A new online resource called Claudisms has emerged, positioning itself as a "living banlist" for the repetitive linguistic patterns often found in AI-generated text. The site, hosted at claudisms.ai, tracks specific phrases and sentence structures that have become synonymous with large language model outputs. By cataloging these "tells," the project aims to provide a practical tool for developers, writers, and editors who want to identify or eliminate synthetic-sounding prose from their workflows.

The Problem with AI Homogenization

As AI integration becomes ubiquitous in software development and content creation, a distinct stylistic fingerprint has emerged. This fingerprint consists of overused transition words, specific adjectives, and sentence structures that lack human nuance. Claudisms addresses this by maintaining a dynamic list of these patterns. While the Hacker News submission is recent and currently has low engagement with only 2 points and no comments, the underlying issue it highlights is significant for the dev community. The homogenization of text makes it harder to distinguish between human-authored documentation, code comments, and AI-assisted drafts.

A Tool for the Builder

For developers building applications that rely on LLMs for content generation, identifying these tells is crucial for quality control. Claudisms serves as a reference point for creating custom linters or post-processing scripts that can flag or rewrite these common AI phrases. The "living" nature of the banlist suggests that as models evolve and change their output styles, the list will update accordingly. This adaptability is key, as static lists quickly become obsolete in the fast-moving AI landscape. The project encourages community contribution, ensuring that the banlist stays current with the latest model behaviors.

Key Takeaways

  • Claudisms.ai catalogs repetitive AI writing patterns to help identify synthetic text.
  • The list is described as "living," implying regular updates to match evolving LLM outputs.
  • Developers can use this resource to build better post-processing tools for AI-generated content.
  • The project highlights the growing need for stylistic diversity in AI-assisted workflows.

The Bottom Line

If your AI-generated docs sound like they were written by a robot that just discovered adjectives, check this list before you ship.