A new tool aims to help AI chatbots create text that sounds more like it’s written by humans, using insights from Wikipedia’s guide for spotting AI-generated content, according to Ars Technica. Developer Siqi Chen designed the tool, called Humanizer, by providing Anthropic’s Claude with the list of indicators compiled by Wikipedia’s volunteer editors to tackle “poorly written AI-generated content.”
Wikipedia’s guide highlights various signs that suggest a text might be AI-generated, such as vague attributions, promotional language (like calling something “breathtaking”), and phrases that imply collaboration, such as “I hope this helps!” Humanizer, a custom skill for Claude Code, aims to help the AI assistant evade detection by eliminating these “signs of AI-generated writing” to make the text feel more natural and human, as noted on its GitHub page.
The GitHub page includes examples of how Humanizer can assist Claude in identifying and modifying these indicators. For instance, it might transform a description of a location from “nestled within the breathtaking region” to simply “a town in the Gonder region.” It also improves vague attributions by changing “Experts believe it plays a crucial role” to “according to a 2019 survey by…”
Chen mentions that the tool will “automatically push updates” whenever Wikipedia revises its AI-detection guide. It won’t be long before AI companies start adapting their chatbots to counter these indicators too, especially since OpenAI has already started addressing ChatGPT’s tendency to overuse em dashes, which has become a telltale sign of AI-generated content.
