Scoring Specification

How the Analyzer converts four category check results into a single overall score — the STATUS_SCORES mapping, equal category weights, the rounding formula, and the grade scale used to interpret the result.

Check status values

Every checker returns a CheckStatus value for its category. The Analyzer maps each status to a numeric score using the STATUS_SCORES constant defined in lib/analyzer/types.ts:

Status	Score	Meaning
pass	100	The category is fully implemented and meets all Analyzer criteria.
warn	65	The category is partially implemented or has a recoverable issue (e.g. thin content, missing optional fields, or implicit rather than explicit configuration).
fail	20	The category has a critical implementation error that actively harms AI visibility (e.g. all crawlers blocked, all critical metadata absent).
not_found	10	The required resource was not found (HTTP 404). The category is not implemented.
unknown	40	The Analyzer could not evaluate the category due to a network error or unparseable response. The score is penalized but less severely than a confirmed failure.

not_found scores lower than unknown

A not_found result (score 10) is penalized more heavily than unknown (score 40) because a 404 is a definitive signal that the resource does not exist, whereas unknown may reflect a transient network condition rather than a missing implementation.

Category weights

The four AnalysisCategory values each contribute equally to the overall score. No category is weighted more heavily than another:

Category	Weight	Checker function
llmsTxt	25%	checkLlmsTxt
aiMetadata	25%	checkMetadata
crawlerRules	25%	checkCrawlerRules
structuredSignals	25%	checkStructuredSignals

llms.txt and AI metadata together account for 50%

Because each category contributes 25%, fixing llmsTxt and aiMetadata from not_found to pass adds up to 45 points to the overall score. These two categories are the highest-leverage starting point for most sites.

Score formula

The calculateScore function in lib/analyzer/score.ts computes the overall score by averaging the four category scores and rounding to the nearest integer:

overall = round((llmsTxt + aiMetadata + crawlerRules + structuredSignals) / 4)

Each category score is the STATUS_SCORES value for that category's CheckStatus. The result is a whole number between 0 and 100 inclusive.

Rounding uses standard half-up rounding

JavaScript's Math.round() rounds 0.5 up. A score of 48.75 rounds to 49; a score of 48.5 rounds to 49; a score of 48.25 rounds to 48.

Grade scale

The Analyzer maps the overall score to a grade label for display in reports and the analyzer UI:

Score range	Grade	Interpretation
90–100	Excellent	All four layers are implemented and passing. The site is fully GEO-ready.
75–89	Good	Most layers are passing. One or two categories have minor issues or are returning `warn`.
50–74	Fair	Some layers are implemented but others are missing or failing. Meaningful improvements are available.
25–49	Poor	Most layers are missing or failing. The site has limited AI search visibility.
0–24	Not optimized	No layers are implemented, or critical failures exist across multiple categories. The site is not visible to AI search engines.

Interpretation

The overall score is a composite signal, not a precise measurement. Use it to track progress over time and to prioritize which categories to address first.

Excellent (90–100). All four categories are passing. The site exposes an llms.txt file, complete AI metadata, explicit AI crawler rules, and valid structured data. No high-priority recommendations are expected.
Good (75–89). The site has implemented most of the specification. A score in this range typically means one category is returning warn — for example, an llms.txt file exists but has thin content, or structured data is present as microdata rather than JSON-LD. Address the warn category to reach Excellent.
Fair (50–74). Two or more categories are returning warn or not_found. The site has partial AI optimization. Prioritize the categories with the lowest scores — not_found (10) and fail (20) have the most room for improvement.
Poor (25–49). Most categories are missing or failing. The site is unlikely to appear in AI-generated answers. Start with llmsTxt and aiMetadata — together they account for 50% of the score and are typically the fastest to implement.
Not optimized (0–24). No meaningful AI optimization is in place, or one or more categories are actively blocking AI access (e.g. all crawlers blocked via robots.txt). Fix any fail conditions first, then address not_found categories.

A high score does not guarantee AI citation

The Analyzer measures implementation completeness against the GEO Specification. It does not measure whether AI engines have actually indexed your site or whether your content appears in AI-generated answers. Those outcomes depend on content quality, crawl frequency, and AI engine behavior outside the scope of this specification.

PreviousStructured Signals NextRecommendations