How To Write With An LLM
Use LLMs as tireless copyeditors to flag mechanical flaws in your drafts, but never adopt their phrasing or accept their praise, so your authentic voice survives while tedious editing work is automated.
Key Points
- Two rules prevent LLM pasteurization: never use a word the model suggests, and never accept its encouragement — both flatten voice into generic output.
- LLMs excel at mechanical copyediting: flagging passive voice, nominalized verbs, filler words (very, unfortunately, really, actually), and paragraphs that improve clarity when moved.
- Reference book “Style: Lessons In Clarity And Grace” turns copyediting into systematic, repeatable passes — comparable to coding refactoring.
- Three-step workflow: (1) ask model to spot problems, (2) rewrite flagged sections yourself, (3) present original and rewrite to a fresh model context for unbiased comparison.
- Author built a local writing workshop tool (Python, HTMX, SQLite, Tailwind, Notion-style editor) to manage editing passes, sidebar commentary, and revision tracking.
- Final corollary: don’t take all copyediting advice — the model may suggest cutting 20% of your piece, but preserving your voice sometimes means keeping the ‘inefficient’ parts.