Generating a draft takes seconds and editing it frequently takes longer than writing from scratch would have. The reasons sit in what the draft supplies and what it conceals.
Why the draft looks finished
Generated prose arrives with correct grammar, consistent formatting and even paragraph lengths. It has the surface properties of finished work.
A human first draft has none of those. It has gaps, notes to self, half-formed sentences and obvious weak spots, and every one of those is a signal about where work remains.
The generated draft suppresses those signals. Weak sections look identical to strong ones, because the fluency is uniform regardless of underlying quality.
Editors are trained to read roughness as a cue, and that instrument stops working on text with no roughness in it.
So the editor must evaluate every sentence deliberately rather than being guided to the problems.
The cost of verifying claims
Every factual assertion in a generated draft requires checking, and checking is slower than writing something you already knew.
A writer working from knowledge produces claims they can defend. A writer editing generated text inherits claims of unknown provenance, each needing independent confirmation.
The claims are stated with uniform confidence, so there is no gradient to prioritise by. A well-known fact and an invented one look the same on the page.
Numbers, dates, names and attributions are the expensive categories, and they are exactly what a model produces most unreliably.
Verification frequently consumes more time than the entire generation step saved.
Why structure is the expensive repair
Sentence-level problems are cheap to fix. Structural problems are not, and generated drafts fail structurally more often than they fail locally.
Typical structural faults include sections in an order that does not build, an argument that states its conclusion before establishing it, or three sections making the same point in different vocabulary.
Fixing structure means moving material, which breaks transitions, creates repetition and requires rewriting the connective sentences throughout.
At that point the editor is doing the work of composition without the benefit of having planned the piece, which is harder than planning it originally.
Structure is decided before writing when a person drafts, and it is decided by the model when generation drafts.
How fluent error hides
Awkward writing signals uncertainty and draws attention. Smooth writing signals confidence and deflects it.
A wrong claim in a well-formed sentence surrounded by correct claims is genuinely difficult to notice, because reading comprehension relies partly on flow.
The effect strengthens with familiarity. An editor who has read three similar generated drafts skims the fourth, and skimming is where fluent errors survive.
Reading aloud, or checking claims in isolation out of order, defeats the effect at the cost of speed.
Teams that adopt those habits report catching substantially more, and they also report that editing is slower than they expected it to be.
Why small edits do not compound
Improving individual sentences in a weak draft produces a weak draft made of better sentences.
Editors often begin at the top and work down, fixing what they encounter, which is efficient on a draft with sound structure and wasteful on one without.
Hours can go into polishing a section that a structural revision would have deleted entirely.
The discipline that helps is a structural read first, with no sentence-level editing permitted until the outline is agreed.
Most editors know this and abandon it under time pressure, because generated text invites line editing by looking nearly ready.
The rewrite threshold
Beyond a certain level of damage, rewriting is faster than repairing, and recognising that point early is the main skill.
Useful indicators are whether the draft has a defensible thesis, whether its sections are in a defensible order, and whether the specifics are ones you can verify.
Two failures out of three usually means the draft is a source of raw material rather than a document to edit.
Sunk cost works against this judgement. Time already spent editing feels like a reason to continue, and it is not.
The generation was cheap, which means discarding it costs almost nothing compared with the editing hours it would otherwise consume.
How briefs shift the work earlier
Most editing burden originates in an underspecified request, and specifying it properly moves the same work to a cheaper stage.
A brief that names the audience, the argument, the sections, the evidence available and the material to exclude produces a draft that needs local editing rather than reconstruction.
Writing that brief takes real time and feels like overhead, particularly since generation appears to remove the need for planning.
The comparison that matters is brief-writing time against editing time saved, and for anything that will be published the trade is usually favourable.
Teams that produce many similar pieces get the largest benefit, because one good brief is reused across dozens of drafts.
Where the time saving is real
Generation is genuinely fast for work with low verification cost and low structural risk.
Reformatting existing text, producing variants of an approved message, drafting routine correspondence and expanding an outline the writer already made are all cases where the model supplies labour rather than judgement.
It is also strong at the blank page problem, producing something to react against, which many writers find faster than starting cold even when they keep none of the words.
The saving shrinks as the required accuracy, originality and structural quality rise, and it can go negative on work that must be correct.
Estimating which category a task falls into before starting is what separates teams that save time from teams that report being busier than before.