ChatGPT for Studying: What It's Great At, and Where It Will Quietly Lie to You

ChatGPT is the best explainer most students have ever had access to, and the worst examiner.

Both of those are true at the same time, and the students who get real value from it are the ones who know which is which. Use it for what it's good at and it will save you hours. Use it for what it's bad at and it will make you feel ready for an exam you're about to fail.

This isn't a hit piece. We build AI tools. We just think you should know where the edges are.

What ChatGPT is genuinely great at

Explaining one thing three different ways

This is the killer feature. Your textbook explains equilibrium once, in one way, and if that way doesn't land you're stuck. ChatGPT will explain it with an analogy, then with a worked example, then as a diagram in words, and it won't get impatient when you ask a fourth time.

For getting a concept to click, it's hard to beat.

Turning your notes into questions

Paste in a page of notes and ask for ten questions on it. Instant quiz. This is a real form of retrieval practice, and retrieval beats re-reading by a wide margin. Just be aware the questions will be in ChatGPT's style rather than your exam's, which we'll come back to.

Summarising and restructuring

Turning a 40-page chapter into a one-page summary, or a messy set of class notes into a clean list of definitions, is a task it does well and fast. It's a good starting point for your own summary notes. Not a replacement for making them, because the making is half the learning, but a good scaffold.

Being available at 11pm

No tutor is. No teacher is. When you're stuck on a homework question the night before it's due, having something that can walk you through the method is a real advantage, as long as you actually do the walking rather than copying the answer.

Where it will quietly lie to you

None of these are dramatic failures. They're quiet ones, which is what makes them dangerous.

It marks you kindly

Give ChatGPT an answer and ask "is this right?" and it will very often find a way to say yes. It's trained to be agreeable. A wrong answer with confident wording gets "you're on the right track". A hedge like "I think it might be 40%?" gets partial credit for the 40%.

A real examiner does neither. Practise with a marker that rewards hedging and you'll hedge in the exam, where it's worth nothing. This is the single biggest way ChatGPT gives students false confidence, and it's the reason we built our own marking to be deliberately unkind.

It gets arithmetic wrong, confidently

Multi-step calculations are not its strength. It will make a slip in the third line and carry it through with total composure. If you're checking maths or physics working against ChatGPT's answer, you may be checking against a wrong answer. Always redo the numbers yourself.

It invents sources

Ask it for a reference and it may produce a very plausible-looking citation that doesn't exist. Never put a ChatGPT citation in an essay without finding the actual source first.

It doesn't know your exam board's mark scheme

ChatGPT has read a lot of exams. It hasn't read your one, and it has no idea how your board allocates marks. When it tells you an essay answer is "excellent", that's a general judgement, not a judgement against the criteria your examiner will use. The gap between those two can be a whole grade band.

Five prompts that actually work

These play to its strengths.

1. The three-ways explainer

Explain [concept] three ways: first with a real-world analogy, then with
a worked numerical example, then in the formal language my textbook
would use. I'm studying for [exam and level].

2. The notes-to-quiz converter

Here are my notes on [topic]. Write 10 short-answer questions that test
whether I actually understand them, not just whether I remember the
words. Don't show me the answers until I ask.

3. The "why is this wrong" prompt

Here is a question, my answer, and the correct answer from the marking
scheme. Don't tell me the correct answer again. Tell me precisely where
my reasoning went wrong and what I misunderstood.

This one's excellent because you supply the truth (the real marking scheme) and only ask it to explain the gap.

4. The examiner roleplay, with strict rules

Act as an examiner for [exam system] in [subject]. Mark my answer
strictly against what a marking scheme would require. If I hedge or
say "not sure", give 0. Do not praise me. State the marks, the specific
error, and the correct approach.

It'll still drift towards being nice over a long chat, so restart the conversation when you notice.

5. The "explain like I'm going to teach it" prompt

I'm going to explain [concept] to you as if you're a student who's
never seen it. Ask me questions when my explanation is unclear or
wrong. Don't help me until I get stuck.

Teaching something is one of the strongest retrieval exercises there is, and this makes ChatGPT the student.

Three prompts that backfire

"Is this answer right?" with no marking scheme attached. You'll get a yes far more often than you deserve. Always give it the truth to check against, or use something that marks honestly.

"Write me an essay on [topic]." Even if you only intend to "use it as a guide", you've skipped the part where you learn to structure an argument. The exam is handwritten, timed, and closed-book. The essay ChatGPT writes will never be sat next to you.

"Give me the answers to these past paper questions." You'll learn nothing and you'll also be trusting its arithmetic, which is the thing it's worst at. If you want past paper answers, your exam board publishes real marking schedules, and those are correct.

The one habit that keeps you safe

Treat everything ChatGPT tells you as a claim to be checked, not a fact to be learned.

That sounds exhausting. In practice it's one small habit: whenever it gives you something you're going to rely on (an answer, a formula, a fact, a source), verify it against something real. Your textbook. Your teacher. Your exam board's published papers and marking schemes. In New Zealand that's NZQA. Every board has an equivalent.

Verifying is not wasted effort. Catching an AI error is a strong form of active learning. Letting one slide past is how you walk into the exam confidently wrong.

The short version

• ChatGPT is excellent at explaining, quizzing from your notes, summarising and being there at 11pm.
• It's bad at honest marking, multi-step arithmetic, real citations, and your exam board's mark scheme.
• Its confidence tells you nothing. Check anything you're going to rely on.
• Use it as a tutor. Don't use it as an examiner unless you force strict rules on it, and restart when it goes soft.