Thomas Ptacek on using LLMs as copyeditors, not ghostwriters
A crisp practical rule-set for preserving voice while still using models to do the exhausting mechanical parts of editing.
Links: Original source · Shared link · Related link 2
Logged at IST: 2026-09-18 02:43 IST
What it is: Thomas Ptacek's essay on how to use LLMs for writing without letting them flatten your prose into model-output voice.
Gist: Ptacek's two rules are blunt: do not use a single word the LLM suggests, and avoid its encouragement. His argument is that readers can smell LLM-selected phrases, and that model praise is especially dangerous because it nudges writers to preserve weak first-draft instincts instead of doing the voice-forming work of rethinking, cutting, and replacing paragraphs.
The useful workflow is to write the piece yourself, then use models as tireless copyeditors. Ask them to flag passive voice, nominalizations, filler words, repeated phrases, bad paragraph order, and other mechanical problems. Rewrite the suspect passages yourself, then compare old and new versions with a model that has not seen your edit history. Ptacek also points to Style: Lessons in Clarity and Grace as the programmer-friendly schematic for this kind of revision.
Newsletter angle: A practical taste-preservation pattern for AI-assisted writing: outsource detection and comparison, not word choice or confidence.