docs-ai-scorer

A single-file Node.js script that measures how AI-friendly a docs site is. It powers the scoring in Which docs platforms are actually AI-friendly? on the Jamdesk blog.

Run it

node score-docs.mjs <platform-name> <page-url-1> <page-url-2>

Outputs JSON with:

• Text-to-HTML ratio
• SSR pass (does the page server-render real content?)
• llms.txt / llms-full.txt discovery — checked at the origin root and every ancestor subpath of the tested URL, so /docs/llms.txt counts, not just /llms.txt. The JSON records where it was found via llmsTxtLocation (root, subpath, or none).
• sitemap.xml and per-page .md-endpoint presence
• AI-crawler robots policy (GPTBot, ClaudeBot, PerplexityBot, etc.)
• Chrome-noise ratio between two pages on the same site

No npm install needed. Just Node.js 24 (or any version with fetch and AbortSignal.timeout — Node 18+).

Reproduce the article

./run-all.sh > results.json

Uses platforms.json to score the seven platforms covered in the article. Results land in results.json.

Submit your platform's results

Run the script against your platform's docs and open an issue with your numbers. If the reproduction stands up, we'll add your platform to the scorecard in the article.

What this script doesn't do

• It doesn't follow JavaScript-rendered content (it makes plain HTTP GETs and parses HTML).
• It doesn't simulate Cursor or any specific AI coding agent.
• It doesn't grade answer quality — that's a separate manual rubric documented in the article.

Changelog

• v1.1 — Fixed the discovery-file probe. v1.0 checked only the bare origin root (/llms.txt), which produced false negatives for docs served from a subpath (e.g. /docs/llms.txt) even though the llms.txt spec allows subpaths. v1.1 checks the root and every ancestor segment, follows redirects, validates the body is real markdown rather than an SPA shell, and reports llmsTxtLocation. The story behind the fix is in the article.

License

MIT. © Jamdesk.