ChatGPT vs DeepSeek for medical students: what AI gets right and wrong

ChatGPT and DeepSeek are both useful for medical students if you use them to explain topics and quiz you, not as your source of facts. Studies don't show that either one is clearly better: each has come out ahead on different exams. Whichever you use, check what it tells you against your textbook and your national guidelines.

Is ChatGPT good for medical students?

Yes, for some jobs. In 2023, an early version of ChatGPT scored 52.4% to 75.0% across the three USMLE exams on public sample questions, close to the pass mark of about 60% (Kung et al., 2023). Newer models score higher, but that doesn't mean every answer is right, or right for Nigeria.

AI chatbots are good at:

• Explaining a slide in simpler words. Paste in the slide and ask for it again at a lower level, or with an example.
• Turning your notes into practice questions. MCQs, short-answer questions and viva-style questions.
• Quizzing you. Ask it to give you one question at a time and wait for your answer.

The last two matter most. A large review of study methods rated practice testing as highly useful, and rereading and summarising as low (Dunlosky et al., 2013). So asking AI for a summary is the weakest way to use it.

How do medical students use AI tools?

Five of the six Nigerian medical students we interviewed in October 2026 used AI tools. Three of them used it in ways worth copying:

• One shrinks his slides with iLovePDF, gives them to DeepSeek and asks for MCQs and theory questions. He also used Grok to make audio viva practice, with a 10-second pause before each answer.
• One records lectures with Granola and gives the transcript to Claude to turn into notes. When a slide is hard, they ask ChatGPT or Claude to explain it.
• One makes flashcards with NotebookLM. For pharmacology, they ask a chatbot to start with how the body normally works, then explain what the drug changes.

Where does AI go wrong for medical students?

It makes confident mistakes

References are a known weak spot. In one study, ChatGPT-3.5 wrote 30 short medical papers with 115 references: 47% of the references were made up, 46% were real but had wrong details, and only 7% were real and correct (Bhattacharyya et al., 2023). Newer models may do better, but never quote a reference from a chatbot without finding it yourself.

It answers from other countries' guidelines

Ask about TB treatment and you may get "RIPE", the way American texts teach it. Nigeria's national guidelines write the same drugs as 2(RHZE)/4(RH): two months of rifampicin, isoniazid, pyrazinamide and ethambutol, then four months of rifampicin and isoniazid. Your examiners will expect the Nigerian form. Tests can differ too: in Nigeria, GeneXpert MTB/RIF is the first test for anyone with presumptive TB.

It doesn't know what your lecturers set

A chatbot can't see your past questions or know which topics your department likes. It may spend pages on something your lecturers never ask and skip what they always ask. Give it your own slides so it works from what you were taught.

It can't replace practice under exam conditions

Reading a good AI explanation feels like learning. The exam is answering questions alone, against the clock. Use AI to make the questions, then put it away and answer them.

Is DeepSeek better than ChatGPT for medical students?

There's no clear answer. Studies from 2025 point both ways, depending on the exam and the language:

None of the studies used Nigerian or West African questions, and the models have changed since.

Pick the one that works for you. One student we interviewed chose based on which tool handled his uploads, not on test scores. How you use it matters more than which one you pick.

How do you use AI safely for medical school?

1. Give it your own slides or notes. Ask it to work only from them. You'll get less off-syllabus material.
2. Ask for questions, not just summaries. "Write 10 MCQs from this" teaches you more than "summarise this".
3. Check facts against your textbook or guideline. Especially drug regimens, doses and anything you'll write in an exam.
4. Check every reference yourself. If you can't find it on PubMed or the journal's site, don't use it.
5. Never paste anything that could identify a patient. No names, hospital numbers, photos of folders or case notes with details. Change or remove details before you ask about a case.

What are good ChatGPT prompts for medical students?

Try these with ChatGPT, DeepSeek or any other chatbot:

• "Using only this slide, explain it to me as if I'm in second year. Then give me one example."
• "From these notes, write 10 single-best-answer MCQs with five options each. Don't show the answers until I reply."
• "Ask me viva questions on this topic one at a time. Wait for my answer, then tell me what I missed."
• "I'm studying in Nigeria. Write the answer the way the Nigerian national guidelines would, and tell me if you're not sure."

The last prompt makes a Nigerian answer more likely, not certain, so still check it.

Practise it

Our tuberculosis module is written and sourced by physicians in line with Nigeria's national guidelines, so you can practise TB questions with the regimen your examiners expect.