PRISMA 2020 (Page et al., BMJ 2021;372:n71) is 27 items across seven sections, and the first thing to understand about it is what it is not: it is a reporting guideline, not a method. It tells you what must appear in the write-up. It does not tell you how to run the review.
That distinction is the useful one, because a reporting standard is a list of things you must be able to say afterwards — which makes it, read backwards, a specification for what your context has to have been carrying all along. You cannot report what you never recorded.
The requirement that disciplines the question is in the flow diagram: you must report records excluded, with reasons. Try to satisfy that with a question like "what do we know about AI in education" and it collapses — the reason for each exclusion would be a judgement call you cannot restate consistently. A question naming the population, the intervention or object, and the outcome survives it, because exclusions then have a stateable reason: wrong population, wrong outcome, wrong design.
So the practical test is not "is this question interesting" but "could I write the exclusion reason for a borderline paper, and would a colleague write the same one?" If not, the question is still too broad, and no amount of reading fixes it later.