🚀 Live Demo — Try it in your browser, no installation needed.
Detect AI slop in text. A living buzzword database, structural pattern matcher, and scoring engine -- available as CLI tool, Node.js library, and Claude Code skill.
🚀 Live Demo — Try it in your browser, no installation needed.
Slop-radar scans text for:
- 200+ AI buzzwords (English) and 120+ German AI phrases
- 14 structural patterns like em-dash abuse, "Let me" starters, bullet overload, and passive voice density
- Fuzzy matching to catch variations and reformulations
It returns a score from 0 to 100:
| Score | Rating | Meaning |
|---|---|---|
| 90-100 | HUMAN | Clean, natural writing |
| 70-89 | MOSTLY CLEAN | Minor AI signals |
| 50-69 | SUSPICIOUS | Multiple AI patterns found |
| 30-49 | LIKELY AI | Strong AI writing signals |
| 0-29 | PURE SLOP | Heavy buzzword and pattern use |
# Check a file
npx slop-radar check essay.md
# Pipe text
echo "Let me dive deep into this transformative journey" | npx slop-radar
# Score only
npx slop-radar score article.txt
# JSON output (for CI/scripts)
npx slop-radar json draft.mdnpm install -g slop-radarOr use directly with npx (no install needed).
slop-radar check <file> Full analysis with color output
slop-radar score <file> Score + rating badge only
slop-radar json <file> Machine-readable JSON
cat text.md | slop-radar Pipe text for analysis
--lang en|de|auto Force language detection (default: auto)
--help Show help
--version Show version
import { detect, score, formatFull } from "slop-radar";
const text = "This transformative journey leverages cutting-edge innovation.";
const detection = detect(text, "en");
const result = score(detection);
console.log(result.score); // 88
console.log(result.rating); // "MOSTLY CLEAN"Starting score: 100
Deductions:
- -2 per buzzword/phrase found
- -weight per structural pattern match (weight varies 1-5)
- -10 for passive voice density above 30%
- -3 per "Let me" / "Here's the thing" opener
Bonuses:
- +5 if text contains questions
- +5 if sentence lengths vary naturally
Score is clamped to 0-100.
Input:
Let me dive deep into this transformative journey. Here's the thing -- in today's fast-paced landscape, it's worth noting that leveraging cutting-edge solutions is crucial. Moreover, this holistic approach empowers stakeholders to unlock unprecedented synergy.
Output:
Score: 28/100 PURE SLOP
Buzzwords found: 12
"dive deep", "transformative", "journey", "in today's fast-paced",
"landscape", "it's worth noting", "leveraging", "cutting-edge",
"crucial", "moreover", "holistic", "empower", "stakeholders",
"unlock", "unprecedented", "synergy"
Patterns detected: 3
let-me-starter, heres-the-thing, worth-noting
Cleaned version:
How do we make our product development faster? We found that using modern tools cut our deployment time by 40%. The team now ships weekly instead of monthly, and customer complaints dropped.
Score: 95/100 HUMAN -- specific, concrete, no filler.
The phrase databases are plain JSON arrays in src/database/:
phrases-en.json-- 200+ English AI phrasesphrases-de.json-- 120+ German AI phrasespatterns.json-- 14 structural patterns with regex
Edit the JSON files and submit a PR. See CONTRIBUTING.md for guidelines on what qualifies as an AI phrase.
Drop skill/SKILL.md into your .claude/skills/ directory to use slop-radar as a Claude Code slash command. See the skill file for details.
AI-generated text has a recognizable style: filler words, hedge phrases, artificial enthusiasm, and predictable structure. Slop-radar does not detect AI authorship (that is a harder, different problem). Instead, it detects AI-style writing -- the buzzwords and patterns that make text feel generic and low-effort, regardless of who wrote it.
Use it to:
- Clean up your own writing
- Audit content before publishing
- Score AI drafts and iterate until they sound human
- Enforce quality standards in a CI pipeline
- claude-cortex — Persistent memory for Claude Code sessions.
- claude-sync — Multi-device sync for Claude Code.
MIT