How Slop Detector compares
Where Slop Detector sits among design-quality and AI-detection tools — and why its methodology is different.
The short version
Most "is this AI?" tools are probabilistic ML classifiers: they output a confidence that something was machine-generated, and that confidence shifts as the model is retrained. Slop Detector takes the opposite approach. It is a deterministic fingerprint: a fixed catalogue of CSS and copy tells, each with a fixed weight, evaluated against a page's live computed styles in real headless Chromium. The same page always yields the same score. There is no model and no randomness — so the output is auditable, reproducible, and safe to gate CI on.
Methodology comparison
| Dimension | Slop Detector | Typical ML AI-detectors | Generic Lighthouse / a11y tools |
|---|---|---|---|
| Output | Deterministic 0–100 weighted fingerprint | Probabilistic % confidence | Perf / a11y / SEO scores |
| Reproducible | Yes — same page, same score | No — drifts with retraining | Mostly |
| What it measures | AI-design + copy slop tells | Generated-vs-human likelihood | Technical quality, not aesthetics |
| Engine | Real Chromium, computed styles | Text / pixel models | Real Chromium |
| Auditable rules | Yes — open catalogue at /api/patterns | Opaque weights | Documented audits |
| AEO axis (AI-readability) | Yes — built in | No | Partial (SEO only) |
| License | Open source (MIT) | Usually closed | Mixed |
| Agent interfaces | API + CLI + MCP | Rare | Rare |
Why deterministic matters
- CI gating. You can fail a build at
--fail-on heavyand trust the threshold won't move under you. - Auditability. Every point in a score traces to a named pattern with a documented weight.
- Comparability. Two designs can be compared on the same fixed yardstick, today and next year.
- No false-confidence. A high score says "this looks machine-made," not "an AI wrote this" — it's a fingerprint, not a verdict.
The research behind it
The pattern weights derive from Adrian Krebs's April 2026 study
of 1,400 Show HN submissions, plus Meng To's
gradient-avatar tell. The catalogue is versioned (DEFINITIONS_VERSION) and served
live at /api/patterns.
When to use which
- Use Slop Detector when you want a reproducible, explainable read on how templated/AI-generated a landing page looks, or to gate design regressions in CI.
- Use an ML AI-detector when you need a probabilistic "was this generated?" verdict on arbitrary text/images.
- Use Lighthouse/axe for performance, accessibility, and technical SEO — orthogonal to slop.
Try it
Web: slop-detect.com ·
CLI: npx slop-detect <url> ·
MCP: slop-detect-mcp ·
API: /openapi.json ·
Source: GitHub