A tool to score any website for common AI design patterns. Basis for the blog post and the HN discussion "Scoring Show HN submissions for AI design patterns".
The scorer loads each site in a headless browser, analyzes the DOM, and reports which of 16 deterministic AI design patterns are found. Manual verification across ~150 labeled sites suggests ~5–10% false positives. Still, take it with a grain of salt :)
Install & run
git clone https://github.com/AdrianKrebs/ai-design-checker cd ai-design-checker && npm install node cli.js https://news.ycombinator.com/
https://news.ycombinator.com/
Clean · score 0/100 · 0/16 patterns
https://engagemii.com/aeo
Heavy · score 38/100 · 6/16 patterns
Triggered:
• Vibe purple
• Gradients
• Perma dark
• 1·2·3 steps
• Eyebrow pill
• FAQ
--json for machine-readable output. Requires Node 18+.
To update an existing clone, git pull from inside the directory.
Run the full corpus
For batch analysis (e.g. the bundled urls.txt with ~1000 Show HN posts) — same install as above, then:
npm run analyze # sequential, ~100 min for 1k URLs node src/run.js --concurrency=4 # 4 parallel, ~25 min, ~600 MB RAM node src/run.js --skip-existing # only fetch URLs not yet cached
Results go to results/:
results/raw/<slug>.json— per-URL signals + scoreresults/screenshots/<slug>.png— full-page screenshotresults/all-results.json— all scored entries combined
Patterns
| # | Pattern | Tell |
|---|---|---|
| 1 | Templated fonts | Space Grotesk, Instrument Serif, Geist, Syne — fine in isolation, default-stack tell |
| 2 | Centered hero | H1 centered + Inter as primary font |
| 3 | VibeCode purple | Filled violet/indigo CTAs |
| 4 | Perma dark | Dark template, or muted grey body text on dark |
| 5 | All-caps headings | ≥ 3 uppercase H-tags |
| 6 | Gradients | linear/radial/conic-gradient + background-clip: text |
| 7 | Colored glow | Saturated box-shadow blur ≥ 15 px |
| 8 | Eyebrow pill | Small badge above the H1 |
| 9 | Accent stripe | Colored left/top stripe on a card with a heading |
| 10 | Icon-card grid | 3–12 cards: icon-in-badge + title + blurb |
| 11 | Numbered steps | "1·2·3" digits in styled badges |
| 12 | Stat banner | "10K+ users · 99.9% uptime · 4.9★" |
| 13 | FAQ accordion | "Frequently asked questions" with ≥ 3 Q/A items |
| 14 | Emoji nav | ≥ 40 % of nav links prefixed with an emoji |
| 15 | shadcn signature | CSS vars (--background), data-slot, Lucide, Radix attrs |
| 16 | Glassmorphism | backdrop-filter: blur(...) on a translucent panel |
The full rule for each pattern lives in src/patterns/<id>.js.
Score
score = round(100 × patternsFlagged / patternsTotal)
tier = ≥5 Heavy · 2–4 Mild · 0–1 Clean
Tools
npm run scan # → http://localhost:7788 paste a URL, see verdict npm run label # → http://localhost:7777 label sites for ground truth npm run eval # precision / recall vs dataset/labels.jsonl (165 labels shipped) npm run report # generate results/index.html — browsable tier-filtered grid npm run fetch # pull the latest 100 Show HN URLs into urls.txt
The scan UI launches a real browser per URL, so the first scan after starting the server is slower. The label UI lets you mark each pattern present / not_present / skip; saves are appended to dataset/labels.jsonl. The eval script compares the detector's verdict against those labels.
Adding a pattern
Drop a file in src/patterns/<id>.js and append it to the list in src/patterns/index.js. Each pattern exports { id, label, shortLabel, description, category, thresholds, extract(ctx), score(signal, T) }. extract is serialized via Function.prototype.toString() and runs in the browser — keep it self-contained (only reference ctx.*, no closures, no imports). score runs in Node and returns { triggered: true/false, evidence: ... }.
License
MIT.