- Why most people use ChatGPT for revision wrong
- The core prompt: turn ChatGPT into a quizmaster, not a textbook
- Prompts for building flashcards that get used
- Prompts for essay and long-answer subjects
- Prompts for maths, science, and anything with steps
- The prompt that catches false confidence before the exam
- What to avoid entirely
- Frequently asked questions
- Sources worth reading
The short version: the ChatGPT prompts that work for studying are the ones that make you retrieve information, not the ones that hand it to you neatly summarised. Below are the exact prompts I use with my own family, plus the uncomfortable bit nobody tells you: if ChatGPT is doing the thinking, you are not learning anything, you are just watching someone else revise.
Why most people use ChatGPT for revision wrong
My nephew Daniel came to me in March last year with his GCSE history notes and a ChatGPT tab open. He'd asked it to "summarise the causes of World War One" and had a beautifully formatted answer with headings and bullet points. He read it twice, felt clever, and closed the laptop. Three days later I quizzed him on the same topic and he could remember maybe two facts.
That's the trap. A tidy summary feels like progress because it's easy to read. But easy to read is not the same as easy to recall under exam pressure. Cognitive scientists have known this for decades, it's called the illusion of fluency, and it's why rereading notes is one of the least effective ways to revise even though it's the most common one. Ironically, ChatGPT makes this worse before it makes it better, because it produces such clean, confident-sounding text that students mistake reading it for learning it.
The fix isn't to stop using ChatGPT. It's to stop asking it to explain things to you and start asking it to test you on things. That single shift changed how Daniel revised, and his mock grade went from a 5 to a 7 in one term, not because the tool got smarter, but because he started using it to practise retrieval instead of consuming summaries.
The core prompt: turn ChatGPT into a quizmaster, not a textbook
This is the one prompt I'd tell any student to bookmark:
- "Act as a strict examiner. I'm revising [topic/subject]. Ask me one question at a time about [specific topic], don't give me the answer, wait for my response, then tell me if I'm right or wrong and explain what I missed. Keep questions at [GCSE/A-level/degree] standard."
The magic is in "don't give me the answer" and "one at a time." Without those two instructions, ChatGPT will happily dump ten questions and ten answers in one go, and you'll skim the answers instead of struggling for them. The struggle is the point. Struggling to retrieve a fact from memory, even if you get it wrong, strengthens that memory far more than reading the correct answer ever will. This is called the testing effect and it's one of the most replicated findings in learning science.
A step-by-step revision session using this method
Here's the actual routine I gave Daniel for his history GCSE, which took about 25 minutes a session:
- Minute 0 to 2: Tell ChatGPT the exact topic, exam board, and level. "I have AQA GCSE History, paper 2, Germany 1890 to 1945."
- Minute 2 to 15: Use the quizmaster prompt above. Answer out loud or on paper before typing anything back.
- Minute 15 to 20: Ask it: "Of the questions I got wrong, pick the two I struggled with most and explain them again using a different example than before."
- Minute 20 to 25: Ask it: "Now give me three of those same questions again, reworded, without telling me which ones they were."
That last step matters more than people realise. Spaced, repeated retrieval of the same facts, worded slightly differently, is what moves things from short-term to long-term memory. One quiz round isn't revision. It's a warm-up.
Prompts for building flashcards that get used
Everyone tells students to make flashcards with ChatGPT. Almost nobody tells them the flashcards it produces by default are usually too long and too easy, which defeats the purpose. Try this instead:
- "Create 20 flashcards for [topic]. Each answer must be one sentence maximum. Make at least a third of the questions require me to compare two things or explain 'why', not just recall a date or definition."
Then, and this is the step people skip, ask it to generate a second batch a week later covering the same ground but phrased completely differently, and test yourself on both sets mixed together. Spaced repetition works on roughly a 1 day, 3 day, 7 day, 14 day rhythm. If you're reviewing the same flashcards every single day, you're wasting time on facts you already know instead of the ones that are fading.
