When your own edit pass triggers the flag
The better an edit, sometimes, the more uniform the result — and uniform is exactly what a statistical detector is built to notice.
Why a good edit can raise a score
Evening out sentence length, cutting stray tangents, and standardizing terminology — all normal, good editing — reduce exactly the variation a detector reads as human. A well-edited piece can end up statistically closer to generated text than the rougher draft it started as.
This is a real limitation, not a loophole to exploit
It means a score is not a reliable signal on heavily edited work either direction — not because editing is suspicious, but because the detector cannot distinguish a good edit from generation using rhythm alone. That is worth knowing, not worth hiding from a client who asks.
What to check before you assume the writer used a tool
Ask for the pre-edit draft. A rougher, more uneven original that you smoothed into a tidier final version is a completely ordinary editorial process, and the draft itself is the evidence, not the final score.
Say this to a client upfront
If you do heavy structural edits as part of your process, telling a client in advance that this can affect a detector score if they ever run one avoids a confusing conversation after the fact.