I think AI feedback is most useful when I already have an outline or a draft to examine. Asking it to identify an unclear claim or a gap in the explanation gives me something specific to consider. Asking it to rewrite the whole essay makes it harder to understand which decisions actually improved the text.
When choosing resources for student writers, I look for support that helps the writer understand a problem and decide how to address it. For example, if a paragraph contains evidence but does not explain how it supports the thesis statement, a useful response would point out that missing connection. The student can then return to the source and develop the explanation.
I would also keep source checking separate from language feedback. A paragraph can sound convincing while misrepresenting a reading, and a correctly formatted citation still needs verification. I would check the original material before accepting any suggestion that changes a factual claim.
One approach I find useful is to ask for questions rather than replacement text: “Where would a reader need more explanation?” or “Which claim in this outline needs stronger evidence?” That leaves the revision with the writer and makes the feedback easier to evaluate.
For those who use AI in teaching or studying, what works better for you: feedback on a complete draft, or smaller questions during planning and revision? How do you decide which suggestions to accept?