Prompts for essay and long-answer subjects
For subjects like English literature, history, law, or philosophy, the risk isn't forgetting facts, it's writing weak arguments. Here's what works:
- "Here's my practice essay answer to [exam question]. Don't rewrite it for me. Instead, ask me three questions that would push me to justify my weakest paragraph, the way an examiner's mark scheme would."
- "I'm about to answer this question: [paste question]. Before I write anything, ask me what the command word means and what the examiner is looking for at top band."
Notice neither of these asks ChatGPT to write anything for the student. That's deliberate. The moment it writes the paragraph, the student has lost the chance to practise the exact skill the exam is testing, which is producing that paragraph themselves under time pressure. If you want a fuller walkthrough of this approach for students specifically, I wrote a longer piece on how students can use ChatGPT prompts to study more effectively that goes into subject-specific examples.
Prompts for maths, science, and anything with steps
For STEM subjects, the failure mode is different. Students ask ChatGPT to solve the problem, copy the method, and feel like they understand it because the steps made sense when they read them. Reading a solved proof and being able to produce one from a blank page are very different skills. Use this instead:
- "Give me a [topic] problem at [grade level], similar difficulty to past paper questions. Don't solve it. Wait for my attempt, then only tell me the first step I got wrong, not the whole solution."
Getting only the first mistake, rather than the full corrected solution, forces a second genuine attempt rather than a copy-paste of the fix. It's slower. It's also how actual understanding gets built, one corrected step at a time rather than one full worked example that gets forgotten by tomorrow.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
The prompt that catches false confidence before the exam
This is the one I wish every student used the week before an exam:
- "I think I understand [topic] well. Ask me the hardest, most awkward question you can on this topic that would trip up someone who only has surface-level knowledge."
Students consistently overestimate what they know, this is well documented and it's got a name, the Dunning-Kruger effect, and it hits hardest right before exams when confidence peaks but depth hasn't caught up. Forcing ChatGPT to hunt for the gap rather than confirm what you already know is uncomfortable, and that discomfort is the whole value of the exercise.
What to avoid entirely
A few habits I've watched backfire, including with my own family:
- Asking ChatGPT to "simplify" a topic repeatedly until it's so dumbed down it no longer matches exam-level detail, then walking into the exam underprepared for the actual wording of questions.
- Pasting past exam papers and asking for "model answers" to memorise word for word. Examiners mark understanding, not recitation, and near-identical phrasing across multiple students' scripts is a real red flag that has led to malpractice investigations in some UK schools.
- Using it as a substitute for past papers under timed conditions. No amount of chat-based quizzing replaces the specific stress of a 45 minute timed paper. Do both.
None of this means ChatGPT doesn't work for revision. It means it works when it's making you produce the answer, not when it's producing the answer for you. That's the whole shift, and it's a smaller change to your prompts than most people think, but a much bigger change to how much you remember in six months' time.
If you want the full set, start at the ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps.
I built a free prompt builder for exactly this.
Frequently asked questions
What is the best ChatGPT prompt for revision?
The single most effective prompt is asking ChatGPT to act as an examiner that quizzes you one question at a time, withholds the answer until you respond, and tells you specifically what you got wrong. This forces retrieval practice, which is proven to build stronger memory than reading summaries or notes.
Can ChatGPT replace flashcards or past papers?
No. ChatGPT is best used to generate and vary flashcards and practice questions, but it should sit alongside timed past papers under exam conditions, not instead of them, because timed pressure is a skill in itself that quizzing alone doesn't build.
Is it cheating to use ChatGPT to revise?
Using it to quiz yourself, generate practice questions, or explain a concept you're stuck on is standard revision support, similar to a tutor or a revision guide. It becomes a problem only when students copy generated answers directly into coursework or exams without understanding them, which most exam boards treat as malpractice.
How often should I use ChatGPT to revise a topic?
Short, spaced sessions work better than long ones. Aim for 20 to 25 minute quiz-style sessions repeated at roughly 1 day, 3 day, 7 day, and 14 day intervals on the same topic, rather than one long session the night before.