more than 40 million people ask ChatGPT for health advice every day, according to OpenAI.

and in a new Oxford trial, people using chatbots to work out what was wrong did no better than people without them.

so AI is already in medicine, whether your curriculum likes it or not. the real question for medical and nursing students is which AI tools help you learn, and where they quietly make you worse.

What's happening with AI in healthcare right now

three things from the last two months:

→ ChatGPT Health launched on 7 January, a health space inside ChatGPT that can connect to medical records and wellness apps (Gizmodo). record linking is US only, and the features aren't offered in the UK, EU or Switzerland → OpenEvidence raised $250M at a $12B valuation in January, and says more than 40% of US physicians use its clinical search tool (HLTH) → misuse of AI chatbots topped ECRI's list of health technology hazards for 2026 (Betsy Lehman Center). ECRI, a US patient safety organisation, tested everyday chatbots like ChatGPT, Claude and Grok

then the reality check. Oxford researchers published a randomised study in Nature Medicine with 1,298 UK participants working through ten medical scenarios. people using chatbots were no better than the control group at judging how urgent a case was, and worse at spotting the relevant conditions.

the wild part, from the preprint version:

  1. the models alone identified the relevant conditions 94.9% of the time
  2. people USING the models got there in under 34.5% of cases
  3. and chose the right course of action in under 44.2%

the knowledge was in the machine. it got lost in the conversation.

What med school won't teach you about AI tools

medical and nursing curricula are packed. AI usually gets a lecture on ethics, if that. what's missing:

  1. how to question a confident answer. chatbots sound like a consultant even when they're wrong
  2. what information a model needs. the Oxford participants often didn't know what to tell it
  3. the difference between a study aid and a clinical tool. a chatbot that helps you revise renal physiology is not the thing that decides a dose
  4. data rules. patient details never go into a public chatbot. your placement trust's information governance policy will say so, so read it before day one

The AI toolkit for medical and nursing students

1. a grounded study assistant

→ use: upload your own lecture notes or guidelines and ask questions only about them. NotebookLM works this way, with citations back to your sources → why: answers stay tied to material you trust → limit: if your notes are wrong, so is it

2. a general chatbot as a sparring partner

→ use: "explain the renin angiotensin system like I'm in first year, then quiz me". or have it play the patient in a mock OSCE station (the practical exams with actors) → why: unlimited practice and active recall → limit: the Oxford study shows conversation is exactly where things go wrong, so treat every clinical fact as unverified

3. clinical evidence search

→ use: tools like OpenEvidence are built for clinicians searching the medical literature. access rules vary, so check what your school or placement provides → why: it shows you how qualified clinicians are starting to look things up → limit: a summary of a paper is not the paper. open the source

4. your verification layer (not AI, and that's the point)

→ use: NICE CKS, the BNF and your local trust guidelines → why: this is what you'll actually be held to on the ward

(and learn what ChatGPT Health can do, because patients will turn up having asked it first.)

The critical-thinking move: 4 checks for any AI answer

  1. would I bet a patient on this? if not, it's a hypothesis, not an answer
  2. what did I leave out of my question? age, history, medication, timing. the Oxford participants struggled right here
  3. where does the guideline disagree? check NICE CKS or the BNF, every single time
  4. am I learning or outsourcing? the same pattern hit junior developers: people who let AI do the work learnt less than people who kept asking "why"

What to do this month

  1. build a grounded notebook from one module's lecture notes and test yourself with it before your next exam
  2. run a mock OSCE with a chatbot, then check every clinical fact it gave you against NICE CKS
  3. catch three AI errors in your subject and write down why each one was wrong (honestly the best revision you'll do)
  4. read your school's AI policy and your placement trust's data rules. as we found across universities, the rules vary module by module

we'll keep covering this on our healthcare hub.

so TLDR: use AI to explain, quiz and organise, never to decide, and always check it against the guideline you'll actually be held to.

which AI tool has been the most useful (or the most wrong) in your revision so far?

Sources

Get the next issue by email

theVANTAGEd is written by students to bring critical thinking to AI across real industries. One issue every two weeks.