Simon Willison Releases LLM Cliché Highlighter to Detect Robotic Writing Patterns
Simon Willison has launched a browser-based utility called LLM Cliché Highlighter. It automatically scans pasted text or loaded URLs and highlights classic AI clichés and repetitive phrases in real-time.
Impact: Low
Why it matters
Developers and technical writers can immediately clean up draft copy of robotic phrases before pushing content to public documentation or repositories.
TL;DR
- 01Detects robotic LLM stylistic habits like 'delve' and repetitive 'no X, no Y' structures.
- 02Operates fully in the browser as a static web tool with real-time text analysis.
- 03Helps refine system prompts by pinpointing exactly which clichés models generate.
Key facts
- Deployment Model
- Static HTML/JS Page
- Cost
- Free
- Developer
- Simon Willison
Real-Time AI Linter
The LLM Cliché Highlighter is an open-source static web tool that processes strings as they are typed or fetched directly from a target URL. Operating fully inside the browser, it requires no installation or server resources.
Structural Chain Detection
Unlike simple key-value word finders, the script targets complex grammatical structures. Sentence configurations like negative list structures get dynamic badges counting the repetition level.
Improving Prompts and Text
Hovering over flagged sentences guides users through the specific rule violation. Technical authors and developers can leverage these flags to identify patterns their generative setups are outputting, helping refine custom prompt rules to discourage models from returning specific clichés in the future.
✓ When to use
- When cleaning up generative AI-produced technical blog posts, project readmes, or customer-facing documentation.
✕ When NOT to use
- For analyzing executable programming code or markdown formatting logic.
What to do today
- Bookmark tools.simonwillison.net/llm-cliche-highlighter to analyze AI-generated technical documentation draft copies.
- Incorporate discovered cliché phrases into your negative system prompts to discourage models from utilizing them.
Sources