Studying With AI
As a tool that tests you, not one that hands you answers
① AI explains so smoothly that a feeling of "I get it" comes easily. But that feeling isn't the same as being able to pull the material out of your own head. In an experiment, students who practiced with answers handed over did worse on an exam once the AI was taken away.
② The knack for using AI to study is not to receive answers but (1) be asked with hints, (2) explain first and have gaps pointed out, and (3) be tested with quizzes.
③ But an AI teacher also teaches wrong things with confidence. Check concepts against textbooks and originals, and do the calculations yourself.
① Getting started — two ways to use AI for studying
The same AI can be used in opposite ways. One is receiving answers — "explain this," "give me the solution." The other is making it test you — "quiz me," "check whether my explanation is right." If researching with AI was about investigating facts, this piece is about making things stay in your own head. Same tool, different goal, so a different way of using it.
② Why "receiving answers" may not be studying
A high school math experiment published in PNAS in 2025 shows the problem directly. Students were given AI while working through practice problems in class. One version gave answers the way an ordinary chatbot does (GPT Base); another was designed to give hints rather than the answer outright (GPT Tutor) (Bastani et al., PNAS 2025).
| Condition | Score while practicing with AI | Exam with the AI taken away |
|---|---|---|
| AI that gives the answer (GPT Base) | About 48% higher | About 17% lower than students who never had AI |
| Hint-giving tutor design (GPT Tutor) | About 127% higher | That harm was largely mitigated |
While practicing, scores rise because the AI is there. But once the AI is gone, it turns out the score was the AI's work, not the student's skill. This echoes a long-established psychology result. In Roediger and Karpicke (2006), students who read a passage were compared when they either reread it or practiced recalling it as if in a test: after 5 minutes rereading did better, but after 2 days and 1 week, the recall group remembered far more. And the rereaders believed they'd remember better (Roediger and Karpicke, Psychological Science 2006). Reading smoothly and remembering for a long time are different things.
The PNAS experiment is a result from one setting — high school math. You can't assume the same size of effect for adult learning or other subjects. The point is the direction, not the figures: practice where AI solves things for you may not translate into skill without AI, and changing the design can sharply reduce that risk.
③ Three principles for study prompts
- Be asked, not told. Don't request the answer; say "give me hints one step at a time." If you only get help where you're stuck, the solving still happens in your own head.
- Explain first, then have the AI find the gaps. Explain the concept in your own words and have the AI point out what's missing and what's wrong. Explaining forces you to retrieve it, which exposes the blanks in your understanding.
- Take the test, then revisit what you missed. Solve questions the AI sets (retrieval), then re-solve only the ones you missed, spaced out over time. That reviewing at intervals sticks longer than cramming is a widely confirmed finding in learning research.
④ Four practical prompts
1. Hint tutor — help only when stuck
"I'm going to solve this problem, so never give me the answer or a finished solution. If I say I'm stuck, give me one hint at a time. When I show you my working, point out only where it's wrong and tell me how to fix it by asking questions.
Problem: [paste]"
2. Explain and get corrected — in my own words
"I'll explain [concept] the way I would to a friend who knows nothing. After reading it, point out separately ① what's wrong ② what key point is missing ③ what's vague. Don't give me a rewritten final version — just ask me questions so I can rewrite it myself.
My explanation: [paste]"
3. Quiz me — only from my own material
"Write 5 quiz questions based only on the material below. Ask one at a time and give the answer and explanation only after I respond. Don't ask about anything not in the material. Keep a list of the ones I miss and ask them again at the end.
Material: [paste]"
4. Sorting mistakes and a review schedule
"The problems I got wrong and concepts I confused today are [list]. ① Classify why I got each wrong in one line (concept misunderstanding / calculation slip / misread the question). ② For the concept misunderstandings only, write review questions for tomorrow, in 3 days, and in a week."
⑤ The limits of an AI teacher — it teaches wrong things confidently
AI is a tool that says wrong things smoothly too. In studying, that's more dangerous: a beginner has no standard for judging whether the AI's explanation is right, and a concept learned wrong is harder to fix later. So hold to three rules.
- Check key concepts against a textbook or official material. Use the AI's explanation to get the feel, and confirm definitions and formulas in the original (how to open the source and check it).
- Do the calculations yourself. AI sometimes fakes arithmetic by prediction, so following its working blindly can go wrong.
- Draw quizzes from material you hold. Paste your material and have it ask only from that, as in prompt 3, which reduces the risk of wasting study time on questions the AI invented.
⑥ Applying it to studying investing
The same principles hold when you study investing concepts. For example, to study P/E, do it like this.
| Step | What you do |
|---|---|
| ① In my words first | Write your own explanation: "a low P/E means a cheap stock" |
| ② Get the gaps pointed out | Ask the AI what's wrong or missing. It may raise that the same P/E means different things depending on industry averages, one-off profits and growth rate |
| ③ Quiz | Get a question like "how would you compare two companies with the same P/E?" and answer it yourself |
| ④ Check the original | Confirm definitions and figures in a textbook or the filing itself |
Asking "should I buy this stock?" isn't studying — it's handing the decision to the AI. The method here goes only as far as making concepts and decision frameworks your own. Study from material like the U.S. ETF beginner's guide for the basics and the economic indicators calendar, fully explained for reading release data, and use the AI as the one who asks you questions and quizzes you.
⑦ Practical checklist
□ Did I ask for hints or questions instead of the answer?
□ Did I write the concept in my own words first, instead of only reading the AI's explanation?
□ Did I have it quiz me from material I hold?
□ Do I have a plan to re-solve the ones I missed, at intervals?
□ Did I check key definitions and formulas against a textbook or the original?
□ Did I do the calculations myself rather than letting the AI?
□ Have I checked that I can still explain and solve it with the AI turned off?
※ The research figures cited here come from specific conditions (a high school math experiment, a memory experiment) and can't be guaranteed to apply at the same size to every learning situation. This piece is general guidance on study methods and is not investment advice. Investment decisions and responsibility rest with the investor.
※ This guide is provided for general educational purposes and simplifies technical details for readability.
New guides, when they land
We publish AI literacy guides twice a week. Subscribe and the next one comes to you — free, unsubscribe anytime.