Yes, ChatGPT is good for studying, as long as it tests you instead of answering for you. Use it as an examiner and a tutor, never as a substitute for your own thinking. The danger is real: in a large field experiment, students with unrestricted GPT-4 did better during practice but worse on the exam once it was taken away. Set it up to quiz you and you get the benefit without the crutch.
A confession first: I didn’t have this during A-levels. I started using ChatGPT to study and learn at Oxford, and I still do. I use it to consolidate information, as a tutor, to make notes on hard topics, and above all to test me. But it must never substitute for thinking. The moment it does the thinking for you, it’s stopped helping. Below are the prompts I’d use if I were back in JC.
Is it okay to use ChatGPT to study?
It’s fine for learning, and risky for anything you hand in. Those are two different questions.
Using AI to understand a topic, quiz yourself or check your reasoning is studying. Using AI to write work you submit as your own is a different matter, and your school’s rules decide what’s allowed. MOE says students must state where they’ve used AI and cite their sources, and are taught to be discerning because AI output can contain errors. Students who pass off AI-generated content as their own “will have to bear the consequences of academic dishonesty”. National exams are proctored, and AI is banned in them (MOE, May 2026).
If you’re unsure whether something is allowed, ask your teacher before, not after.
What does “letting AI think for you” actually look like?
It looks like getting the answer, nodding, and moving on. It feels productive. It isn’t.
Researchers gave nearly a thousand high-school math students in Turkey access to GPT-4 during practice (Bastani et al., PNAS). With a plain chatbot, practice scores rose 48%. But on the later exam, without AI, those students scored 17% lower than students who never had it (Wharton summary). Many had used it to get answers rather than to learn.
A second group used a version designed to give hints instead of answers. Their practice scores rose even more (127%), and the harm on the exam largely disappeared.
Same AI. Different instructions. Completely different outcome. That’s the whole lesson: your prompt decides whether AI is a tutor or a crutch.
How do you stop AI doing your thinking?
The Examiner Rule: every AI study session ends with the AI asking you questions, not you asking it.
Asking AI to explain is fine as a start. But explanation is input, and grades measure output: what you can retrieve and use on your own. Testing yourself is one of the best-supported study methods there is (the research is here). AI makes that testing almost free.
A simple check: before you ask, write down your own attempt, however rough. If you can’t be bothered to try first, you’re not studying. You’re outsourcing.
How do I use ChatGPT as a study tool? Four jobs, with prompts
These are prompts I’d use, written for a JC or upper-sec student. Copy them into ChatGPT or any chatbot, and replace the bits in [brackets].
Job 1: Consolidate your notes
Lesson slides, your own notes, a textbook chapter and a worksheet often say the same thing four different ways. AI is good at merging them into one clean page.
The “use only what’s in my notes” line matters. It cuts down on made-up content, and the ⚠️ flags tell you exactly what to check.
Job 2: A tutor that won’t hand you the answer
This is the “hints, not answers” design from the study above, built by you in one prompt. It keeps the thinking on your side of the screen. Some AI apps now have a built-in study or learning mode that does something similar (example from BYU). Turn it on if yours has one.
Job 3: Notes on the topic you keep avoiding
Then check the notes against your textbook or teacher’s notes. AI-made notes are a first draft, never the final word.
Job 4: The examiner (use this one the most)
Then, to make it “remember” your answers between sessions:
Paste that list into your next session and start with: “Here are my weak spots from last time. Test me on these first.” That’s spaced repetition with an AI doing the admin. (Some apps can remember past chats on their own. Don’t rely on it; the pasted list is more reliable.)
Bonus: Feynman with a sceptical listener
Explaining in plain words until the gaps show is the Feynman technique (Ohio State’s guide). AI makes a tireless listener.
Can AI mark my answers?
Roughly, yes. Precisely, no. AI doesn’t know exactly how your paper is marked, and it can be confidently wrong about what earns credit.
If you use it for marking, paste in the official mark scheme or your teacher’s model answer and ask it to compare your answer against that, point by point. Treat its feedback as a second opinion. Your teacher’s comments outrank it.
What are the risks of studying with AI?
Three big ones. All are manageable if you know about them.
- It makes things up. AI can state wrong facts, wrong formulas or fake sources fluently. Fix: give it your own notes as the source, ask it to quote which part it used, and check anything that matters against your textbook.
- It does the thinking you needed to do. This is the Bastani finding. If AI solves it, you didn’t learn it. Fix: the Examiner Rule. Hints before answers, and always end with you being tested.
- It can break your school’s rules. Submitting AI-written work as your own can count as academic dishonesty. Fix: know your school’s policy, acknowledge AI use where it’s allowed, and keep AI out of graded work unless your teacher says otherwise.
A smaller fourth: don’t paste personal details or other people’s work into a chatbot.
Who shouldn’t rely on AI for studying?
- If you’re learning something for the very first time, read the textbook or listen in class first. AI is far better at testing and clarifying than at building a foundation from scratch.
- If you notice you’re asking AI before trying yourself, stop for a week. Struggle first, then ask.
- For practice-heavy subjects like Math, AI can explain a step, but you still need to grind through problems by hand. See practice subjects vs concept subjects.
Sources
Checked 5 October 2026. If one of these has changed, tell me and I'll fix the article.
- doi.org/10.1073/pnas.2422633122
- knowledge.wharton.upenn.edu/article/without-guardrails-generative-ai-can-harm-education
- moe.gov.sg/news/parliamentary-replies/20260506-ai-usage-in-schools
- dennislearningcenter.osu.edu/the-feynman-technique
- genai.byu.edu/chatgpt-study-mode
Know someone who needs this?
Send it to a friend, a parent, or your kid.