Ace
Short answer: give ChatGPT your own course material and the format of your real exam, ask it to write a full practice test without showing any answers, then sit it timed and closed-book. Paste your answers back and ask it to mark them against your notes, quoting the line that supports each mark. Keep every mistake in an error log and ask for a fresh test on those topics a few days later.
Why a practice test beats another read-through
Taking a test is not just a way to measure what you know; it changes what you remember. In Roediger and Karpicke's (2006) experiment, students who practised recalling a passage remembered much more of it a week later than students who spent the same time rereading it. A large meta-analysis by Adesope, Trevisan and Sundararajan (2017) found the same pattern across many studies and formats: practice tests beat restudying.
The catch with AI is that it is very easy to let it do the work. In a field experiment with nearly a thousand high-school maths students, Bastani and colleagues (2025) found that students with unrestricted GPT-4 access did better during practice but 17% worse on the later exam once access was removed, compared with students who never had it. A version of the tool with guardrails, which gave hints instead of answers, largely avoided that harm. So the rules below are the guardrails: ChatGPT sets and marks, you answer alone.
Step 1: build the test
Start a new chat and upload or paste the material the exam covers: lecture slides, notes, a textbook chapter. Then describe the real exam. Something like:
"Using ONLY the material I've uploaded, write a practice exam for [course]. Format: 15 multiple-choice questions with four options, 4 short-answer questions and 1 longer question, like my real exam. Make the wrong options mistakes a half-prepared student would make. Cover every section in proportion to how much space it gets in the material. Do NOT show any answers or hints. Number the questions."
If you have past papers, upload one as well and add: "Match the style and difficulty of this past paper, but do not copy its questions." That keeps the test close to what your examiners actually ask. Our guide to using past papers properly covers how to fit both into a revision plan.
If the test is too easy, say so and ask for harder questions that need applying an idea, not just recognising it.
Step 2: sit it like the real exam
- Copy the questions out of the chat or print them, so the answers never appear on the same screen.
- Set a timer for the real exam's time per mark.
- Notes closed, chat closed. Write your answers in a separate document or on paper.
- Mark your confidence next to each answer: sure, unsure or guess. This makes the marking more useful.
It will feel harder than revising. That difficulty is the point; struggling to retrieve something is what strengthens it.
Step 3: get it marked honestly
Paste your answers back into the same chat with this prompt:
"Mark my answers strictly, as an examiner would. For each question, say right, partly right or wrong, give the correct answer, and quote the sentence from my material that supports it. If my material does not cover something, say so rather than using outside knowledge. Then list the topics I lost the most marks on."
The quote is your safety net. ChatGPT can mark confidently and still be wrong; if the quoted sentence does not actually say what it claims, check your notes before accepting the mark. Our guide on checking AI study answers goes further.
Compare the marking with your confidence ratings. Wrong answers you were sure of are the most important ones to fix, because those are the misconceptions that survive into the exam.
Step 4: the error log and the re-test
Ask ChatGPT for a table of every question you missed, with the correct answer and the supporting quote. Copy it into a document you keep. Then, two or three days later, start a new chat with your material and that log:
"Write a new practice test focused on these topics. Use different questions that test the same ideas from another angle. No answers until I've finished."
Spacing the re-tests out over days, rather than doing them straight away, is what makes the learning last (Cepeda et al., 2006). One mock a week, plus a short re-test of the error log in between, is a sensible rhythm for the last month before exams.
If building and marking the test by hand gets tedious, Ace is designed for the same loop: upload your PDF, slides or lecture video, choose the chapters, and it writes practice questions from that material only, each answer backed by the sentence from your own file.
Frequently asked questions
Should I use ChatGPT's study mode for a practice test?
Study mode is built for guided learning, asking you questions and giving hints. It is useful for working through a topic. For a mock exam, a normal chat with the instructions above gives you more control over format and timing.
Can ChatGPT predict what will be on my exam?
No. It can only test you on the material you give it. Cover everything your course says is examinable, and use real past papers to see the style.
How long should a ChatGPT practice test be?
For a full mock, match the length of the real exam. For daily practice, 10 to 15 questions on one topic is enough.
Is it cheating to use ChatGPT this way?
Making yourself practice questions for your own revision is generally fine. Check your university's rules, and never use AI on assessed work where it is not allowed.
Sources
- Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255. PubMed
- Adesope, O. O., Trevisan, D. A., & Sundararajan, N. (2017). Rethinking the use of tests: A meta-analysis of practice testing. Review of Educational Research, 87(3), 659–701. SAGE
- Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26). doi.org/10.1073/pnas.2422633122
- Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132(3), 354–380. PubMed
