Nearly every flagged essay I have looked at has the same shape: the introduction is fine, the middle is fine, and the conclusion lights up. There is a structural reason. A conclusion is written to summarise, so it repeats the argument in the same formal register, and the sentences come out the same length. Uniform, abstract and predictable is the exact profile a detector is trained to notice.
The fix is not to make the conclusion vague. It is to make it specific again. Name the concrete finding, then say what it changes. If the essay argued that a policy failed for a particular reason, the conclusion should say that reason once more in a sentence only this essay could write.
Read it aloud afterwards. If it sounds like a summary of a summary, cut the first sentence and start from the finding. More detail in this guide to humanizing AI text and a free checker is at the AI detector page.