There’s a failure mode that traps most authors who try AI editing.
They paste a chapter in and type: “Edit this.”
What comes back is confident, plausible, and mostly useless. The AI performs edits to justify its existence – smoothing things that didn’t need smoothing, proposing changes that technically work but make nothing better. It’s a vending machine that dispenses the appearance of helpfulness.
“Edit my book” gives you the same results as “Write me a book.”
Precision is the only way in. And precision requires knowing what editing actually is – which most authors, and every AI prompting guide I’ve ever read, skips entirely.
So I did what an engineer does with a system he doesn’t understand. I read the manuals.
The Research
Three books. Not craft books – editing books.
The Artful Edit by Susan Bell. Dreyer’s English by Benjamin Dreyer. The Copyeditor’s Handbook by Amy Einsohn.
I read them like API documentation. I was looking for specific, nameable operations – not taste, not instinct, not vibes. Discrete editorial activities precise enough for a language model to execute without hallucinating its way toward “helpfulness.”
What they revealed: editing has distinct layers that should never run simultaneously. Mixing them is why AI editing fails. It’s also why human editors charge differently for developmental work versus copyediting. The tasks are not the same thing. They don’t even ask the same questions.
