A comprehensive catalog of AI writing patterns has been compiled to help developers detect and avoid machine-generated text. The single-file markdown document, titled 'LLM Writing Tropes.md', contains hundreds of overused phrases and structural patterns commonly found in AI-generated content. Created with AI assistance, the file serves as a system prompt reference for anyone building AI writing tools.

The catalog identifies specific word choices that AI models overuse, including 'quietly', 'delve', 'tapestry', and 'landscape'. These terms appear disproportionately in AI text compared to human writing. The document also catalogs structural patterns like negative parallelism ('It's not X -- it's Y'), tricolon abuse, and the 'The X? A Y.' rhetorical question format. Many of these patterns emerged from reinforcement learning with human feedback training.

Developers can add this file to their AI's system prompt to help models avoid these telltale patterns. The creator notes that AI models tend to repeat these structures because they're rewarded for 'readable' writing that breaks complex ideas into simple fragments. By flagging these patterns, developers can create more natural-sounding AI text that doesn't immediately signal its machine origin. The file represents a practical tool for improving AI writing quality and avoiding the most obvious machine-generated text giveaways.