The detectors do not work well enough to optimise against
Published false-positive rates on human-written text are high enough that respected institutions have stopped using them for grading. Optimising your copy against a noisy classifier means making changes that have no relationship to whether the writing is good.
What readers actually notice
Generic openings, the absence of specifics, and sentences that would be equally true of any competitor. None of those are detection problems. They are drafting problems, and they existed long before generated text did.
The fix is boring
Add the thing only you know: the number, the customer quote, the mistake you made. Cut the first paragraph. Name the audience precisely in the brief. This is the same editing that improved human first drafts, applied to a first draft that arrived faster.