An AI assistant is unusually good at some parts of studying and actively harmful to others. The difference is not the tool, it is whether it is doing the thinking you were supposed to do.
Here is a practical split.
Six uses that genuinely help
1. Explain this to me a different way
The strongest use. When a textbook explanation does not land, asking for the same idea framed through an analogy, a worked example, or from the opposite direction often works where re-reading does not. Ask for three explanations and pick the one that clicks.
2. Generate practice questions
Paste your lecture notes and ask for ten questions of increasing difficulty. Answer them closed-book. This converts passive material into retrieval practice, which is the mechanism that actually builds memory.
3. Check your reasoning after you have attempted something
The order matters enormously. Attempt it, then ask “here is my reasoning, where is it wrong?” You get the benefit of feedback without skipping the struggle that produces learning.
4. Interrogate your own understanding
Ask it to quiz you, or to play the role of an examiner asking follow-up questions. Being asked “why does that hold?” three times in a row exposes the places where you have a memorised phrase rather than an understanding.
5. Summarise something you have already read
Useful as a check: read it, write your own summary, then compare. The gaps between the two are the parts you missed.
6. Unstick yourself on mechanics
Syntax you have forgotten, a formatting problem, a citation style. Low-value, high-friction tasks where the learning content is near zero. For larger writing projects, AI can also assist with outlining and developing long-form content. AI book writing tools, for example, can help writers turn initial ideas into a structured book outline and develop chapters that can then be reviewed and refined manually.
Three uses that prevent learning
1. Producing work you submit
Aside from being against the rules almost everywhere, the practical problem is that the skill was the point. The gap shows up in the exam, where the assistant is not available and you have never done the thing unaided.
2. Getting the answer before attempting the problem
The productive part of a problem set is the twenty minutes of being stuck. Reading a correct solution produces a strong feeling of understanding and very little actual learning — you recognise the solution rather than being able to generate one.
3. Treating output as verified
Assistants produce confident, plausible, wrong answers, particularly on specific figures, citations, niche technical details and anything recent. They do not signal uncertainty reliably. Anything going into work that matters needs checking against a source.
The test
Ask yourself: after this exchange, can I do the thing without the assistant?
If yes, it was a tutor. If no, it was a substitute, and you have borrowed against an exam you will sit alone.
A second, more social test: would you be comfortable if your tutor watched the whole conversation? That one catches most of the cases where you already know the answer.
Prompts that work better
- “Explain X as if I understand Y but not Z.” Specifying your starting point improves the answer more than any other single change.
- “Give me the three most common misconceptions about this.” Surfaces the errors you might be holding.
- “Here is my answer. Do not give me the correct one — tell me which step is wrong.” Preserves the struggle while getting the correction.
- “What would I need to know to check whether this is right?” Turns an answer into a method.
What it is still bad at
Knowing your specific institution’s rules. Marking to your actual rubric. Anything requiring current, verifiable facts. And arithmetic where precision matters — for a weighted average that has to be exactly right, use a purpose-built calculator rather than asking a language model to do the sums.