# How Slop Detector compares

> Where Slop Detector sits among design-quality and AI-detection tools, and why its
> methodology is different. HTML version: https://slop-detect.com/compare

## 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 works the other way, as 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. Same page in, same score out. No
model, no randomness, auditable, reproducible, safe to gate CI on.

## Methodology comparison

| Dimension | Slop Detector | Typical ML AI-detectors | Generic Lighthouse / a11y |
|--------------------------|------------------------------------------|------------------------------|----------------------------|
| 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          |
| Engine                   | Real Chromium, computed styles           | Text / pixel models          | Real Chromium              |
| Auditable rules          | Yes, open catalogue at /api/patterns     | Opaque weights               | Documented audits          |
| AEO axis                 | 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: fail a build at `--fail-on heavy` and trust the threshold won't move.
- Auditability: every point traces to a named pattern with a documented weight.
- Comparability: two designs measured on the same fixed yardstick, today and next year.
- No false confidence: a high score means "looks machine-made," not "an AI wrote this."

## The research behind it

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 https://slop-detect.com/api/patterns

## When to use which

- Slop Detector: a reproducible, explainable read on how templated or AI-generated a landing page looks; CI gating.
- ML AI-detector: a probabilistic "was this generated?" verdict on arbitrary text or images.
- Lighthouse / axe: performance, accessibility, technical SEO. Orthogonal to slop.

## Try it

- Web: https://slop-detect.com
- CLI: `npx slop-detect <url>`
- MCP: `slop-detect-mcp`
- API: https://slop-detect.com/openapi.json
- Source: https://github.com/ravidsrk/slop-detect