How We Score a Website for AI Visibility: The GeoTest Scoring Methodology

Abstract

Traditional search engine optimization produced a well-understood set of signals: backlinks, keyword density, page authority, Core Web Vitals. Generative Engine Optimization requires a different measurement framework. When the output of a search interaction is a synthesized paragraph rather than a list of ranked links, the signals that predict success change fundamentally. A page can rank #1 on Google and never be cited in an AI-generated answer.

This document describes the GeoTest Scoring Engine v3.0 — what we measure, how we measure it, why each category was included, and what the resulting scores mean in practice. The category weights below mirror the production code in supabase/functions/analyze-site/scoring-v3.ts: the scoring contract is published rather than proprietary.

1. Background: why a new scoring framework was necessary

Traditional SEO is optimized for retrieval; GEO is optimized for extraction. A search engine retrieves the most relevant pages. An AI citation engine extracts the most useful content from the pages it can read. Every category in the engine traces back to one question: can an AI agent find, read, parse, and extract a citable answer from this page?

What existing research established

• Chartbeat's analysis of traffic across more than 2,500 publisher websites found Google search referral traffic down 33% globally and 38% in the US year-over-year as of early 2026. Publishers with fewer than 10,000 daily pageviews absorbed a 60% decline; mid-sized publishers are down 47% (Chartbeat, 2026).
• Similarweb's tracking of zero-click searches shows the rate rising from 56% to 69% between May 2024 and May 2025. When Google's AI Overviews surface, 83% of searches end without a click; in AI Mode, 93% (Similarweb).
• A Pew Research Center study of 68,000 real search queries found users clicked results 8% of the time when an AI summary appeared, versus 15% without one — a 47% reduction in click probability (Pew Research Center).

How the framework was built

We ran queries against Google AI Overviews, Gemini, ChatGPT, and Perplexity, tracked which pages were cited, and compared the structural properties of cited versus non-cited pages in the same category and competitive set. The patterns that recurred consistently enough to become scoring categories: valid JSON-LD structured data, FAQPage schema on question-format queries, page depth above roughly 800 words, server-rendered rather than JavaScript-injected content, and off-site brand mentions correlating more strongly with citation than backlink profiles.

2. The six-category framework

Total: 100 points. No bonuses; penalties act as absolute caps. Sections are sequenced with crawler access first, since it gates everything downstream — a site with poor crawler access or broken schema cannot benefit from strong content quality.

Score interpretation scale

In practice, the majority of mid-market business websites we audit score between 45 and 65 — functional sites with real content that are nonetheless structurally invisible to AI citation engines.

3. Category detail, testing protocol and limitations

The full article walks each category in turn — Crawler Accessibility (8 pts), Structured Data (18 pts), Content Quality (25 pts), Citation Worthiness (20 pts), Authority Signals (17 pts) and FAQ & Q&A Format (12 pts) — followed by the testing protocol, scoring and reporting model, and the framework's stated limitations. Crawler Accessibility covers indexing and crawlability (noindex directives, robots.txt, sitemap currency, crawl traps) plus content structure for AI parsing: heading hierarchy as a structural map, and whether text is present before JavaScript executes.

Read the complete methodology at geotest.ai/methodology.