Medicine & health

How to use AI in medical school without learning less

Medical students have adopted AI fast, mostly as a summary machine before exams. The research on how memory works suggests pointing it the other way.

By the Ace team · November 10, 2026 · 6 min read

How to use AI in medical school without learning less — Ace blog

Short answer: use AI to question you on your own lecture slides and notes, not to summarise them for you. Surveys show most medical students already use ChatGPT, mostly for summaries and explanations and most heavily around exams, but reading a summary is passive and AI errors matter in medicine. Ask it for short-answer questions from your material, answer from memory, check every answer against the source, and keep doing your question banks.

How are medical students using AI now?

Mostly as a shortcut to understanding. In a 2024 survey of 415 US medical students, Ganjavi and colleagues found 52.0% had used ChatGPT for medical school tasks and 38.4% had used it to study. Every student who studied with it asked it specific content questions, and 63.4% asked it for high-yield overviews of broad topics.

A smaller German survey by Fußhöller and colleagues (2025), with 84 complete responses, found 76.2% had used ChatGPT for their medical education. The most common use was summarising information (60.7%), followed by literature research and clarifying concepts, and students reported using it significantly more during exam preparation than during the rest of the semester. Most of them were also uneasy: 73.3% of those asked worried about incorrect or inaccurate information.

So the typical pattern is: exams are coming, there is too much to read, and AI condenses it. That is understandable. It is also the use that does the least for your memory.

AI can pass board-style questions, so why not learn from it?

In early 2023, Kung and colleagues tested ChatGPT on 350 text-only questions from the three USMLE Steps. Without any specialised training, it performed at or near the passing threshold of about 60% on all three.

That shows AI can produce plausible medical answers. It does not show that reading those answers makes you able to produce them. The skill you need on the ward and in the exam hall is retrieving the right fact or mechanism yourself, under pressure, often from a vignette that does not name the topic. A summary hands you the conclusion and skips the retrieval.

And "near the passing threshold" also means a meaningful share of answers were wrong. In most subjects a wrong summary costs you a mark. In medicine a confidently wrong drug interaction or dose, learned and remembered, is the kind of error you do not want to carry onto a ward.

What is a better way to use AI in medical school?

Turn it from a summariser into an examiner. The case for this is old and well established. In their review of test-enhanced learning in medical education, Larsen, Butler and Roediger (2008) summarise the research: repeated testing produces better retention than repeated study, especially when the tests are spaced out over time; tests that need effortful recall, such as short-answer questions, beat recognition tests such as multiple choice; and feedback is crucial to learning from tests.

That maps neatly onto how to use a chatbot:

  • Feed it your own material. Upload or paste the lecture slides and notes your school actually teaches from. Your exams follow your curriculum, and keeping AI inside it limits how much it can wander off or invent.
  • Make it ask, not tell. Short-answer questions first, one at a time, answered before you see anything.
  • Get feedback from the source. After each answer, ask it to quote the line from your slides that supports the correct answer, then look at that line yourself.
  • Come back later. Re-quiz yourself on the same lecture a few days later, then a week later. The spacing matters as much as the testing.

None of this replaces question banks or your flashcard routine. Our guide on how to study in medical school covers how to build those into your week; AI quizzing fits in as the same-day step after a lecture, before material reaches your spaced reviews.

Which prompts work for medical school?

  1. Lecture recall: "Using only these slides, ask me ten short-answer questions on the learning objectives. One at a time. Do not give the answer until I have tried."
  2. Mechanisms: "Ask me to explain the mechanism on slide 12 step by step. Then tell me which step I left out, quoting the slide."
  3. Vignettes: "Write a short clinical vignette based on this lecture without naming the condition. Ask me for the most likely diagnosis and the next step, then check my answer against the slides."
  4. Pharmacology: "From my notes, give me a drug name. I will state its class, mechanism, one key side effect and one contraindication." Our pharmacology study guide has more on organising drugs by class.
  5. Anatomy: "Describe a structure's relations from my notes and ask me to name it." Pair this with labelled diagrams; see how to memorise anatomy.
  6. Error review: "Here are the five question bank items I got wrong this week. Ask me a new question on each underlying concept, using my notes."

What should you never trust AI with?

  • Doses, thresholds and lab values. Check every number against your lecture material, your school's recommended references or national guidance.
  • Guidelines. Guidance differs between countries and changes over time. A chatbot may give you another country's version or an old one.
  • References. In the German survey, 16% of those asked raised concerns about unreliable or non-existent sources. If AI cites a paper, find it before you rely on it.
  • Real patients. Never paste identifiable patient information into a general AI tool, and follow your school's and placement's rules on AI use.

The fiddly part of all this is keeping questions tied to your own slides and checking each answer. Ace is built around that step: it turns your PowerPoint, PDF or Word lecture material into practice questions written from that material only, checks every answer against your file and shows the sentence that backs it, and drops questions that fail the check. Label-the-diagram questions use screenshots from your own slides, and you pick which lectures to study. See how PDFs become practice questions.

Frequently asked questions

Is it OK to use ChatGPT in medical school?

For private study, most schools allow it, but policies differ, especially for assessed work and anything involving patient information. Check your school's guidance. For learning, use it to test yourself on your own material rather than to read summaries.

Can ChatGPT replace question banks?

No. Question banks are written to match the format of your exams, while a chatbot's questions are only as reliable as its last answer. Use AI for same-day quizzing on your lectures and keep question banks for exam-style practice.

Can AI pass the USMLE?

In Kung and colleagues' 2023 study, ChatGPT performed at or near the passing threshold of about 60% on text-only USMLE-style questions from all three Steps. Passing questions is not the same as being reliable for learning, so check what it tells you.

Are AI summaries useful at all for medical students?

They can give a quick map of a topic before a lecture. The trouble is relying on them for revision, because reading a summary involves little retrieval, and any error in it goes straight into your notes.

Sources

  1. Ganjavi, C., Eppler, M., O'Brien, D., Ramacciotti, L. S., Ghauri, M. S., Anderson, I., Choi, J., Dwyer, D., Stephens, C., Shi, V., Ebert, M., Derby, M., Yazdi, B., & Cacciamani, G. E. (2024). ChatGPT and large language models (LLMs) awareness and use. A prospective cross-sectional survey of U.S. medical students. PLOS Digital Health, 3(9), e0000596.
  2. Fußhöller, A., Lechner, F., Schlicker, N., Muehlensiepen, F., Mayr, A., Kuhn, S., Hirsch, M. C., & Knitza, J. (2025). Perceptions, usage, and educational impact of ChatGPT among medical students in Germany: Cross-sectional mixed methods survey. JMIR Formative Research, 9, e81484.
  3. Kung, T. H., Cheatham, M., Medenilla, A., Sillos, C., et al. (2023). Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models. PLOS Digital Health, 2(2), e0000198.
  4. Larsen, D. P., Butler, A. C., & Roediger, H. L. (2008). Test-enhanced learning in medical education. Medical Education, 42(10), 959–966.

Turn your notes into practice questions

Drop a PDF, slides or a YouTube lecture into Ace and get practice questions checked against your own material — then play them solo or against a friend.

Try Ace free