Slop-Radar: A Buzzword Detector for the Age of AI Slop

You've read it. You've probably written it. That peculiar style of text that screams "an AI wrote this": the em-dashes everywhere, the "it's worth noting that", the "let's delve into", the "in today's rapidly evolving landscape".

I call it slop. And I was producing it.

I run a blog generation pipeline for the Transceiver Intelligence Platform. AI-assisted writing, human-reviewed. And I noticed that even after review, the published posts had a certain... sameness. They were technically correct but felt like they were written by a committee of cautious middle managers.

I needed a tool that could score text for AI-style writing patterns. Not to detect AI authorship — that's a different problem — but to flag lazy, generic, filler-heavy writing regardless of who (or what) produced it.

// how it works

slop-radar does two things: buzzword detection and structural analysis.

Buzzword detection catches phrases like "leverage synergies", "it's important to note", "at its core", "navigate the complexities". The database is plain JSON — easy to contribute to and extend. Fuzzy matching catches variations so "delving into" and "let's delve deeper" both trigger.

Structural analysis identifies 14 patterns: em-dash overuse, passive voice density, artificial openers ("In today's..."), hedge phrase clustering, list-heavy formatting, and more. These are the fingerprints of text that was generated rather than written.

// the scoring system

Every text gets a score from 0 to 100:

• 0–20: HUMAN — natural, authentic writing
• 21–40: MOSTLY CLEAN — minor patterns detected
• 41–60: SUSPICIOUS — noticeable AI-style patterns
• 61–80: SLOPPY — heavy use of generic filler
• 81–100: PURE SLOP — reads like unedited ChatGPT output

// integrated into my blog pipeline

Every blog post I publish goes through slop-radar before publication. It's a Claude Code skill — I can run /slop-check on any text and get instant feedback. Anything above 40 gets rewritten. The TIP blog generation pipeline runs it automatically and rejects drafts that score too high.

It's also available as a CLI tool for quick checks and as a Node.js library for CI/CD integration. Some teams use it as a pre-commit hook for documentation.

The German buzzword database is something I'm particularly proud of. German AI slop has its own flavor — "Es ist wichtig zu beachten", "In der heutigen dynamischen Welt" — and nobody else was detecting it.

npx slop-radar check article.md — GitHub