https://github.com/jamdesk/docs-ai-scorer
Single-file Node.js script that measures how AI-friendly a docs site is.
https://github.com/jamdesk/docs-ai-scorer
Last synced: 3 days ago
JSON representation
Single-file Node.js script that measures how AI-friendly a docs site is.
- Host: GitHub
- URL: https://github.com/jamdesk/docs-ai-scorer
- Owner: jamdesk
- License: mit
- Created: 2026-05-13T19:27:13.000Z (about 1 month ago)
- Default Branch: main
- Last Pushed: 2026-06-01T20:28:05.000Z (14 days ago)
- Last Synced: 2026-06-01T22:17:59.754Z (14 days ago)
- Language: JavaScript
- Size: 9.77 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# 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?](https://www.jamdesk.com/blog/ai-friendly-docs-platforms-scored) on the Jamdesk blog.
## Run it
```bash
node score-docs.mjs
```
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
```bash
./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](../../issues/new?template=submit-results.yml) 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](https://llmstxt.org) 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](https://www.jamdesk.com/blog/ai-friendly-docs-platforms-scored).
## License
MIT. © Jamdesk.