How to use AI for studying without skipping the learning

How to use AI to study, in short

Use AI after you've tried, not instead of trying. Answer a question from memory first, then ask the AI and compare. Ask it to quiz you on your own notes rather than summarize them for you. When you're stuck, ask for a hint or an explanation of the step you're missing, not the finished answer. Check what it says against your course material, because your exam is marked against that. And before any of this, check what your course allows.

The reasons behind each of those rules come from one simple finding: the effort of pulling an answer out of your own memory is a large part of how learning happens. A tool that removes the effort can remove the learning with it.

The risk: AI can do the part that does the learning

A 2025 field experiment gave nearly a thousand high school maths students access to an AI tutor during practice (Bastani et al., 2025). One version worked like a standard chat interface. The other was prompted to safeguard learning. With access during practice, grades improved: 48% for the standard version and 127% for the safeguarded one.

Then access ended. Students who had used the standard version performed worse than students who never had access, with a 17% reduction in grades. The authors describe students using the tool as a "crutch" during practice. The safeguarded version largely avoided that harm.

With the standard version, practice scores went up while learning went down. That's the trap. An AI answer makes practice feel easy, and easy practice feels like progress, because you can see the right answers piling up on the page even when very little of the method is staying in your head.

Rule 1: Try first, then ask

The testing effect is one of the most reliable findings in learning research. In a classic experiment, students who restudied a passage did better on a test five minutes later, but students who had been tested on it remembered substantially more two days and a week later, even though restudying made the first group more confident (Roediger & Karpicke, 2006). A large review rated practice testing as high utility and rereading as low utility (Dunlosky et al., 2013).

Reading an AI's explanation is closer to rereading than to testing. So write or say your own answer first, even a rough one. Then ask. The comparison shows you exactly what you're missing.

Rule 2: Ask it to quiz you, not to tell you

Most AI tools are good at writing questions. Use that. A few prompts that keep you doing the recall:

• "Ask me five questions on this lecture, one at a time. Wait for my answer before the next one."
• "I'm going to explain how the kidney filters blood. Tell me what I got wrong or left out, but don't explain it yet."
• "Give me a question that mixes this week's topic with last week's."
• "Here's my answer to a past exam question. Which parts would lose marks?"

Rule 3: Ask for hints, not answers

When you're stuck on a problem, the full worked solution is tempting. Ask for the smallest nudge instead: "What's the first step?" or "Which concept does this question test?" Then try again. If you still can't do it, read the solution, close it, and solve the problem yourself from a blank page.

Rule 4: Check answers against your course material

AI answers can be wrong, and they can also be right in a way your course doesn't use. Your lecturer may define a term more narrowly, use a particular model, or expect a specific method. An exam is marked against the course, not against the internet.

So treat each answer as a claim to verify. If the tool cites sources, open the one that supports the sentence you care about. If it doesn't, find the relevant page in your notes or textbook yourself. Sources help, but they aren't proof: one audit of AI search engines found that only 74.5% of citations supported their associated sentence (Liu, Zhang & Liang, 2023). More on that in why AI answers make things up.

Rule 5: Know your course's AI rules

Rules on AI vary between universities, between departments and often between courses. Carnegie Mellon's teaching centre publishes example course policies that range from "Students may NOT use generative AI in any form" to "Students are fully encouraged to use generative AI" (Carnegie Mellon University). Yours could be anywhere on that range.

1. 1Read the syllabus and assignment brief for an AI policy.
2. 2If it's unclear, ask your lecturer or tutor, in writing if possible.
3. 3Keep studying (quizzing yourself, checking understanding) separate from assessed work, which often has stricter rules.
4. 4If your course requires you to acknowledge AI use, keep a note of what you used and how.

Good and bad ways to use AI for studying

• Helpful: quizzing you on your own notes, after you've tried to recall them.
• Helpful: explaining a step you got wrong, once you know which step it is.
• Helpful: generating practice questions that mix topics.
• Harmful: summarizing a lecture you haven't reviewed, then reading the summary.
• Harmful: solving the problem set while you watch.
• Harmful: answering from general knowledge when your course uses a specific approach.

Using your own notes as the source

Many of these problems shrink when the AI answers from your course material instead of the open web. That's how Lumyo works: you choose which notes a chat reads, answers cite the notes they used, and tapping a source opens the exact line. When your notes don't cover a question, it says it doesn't know instead of guessing. Use it the same way as any tool: recall first, then ask and compare. For a fuller method, see how to study from your own notes before an exam.