Tech entrepreneur Siqi Chen released an open-source plugin for Anthropic's Claude Code that instructs the AI to avoid common AI writing patterns. The tool, called Humanizer, uses a list of 24 language tells compiled by Wikipedia's AI Cleanup volunteers. Chen published it on GitHub, where it quickly gained over 1,600 stars.

The source material is a guide from WikiProject AI Cleanup, a group of editors who have been tagging AI-generated articles since late 2023. Their list includes inflated phrases like "marking a pivotal moment" and overuse of "-ing" clauses. Chen's plugin is a skill file for Claude Code, which adds these instructions to the model's prompt with more precision than a standard system prompt.

While the tool can make AI output sound more casual, it has notable drawbacks. It won't improve factuality and might harm coding ability or technical documentation. The project highlights a broader irony: the web's most referenced rules for detecting AI writing are now being used to subvert detection. This underscores the ongoing cat-and-mouse game between AI generators and detectors.