Step one: find the edge
Nothing useful happens until we know what is already understood. So the first question is diagnostic rather than instructional — What answer are you hoping for, and why do you want it to be true?
An answer of "I don't know" is a perfectly good answer here. It is information, and it narrows the search immediately.
Step two: test the reasoning, not the answer
A right answer and a lucky answer look identical. The second question separates them: How would you check whether this AI response is actually correct?
This is the step most tutoring skips, and it is the one that decides whether anything transfers to the next problem.
Step three: work it aloud
Understanding gets built by asking the model to argue the opposite case, then judging which argument holds. Saying reasoning out loud exposes the gaps that silent reading hides — including to the person doing the reasoning.
Where a step cannot be justified, that is the step to work on. Everything else can wait.
Step four: check it held
The same idea is returned to later in an unfamiliar form. If it survives that, it was understood; if it does not, we know precisely where it broke.
Across prompt design and checking AI output alike, this is the step that turns a good session into a durable one.
What you should notice
After a few sessions the change worth watching for is the learner prompts for reasoning and sources rather than conclusions.
At that point a learner is running the method rather than receiving it, which is the whole intention.