Use AI as a copyeditor, not a ghostwriter

Readers spot AI-generated text in parts per trillion, but language models still make sharp editors if you handle them right.

Writing on sockpuppet.org in a post that hit 619 points on Hacker News, the author outlined a practical method for writing with LLMs. The core idea: write your own draft first, then use a model strictly as a copyeditor to catch your mistakes.

Why it matters: Frontier models naturally write like magazine headlines, which flattens your voice into generic output. Using a model only to audit your text lets you catch structural weak spots and repetitive filler without handing over the keyboard.

Know this: Two rules keep the AI from ruining your work:

• Reject all suggested phrasing. Any specific turn of phrase the LLM offers is completely off-limits.
• Forbid encouragement. Tell the model to skip the praise, which only tricks you into keeping bad first drafts.

Use clean model sessions without editing context to evaluate your original paragraphs against your own rewrites. For a solid offline guide to prose mechanics, the post points to the book Style: Lessons in Clarity and Grace.

Let the model spot your weak sentences, but keep your hands on the keyboard.