How to Use AI to Study (Without It Doing the Studying for You)

Test on Friday. Forty pages of notes. An AI that can summarize all of it in about eight seconds.

Tempting. Also, according to a pretty big pile of research, the slowest way to actually remember any of it.

The short answer: Use AI to make you remember things, not to read things for you. Get it to quiz you, check your answers and find your weak spots, using your own class material. Try before you ask, and test yourself with the AI closed before the real thing.

The one rule that beats every study hack

The most reliable finding in learning science sounds almost too simple: pulling information out of your brain teaches you more than putting it back in.

In a classic experiment, students read a short science passage. One group read it four times. Another read it once, then practiced recalling it three times. Five minutes later, the rereaders were slightly ahead, 83% to 71%. A week later, the recall group remembered 61% and the rereaders only 40%. And the rereaders were the ones who predicted they'd remember the most [1].

A later study pitted recall practice against making detailed concept maps. A week later, the recall group scored about 50% higher. Most students had bet on the opposite [2].

So here's the whole strategy in one line: get AI to make you recall more and reread less.

What the best AI tutor did differently

In a study published in 2025, Harvard ran a real experiment in its biggest intro physics class. For one topic, students learned from a custom AI tutor at home. For another, they had an in-class lesson from experienced instructors. The AI-tutored students learned more than twice as much, and most finished in under an hour [3].

But that tutor wasn't plain ChatGPT. It walked students through problems one step at a time instead of handing over answers. It was built on expert-written solutions. And it matched the class exactly. The researchers spent months building it, and they say it might not beat great teaching for really complex, big-picture thinking.

You can borrow the same three ingredients: step by step, grounded in real material, matched to your class.

Your AI study playbook

Before class (10 minutes). Give the AI the reading or slides and ask: "What are the three ideas I most need to get from this, and what question should I be able to answer afterward?" You're not learning it yet. You're building a hook to hang it on.

Right after class (15 minutes). Close your notes. Write or say everything you remember. Then ask the AI to compare your recall with the material and tell you what you missed. This one step beats an hour of rereading.

Midweek (20 minutes). Ask for practice questions, one at a time, on last week's material. The gap is the point. A little forgetting before you practice is what makes it stick.

When you're stuck. Try first, properly. Then: "Here's my attempt. Give me a hint for the next step, not the answer." Still stuck after the hint? Get the full solution, then do a similar problem on your own to prove you can.

Test week. Ask for a mixed quiz across every topic, weighted toward what you got wrong before. Then do one full practice run with the AI closed. That's the only honest check.

Prompts you can steal

Three rules that keep AI honest

1. Feed it your material. General chatbots answer from general knowledge, which might not match your teacher's definitions or what they emphasize. Give it your notes and tell it to stick to them.
2. Double-check anything that's graded. AI can be confidently wrong. If a fact matters, find it in your textbook or notes.
3. Know your class's AI rules. Fine for studying doesn't always mean fine for homework you turn in.

What to stop doing

Rereading AI summaries feels productive and mostly isn't. Copying worked answers is worse: in one big study, high school students who practiced with an answer-giving AI scored 17% lower on the exam than students with no AI at all [4]. Cramming only this week's topic means you'll lose week two by week ten.

Where Crammm fits

You can run this whole playbook with a general chatbot if you're disciplined about it every single day. Crammm is built to take the discipline problem off your plate.

You upload your syllabus, notes and readings into a Notebook, and every tool works from that material. It quizzes you, typed or out loud. It tracks what you forget and when you tend to answer well, then moves hard topics into your best hours and brings weak ones back before they fade. After every session, Course Readiness shows where you stand, so test week isn't a surprise.

The honest limits: it's only as good as what you upload, it can summarize as well as quiz, and the free plan is just a taste. But if the hard part is remembering to study the right way every day, a tool that already knows your classes and asks the questions for you is the easiest fix we know.

Try it before your next test. Crammm is opening up to beta testers. Start free with 1,000 AI credits and a day of unlimited access. Upload one class, do one quiz session, and tell us honestly how it went.

Join the Crammm beta →

FAQ

Is it cheating to use AI to study?

Studying with AI, like quizzing yourself or getting a concept explained, usually isn't cheating. Check your rules before using it on graded work. For where the line actually sits, see Is Using AI Cheating?

Can Crammm make flashcards and quizzes from my notes?

Yes. Upload notes, slides, readings, photos, links, audio or video to your Crammm Notebook, and it builds flashcards and quizzes from that material. It keeps your material for the semester, so you don't have to paste it in every time.

Is Crammm free?

There's a free plan with 1,000 AI credits (one time), 24 hours of unlimited access, group chats and the calendar and scheduler. Pro is $8.99 a month and includes 10,000 AI credits a month.

When should I use AI while studying?

After you've tried on your own. Use it to check, quiz and correct you, not to go first.

References

1. Roediger and Karpicke, Test-enhanced learning, Psychological Science 17(3), 2006, Experiment 2
2. Karpicke and Blunt, Retrieval practice produces more learning than elaborative studying with concept mapping, Science, January 2011, via Purdue University
3. Kestin et al., AI tutoring outperforms in-class active learning, Scientific Reports, 3 June 2025
4. Bastani et al., Generative AI without guardrails can harm learning, PNAS 2025, summary via Wharton