Clinical placements are where knowledge becomes judgement, and AI on the ward can push in either direction: it can deepen the learning loop around every patient you see, or it can quietly outsource the very reasoning you are there to build. The difference is not the tool. It is where in the loop you deploy it, and two hard boundaries you never cross.
The two boundaries first
Boundary one: patient data. Never enter identifiable patient information into any AI tool, full stop; that includes names, dates of birth, hospital numbers, and combinations of details that could identify someone. Reflect on cases in genuinely anonymised terms or not at all.
Boundary two: decision-making. As a student, AI output is not a basis for real-time clinical decisions, and suggestions affecting actual patients go through the team, not around it. An educational tutor and a point-of-care clinical tool are different categories: the first exists to change you, the second to support qualified decision-makers, and a student on placement is squarely in the market for the first.
Before the placement
Use AI to arrive less lost. Ask a tutor to quiz you on the common presentations of the specialty, run a question-bank block on its core conditions, and rehearse the structure of its typical assessments. Ten adaptive questions on heart failure before a cardiology block converts week one from vocabulary acquisition into actual learning.
During the day
The high-yield habit is the deferred question list. When something puzzles you on the ward round, capture it in five words and keep moving; placement time is for watching, doing and asking humans. At lunch or on the commute, work the list: a guideline-grounded answer for the factual ones, so the trail ends in NICE or CKS rather than model vibes, and your own differential first for the reasoning ones, before any AI opines. Practising differentials on cases you have genuinely seen, anonymised, with an AI that asks what you would do next, is the single best use of these tools on placement.
After the case
Post-case reflection is where AI earns its place. Take the anonymised case, write your reasoning as it actually happened, then let a Socratic tutor interrogate it: what argued against your leading diagnosis, what would have changed management, what did the registrar see that you did not. Then close the loop with retrieval: turn the case's teaching points into question-bank practice so the learning survives the week. This is the continuity a platform like iatroX is built for, preparing with the bank, clarifying against cited guidance with askiatroX, and letting spaced repetition carry the case forward, but the loop matters more than the logo.
The self-test
One question keeps you honest: is the AI making your reasoning more examined, or less necessary? Tools that ask you questions, demand your differential first and force you back to primary guidance are building a clinician. Tools that hand you conclusions you then repeat are building a courier. On placement, of all places, be the clinician.
Frequently asked questions
Can I cite an AI answer on a ward round or in a presentation?
Cite what the AI cited, never the AI. "askiatroX says" is not a source; the NICE guideline it linked is. This is not pedantry: the habit of resolving every claim to its primary source is precisely the appraisal skill placements exist to build, and it is also how you catch the occasions any AI is wrong before a consultant does.
When should I ask the AI versus ask the team?
Default to humans for anything touching a real patient's care, for local practice, and for the questions where the answer matters less than the discussion. The AI's honest niches on placement are the questions too small to interrupt a busy registrar for, the fifth "why" you were embarrassed to keep asking, and the post-case reflection no one has time to supervise. A useful rule: the AI gets your curiosity, the team gets your uncertainty.
Should I tell anyone I am using AI to study?
There is nothing to hide about educational use within the boundaries above, and your school or trust may have explicit policies worth knowing. What deserves active caution is the appearance of real-time reliance: consulting a chatbot mid-consultation reads very differently from working a case afterwards, whatever you were actually doing.
What about using AI for portfolio reflections and write-ups?
Drafting assistance is widely acceptable where your institution permits it; outsourced reflection defeats itself, since the reflection is the learning event, not the paperwork of it. A defensible pattern: write the substance yourself, let AI challenge and tighten it, and keep every clinical detail anonymised exactly as you would anywhere else.