Simon Willison built an LLM cliché highlighter tool to identify AI-generated writing patterns

Main idea

LLM-generated text has recognizable stylistic fingerprints — recurring phrases and patterns that are becoming increasingly common and predictable. Willison built a practical tool to detect and highlight them.

Context

Willison is one of the most prolific AI bloggers and tool builders (author of Datasette, LLM CLI). As more content is AI-generated, the linguistic fingerprints of models are becoming easily identifiable to experienced readers. The tool runs entirely in the browser without a server, with localStorage support.

Why it matters

This is a small but significant cultural artifact of the era — when AI writing becomes sufficiently homogeneous to need a 'detection tool'. It could become a reference resource for editors, educators, and researchers studying AI content.

Details / arguments

• 10 detected patterns: 'no X, no Y' chains, phrases like 'sit with that', and others
• Runs purely in the browser (no server), localStorage for persistence
• Users can toggle individual patterns on/off and see counts
• Published at simonwillison.net/2026/Jul/17/llm-cliche-highlighter/
• Willison categorized it as a practical response to the 'AI slop' problem