How to Write with an LLM: Never Use Its Words, Never Trust Its Praise

How to Write with an LLM

Thomas Ptacek argues that LLMs work best as copyeditors, not ghostwriters. His two rules: never use a single word the model suggests—frontier models turn everything into magazine headlines—and avoid encouragement, because praise locks in bad first-draft instincts. The real value is mechanical flaw detection: passive voice, filler words, paragraph ordering. He recommends pairing the model with the book 'Style: Lessons in Clarity and Grace' and built a custom Python+HTMX workshop tool to run editing passes without the model knowing which version is the rewrite.

Why it matters: Thomas Ptacek is a well-known security researcher with a track record of sharp technical writing. The piece offers concrete, actionable rules rather than vague advice. Downside: it's a personal essay, not an industry event, and the full body isn't available—but the core argume...