How to Write with an LLM

This lays out a practical workflow for using large language models to improve prose while preserving voice: write the draft yourself, then use an LLM strictly as a copyeditor. Two hard rules prevent the model from hollowing out your voice: never adopt any specific turn of phrase an LLM suggests, and forbid encouragement or praise from the model that would validate first-draft impulses. Models are superb at producing polished, headline-like language and at reassuring you, both of which nudge writers into an uncanny, artificially flavored middle ground. Be strict: don’t let suggested wording or flattering feedback replace the hard rethinking that forms your voice.

Use models for what they do best: spotting mechanical and structural problems - excess passive voice, nominalizations, repeated words and filler adverbs, and paragraphs that belong elsewhere - then run iterative editing passes. Read a style manual (the author recommends Style: Lessons in Clarity and Grace), create a checklist of editing prompts, and feed those prompts to the model in discrete passes. Compare original and rewritten options using a model that doesn’t know your edit history, or build a small workshopping tool to manage passes. Finally, accept only the edits that serve your voice; decline wholesale trimming or smoothing even when the model insists it’s “better.”