AI doesn't save the work. It moves it downstream.
Generation gets faster; verification, re-voicing and restructuring get bigger. Researchers call it the efficiency paradox. Knowing where the work went is how you stop losing to it.
Everything in this pillar
Cutting filler, fixing rhythm, and getting a generic draft to sound like a person wrote it.
The first guides in this pillar are being written now.
What editing an AI draft actually involves
Qualitative research on creators working with AI identifies four recurring forms of repair. Skipping any one of them is what produces the texture people call AI slop.
-
01
Verify the claims
Every fact, number and reference the model supplied gets checked against something real. This is the repair people skip most and regret hardest.
-
02
Re-voice the language
Replace the model's register with your own. Not synonym swapping — reading aloud and rewriting anything that would not come out of your mouth.
-
03
Restructure the argument
Models produce locally coherent, globally flat text. The order that serves your point is usually not the order it generated.
-
04
Re-perform the authority
Put back what the model cannot have: your specific experience, the example only you saw, the opinion you are willing to defend.
Questions people actually ask
Why does AI writing feel flat even when nothing is wrong with it?
Because correctness and voice are different properties. A model optimises for plausible continuation, which produces prose that is locally fine and globally characterless. Research on AI-assisted creative work finds that generated material still needs deliberate filtering, recombination and stylistic refinement before it constitutes usable work.
Is it faster to edit an AI draft or write it myself?
It depends on which repair dominates. If the piece is mostly structure and language, editing wins. If it is mostly claims that need verifying, you may spend more time checking the draft than you would have spent writing it — and that is the efficiency paradox in one sentence.
How much editing is enough?
A practical test: could a reader who knows your writing tell this piece is yours? Studies of creators working with AI describe exactly this as invisible authenticity labour — the work of making assisted output plausibly your own. If that work has not happened, the draft is not finished.