A second opinion, not a verdict — using a detector in your edit pass
The most useful thing a detector gives an editor usually isn’t the score. It’s the margin notes.
Run it after your first pass, not instead of one
A detector is cheapest to use as a second read on a draft you’ve already edited once — by then you have your own sense of the piece, and the report either confirms it or points somewhere you missed.
The two signals worth reading regardless of score
Stock transitions. “Moreover,” “it is important to note,” “in conclusion” — connective filler you’d cut in any edit, whoever wrote it. If the report flags these, that’s a tightening note, not an accusation.
Repeated three-word phrases. The same short phrase turning up more than once in a piece is a copy-edit flag independent of where it came from. Worth fixing either way.
Where the score itself is less useful
Heavily edited AI output lands in the middle of the range or lower — a borderline score tells you little on its own. And formulaic house styles (press releases, structured reviews, templated product copy) can read as templated to any statistical detector without a word of AI involvement. Treat a mid-range score as “worth a closer read,” not as a finding.
Write the policy down before you need it
Decide what counts as acceptable AI use for your publication before a submission forces the question, and tell contributors. A score is only useful measured against a policy someone agreed to.