowly or intermittently loses its slot to a source that responds fast. Response latency, error rates, redirect chains, and stability under automated access determine whether a page is cheap enough to read at the moment the engine needs it.

4. Structured Data measures whether the page declares what it is in extraction-ready form. JSON-LD entities, article metadata, organization identity: the difference between a model inferring what a page describes and reading a direct statement of it. Inference can be wrong; a declaration that agrees with the visible content is corroborating evidence.

5. AI Readiness measures the declaration surfaces built for agents: llms.txt and related machine-readable files, quotable statements, literal status vocabulary. Whether the site states what it is in a form a model can consume directly rather than reconstruct from prose and layout.

6. Static Content measures the body itself: whether the substance lives in the HTML the fetcher receives, whether claims are stated declaratively, whether entities are named rather than implied, and whether the prose is citable sentence by sentence. This is the surface the synthesis stage actually quotes.

Why the Pillars Are Independent

Each surface answers a different question, and no score on one can compensate for failure on another. Indexability answers whether the engine can reach the page. Governance answers whether reading it is permitted and deliberate. Technical health answers whether fetching it is affordable under a latency budget. Structured data and AI readiness answer whether the engine can interpret it without guessing. Static content answers whether anything on the page is worth quoting.

Citation probability behaves closer to a product than a sum. A flawless structured-data score cannot rescue a page the retriever never fetched; flawless indexability cannot manufacture a citable claim the text does not contain. The pillars correlate with citation behavior because each removes a distinct reason for the engine to drop the source — not because strength in one creates strength in another.

Why Pillar Scores Beat a Blended Number

A single composite hides the failure it is meant to locate. Two sites can return the same blended score with opposite problems — one unreachable, one unreadable — and require entirely different work. The composite says something is wrong; the pillar scores say which stage of the pipeline is dropping the site, and therefore which fix ships first.

This is the scoring model Geolify, the studio's Live Product GEO platform, implements: a GEO Score from 0–100 reported alongside per-pillar results across the six surfaces, with the Action Center translating each failing pillar into an ordered fix — issue, recommendation, expected impact. The audit's output is a work queue, not a grade.

Audit-First as an Operating Loop

The anatomy is what makes audit-first an operating discipline rather than a slogan. A site cannot fix a surface it has not measured, and the measurement has to be per-surface to be actionable. The loop follows directly: audit the gaps, fix what matters, track the lift — then audit again, because every change to the site is a change to one of the six surfaces the engines read.