Writers routinely pass a draft through one tool for structure and another for polish. The second pass rarely improves on the first in the way the sequence suggests.
Each pass rewrites toward its own defaults
A model asked to edit produces text resembling what it would have produced from scratch. Editing and generating are not separate operations inside the system.
So the second tool does not refine the first tool's choices. It substitutes its own sentence rhythms, transition habits and paragraph lengths for whatever it received.
The deliberate decisions from the first pass disappear silently, because nothing marks them as decisions rather than as raw text to be improved.
Specificity is the first casualty
Concrete detail carries no signal that it matters. A street name, an unusual figure or a deliberately blunt sentence looks to the second pass like something to smooth.
Rewriting tends toward the middle of the distribution, which means the distinctive phrase gets replaced by the common one and a precise claim gets softened into a general one.
Two passes therefore produce prose that is more fluent and less informative than one, and the loss is difficult to notice when reading only the output.
Instructions do not survive the handoff
Constraints given to the first tool exist only in that conversation. The second tool receives a finished document with no record of what it was supposed to preserve.
A rule such as keeping the opening claim unhedged has to be restated, and restating it in the abstract works less reliably than it did during original drafting.
Writers often discover the violation only at the final read, after the sentence they cared about has been replaced by a reasonable-sounding equivalent.
The pattern repeats at the team level
In American workplaces the sequence is frequently organizational rather than chosen. A contractor drafts with one tool, an in-house editor cleans up with another, marketing runs a third pass.
Each participant improves the document by their own measure while the accumulated effect is homogenization. Nobody sees the full sequence, so nobody observes the drift.
Documents that pass through several such stages arrive sounding interchangeable with every other document that took a similar route.
One pass plus human editing works better
The practical alternative is to use a generative pass once, at the stage where it adds the most, and to do the remaining work by hand.
Human editing preserves intent because the editor knows which choices were deliberate. That knowledge is exactly what the second model lacks and cannot infer.
The sequence that fails is not two tools but two rewrites, and a rewrite by a person who understands the piece is a different operation entirely.