Thomas Ptacek proposes a strict rule for using large language models as copyeditors rather than writing assistants. He states that writers must never use a single word or phrase suggested by the AI. This approach acts as intellectual personal protective equipment against the distinctive, often uncanny texture of machine-generated prose. Ptacek admits he does not allow these tools to draft content for his own blog. Instead, he employs them for fact-checking, spelling, grammar, and occasional synonym searches. He finds the restriction of using only original phrasing helpful for maintaining discipline in the writing process.
The core value lies in preventing the subtle drift of voice that occurs when relying on algorithmic suggestions for specific turns of phrase. By forcing the writer to construct every sentence independently, the final text retains a consistent human fingerprint. This method also encourages deeper engagement with the material, as the author cannot simply accept the first generated option.
* Ptacek shares a screenshot of his personal LLM copyediting tool.
* He provides a prompt to help others build similar proofreading systems.
* The rule applies strictly to any specific turn of phrase the model suggests.
