Methodology
GeoPageSpeed reads a page the way most AI crawlers do: as raw HTML, without running JavaScript. Every result comes from fixed rules (parsing, counts and thresholds), so the same page always gets the same report. No AI model is used anywhere.
Evidence tiers
Advice about AI search ranges from documented fact to guesswork. Every check carries one of four tiers, and the tier changes how much it counts.
| Tier | Meaning | Weight |
|---|---|---|
| Confirmed | Documented by the search engine or AI operator itself. | 100% |
| Supported by research | Backed by published studies or large-scale measurements, not by the platforms. | 100% |
| Good practice | Widely recommended, but no platform has confirmed it matters for AI answers. | 50% |
| Unconfirmed benefit | Proposed, but with no evidence yet that AI systems use it. Shown, never scored. | 0% |
Checks in the Confirmed and Supported tiers link to their sources. Some checks are informational: they explain a choice you made (such as blocking AI training crawlers) without judging it, and never count.
Scoring
- Each check scores pass = 1, warn = ½, fail = 0, weighted by severity (critical 10, warning 5, info 1) and by its evidence tier. Results we're less sure about (low confidence) count half. N/A and errors don't count.
- The overall score is a weighted average of the modules: Crawlability & AI-bot access 25%, Structured data 10%, Answer-extraction readiness 30%, Technical SEO basics 10%, Page performance proxies 10%, Authority & trust signals 15%. A module whose checks are mostly “Convention” counts for less in that average, down to half its weight when all of them are, so best practices without AI-specific evidence can't pull the score down as hard as confirmed problems.
- Grades: A ≥ 90, B ≥ 75, C ≥ 60, D ≥ 40, F below 40.
Gates
Some failures make the rest of the page irrelevant. If the page can't be fetched, the score is 0. If one of these checks fails with Confirmed evidence and high confidence, the score is capped at 49:
- `User-agent: *` allows this page
crawl.robots_wildcard_block - Googlebot and Bingbot allowed
crawl.search_engines_allowed - AI search crawlers allowed
crawl.ai_search_bots_allowed - Page is indexable
crawl.meta_robots_index - Content works without JavaScript
crawl.js_dependency - No anti-bot challenge
crawl.challenge_page
The three verdicts
Above the score, the report answers three questions: can AI bots reach the page, can they read it without JavaScript, and can they lift a clean answer from it. Each answer links to the checks it is based on.
Limits
- One page per scan. We don't crawl the rest of the site.
- Static analysis only: content that appears only after JavaScript runs is invisible to us, as it is to most AI crawlers.
- Filler, definition and readability checks work on English pages only. On other pages written with spaces between words, questions and FAQs are found by their question marks alone; Japanese, Chinese and similar scripts get N/A for those checks.
- Bylines and dates are found heuristically. Thresholds are starting values.
- We can't see whether AI systems actually cite you. That needs live queries, which are planned for Pro.