Building a Six-Pillar Audit Pipeline

The architecture behind the GEO Score: fetch a site the way an AI engine retrieves it, run six independent pillar checks, and compile the failures into a prioritized fix list. Independence is the design requirement — a strong average can hide a blocked crawler.

What the Pipeline Has to Prove

An audit is only useful if two conditions hold: the findings point at something fixable, and running it again next week produces a comparable result. Geolify's audit pipeline is built around those two requirements. It fetches a site the way an AI engine's retrieval layer sees it, checks whether the content is actually present in what comes back, scores six independent pillars, and compiles the failures into a prioritized action list. The output is not a grade. It is a work order.

The six pillars — AI Indexability, AI Crawler Governance, Performance and Technical Health, Structured Data, AI Readiness, and Static Content — each answer a different question about how engines like ChatGPT, Perplexity, Gemini, and Claude can talk about a site. They share a pipeline but not a score.

The Stages, in Order

The pipeline runs as a sequence of stages, each writing evidence that the next stage consumes:

• Fetch. The pipeline requests the site's pages and records what actually comes back: status codes, headers, redirect chains, and the raw HTML payload. This stage mimics the retrieval path an answer engine depends on — what is returned over the wire, before any rendering happens.
• Renderability checks. The fetched HTML is examined for substantive content: real text, real links, structured data blocks. A page that ships an empty shell plus a JavaScript bundle fails here, because most AI retrieval does not run a browser.
• Pillar scoring. Six independent scorers evaluate their own evidence sets. Crawler governance reads robots directives and headers; structured data parses schema blocks; static content measures what the page contains without client-side execution.
• Action synthesis. Failures across all six pillars compile into the Action Center: each item carries the issue, a concrete recommendation, and the expected impact, ordered so the fixes that matter most come first.

Fetch → Renderability → Six Independent Pillar Scores → Action Center (prioritized fixes)

Why Independent Scores Beat a Blended Number

A single blended score answers "how are we doing" and nothing else. Worse, it lies by averaging. A site with excellent structured data, fast responses, and clean markup can still be invisible if its robots directives block the crawlers that matter. Blended into an average, that hard zero looks like a respectable middle. The dashboard says healthy; the engines see nothing.

Independent pillar scores prevent that failure by construction. Each pillar measures a different mechanism with its own evidence, so a failure cannot be diluted by strengths elsewhere. The vector is the diagnosis: a low Crawler Governance score beside five healthy ones is a specific, fixable condition rather than a vague sense of underperformance. The GEO Score still rolls up to a 0–100 number for tracking over time, but the pillars underneath stay visible, because the pillars are where the work happens.

What Makes an Audit Repeatable

Repeatability is an engineering property, not a promise. Three things produce it:

• Deterministic checks. The same page state produces the same findings. Each pillar check is a rule applied to fetched evidence, so a re-run on an unchanged site reproduces the previous result rather than re-rolling it.
• Fixed evidence sets. Each pillar declares what it reads — headers, robots directives, schema blocks, HTML payloads. The evidence behind a score can be inspected, which is what makes a finding disputable and therefore trustworthy.
• Diffable runs. An audit that cannot be compared to last week's is a snapshot, not a loop. Runs are compared directly: new issues, resolved issues, persistent issues. The weekly motion only works because each run is a valid diff against the last.

Between audits, the Prompt Monitor tracks what the engines are actually saying: which prompts surface the brand and how mentions shift. The audit measures the site; the monitor measures the engines' response to it. The loop needs both.

The Property That Matters

The design goal is a pipeline where every number on the report traces back to a fetched artifact and a named rule. That traceability separates an audit from an opinion — and it is what lets a team act on a finding this week and verify the lift on the next run.