The next contest in medical AI education will concern what happens after a learner gets a question wrong. A fluent explanation is easy to appreciate. The more valuable intervention identifies why the learner chose that answer, changes the underlying reasoning and creates another opportunity to use the corrected principle.
OpenEvidence describes Preceptor as a Socratic tutor using simulated cases, part of its new Learn offering. Launch announcement. Our OpenEvidence Learn review covers the wider product, eligibility and access conditions.
iatroX approaches clinical learning through attempted questions, guided tutoring, available simulations and professional learning. The interesting comparison concerns the educational work these activities can support. Neither a product description nor a convincing conversation establishes how much a learner will retain or how well they will perform independently.
The same wrong answer can conceal different problems
Imagine three learners selecting the same incorrect response in an examination vignette.
The first cannot recall an important fact. The second remembers the fact but applies it to the wrong situation. The third understands the clinical distinction and misreads the question’s instruction. A fourth learner selects the correct answer while using an unreliable shortcut.
An explanation of the answer may help the first learner and leave the others’ problems intact. A dashboard organised solely around correctness also misses the fourth learner’s vulnerability. That person may appear ready until an unfamiliar case exposes the shortcut.
This is why the learner’s attempt is useful evidence. It provides something specific to investigate: a choice, a justification, a discarded alternative or a decision made too quickly. A tutor can then decide whether to explain, challenge, contrast or practise.
iatroX Tutor begins beside a question the learner has already attempted. It supports guided discussion and direct explanation, with editable highlights and cards available to preserve an insight. How iatroX Tutor works. This gives the conversation a defined starting point without requiring the learner to invent an effective teaching prompt.
Socratic questioning needs a teaching decision behind it
“Socratic” describes an interaction style. Its educational value depends on how the questions respond to the learner.
A helpful prompt might ask which finding would change an interpretation, or why the closest alternative fits less well. These questions can expose an assumption that remains hidden when a tutor asks only for the final answer.
There is a failure mode: increasingly obvious clues can lead someone to the intended answer while doing much of the reasoning for them. Completing the conversation then looks like success, although the learner has mainly practised following hints.
The opposite failure is endless questioning after the learner needs a clear explanation. Someone who lacks the relevant concept cannot reason effectively from information they have never learned. More questions can increase frustration without increasing understanding.
An effective tutor therefore needs to change its teaching move. Request an independent attempt, investigate the gap, explain when necessary, then return responsibility to the learner through a fresh decision. A shorter conversation can be more useful than a longer one if it resolves the actual confusion.
Better discussion and durable learning are different outcomes
A recent medical-education pilot illustrates why those distinctions matter. In an October 2026 crossover study, 44 second-year medical students used a curriculum-aligned custom GPT for one pathology case and conventional preparation for another. Faculty-rated preparation and discussion scores were higher with the custom GPT. Delayed examination-item retention did not differ significantly in the primary analysis. The small, voluntary, single-school study covered two cases and was not publicly preregistered. It evaluated that instructional package, not Preceptor or iatroX. Roark and colleagues, JMIR Medical Education.
The finding gives an education platform several separate questions to answer. Does the interaction help a learner explain their reasoning now? Can they make the decision later without assistance? Does the skill transfer to a case with different wording and fewer prompts?
A learner’s satisfaction and a trainer’s assessment of a discussion are useful measures. Each describes a particular outcome. A serious evaluation should also examine the independent performance that the learning activity was intended to develop.
Feedback must teach a distinction that survives the example
Consider a fictional question asking which piece of information best separates two explanations. After correcting the response, the tutor could present a second scenario in which the previously useful clue is absent or misleading. The learner must explain the distinction instead of recognising the original wording.
A useful follow-up could ask what additional information would change the choice, which competing interpretation remains plausible and how the next action would change. The aim is to establish the boundary of the principle, including the circumstances in which it stops being useful.
This is where a question bank and a tutor can reinforce each other. The bank supplies relevant decisions and alternatives. The conversation makes the learner’s interpretation visible. A later question provides another opportunity to discover whether that interpretation changed.
For connected learning platforms, a promising direction is to make those links increasingly precise: the misconception identified, the contrast that might challenge it and the fresh task that would provide evidence of progress. That is a product-development opportunity, rather than a claim that any current service has solved every step automatically.
Spaced review needs something useful to retrieve
A return to material becomes more purposeful when the learner understands what deserves attention.
A card repeating an isolated answer can strengthen recall of that answer. A card preserving a useful distinction can support a wider range of decisions. “What additional information would make the alternative explanation more likely?” calls for a different kind of retrieval from recognising a diagnosis beside a familiar vignette.
The learner should review the card as carefully as the original answer. An ambiguous prompt, an omitted condition or an oversimplified rule can make later practice less useful. Editing the retained item is part of learning, especially when an AI system helped create it.
Keeping a concise insight also gives a learner something manageable to revisit between longer sessions. The purpose is to make the important decision available again, with enough context to use it appropriately.
Simulation asks whether reasoning appears in an encounter
A multiple-choice question supplies a defined set of alternatives. In a simulated consultation, the learner must decide what to ask, which uncertainty to resolve and how to explain a plan. Knowing the relevant material and using it at the right moment are related abilities that require different practice.
An October 2026 randomized trial studied SOPHIE, a virtual patient for serious-illness communication. Fifty-one trainees and clinicians received interactive practice with feedback or equal-duration reading. Blinded assessments of subsequent human standardized-patient encounters found greater communication improvement after SOPHIE. Assessment was same-day, in one scenario; longer-term and real-patient effects remain unknown. Reading was the comparator, so the trial does not establish superiority over human coaching. It did not test either platform discussed here. Haut and colleagues, JMIR Medical Education.
That evaluation approach is useful: examine what a learner can carry into a separate encounter. Engagement with a virtual patient alone would not answer that question.
iatroX’s current MRCGP SCA practice supports voice and text, coached practice or uninterrupted exam mode, encounter-linked feedback and Tutor after the encounter. Those are concrete ways to rehearse and inspect decisions. Their value should be assessed through the learner’s subsequent performance, with clear separation between practice feedback and a validated prediction of examination readiness.
Examination context gives tutoring a purpose
The common learning problem travels across borders: clinicians must integrate information, choose among plausible alternatives and justify decisions. The assessment context changes how those capabilities are expressed.
A tutor needs to respect the selected examination and the task being assessed. A biological explanation may be correct yet incomplete preparation for a consultation requiring prioritisation, communication and a negotiated plan. Written preparation and practical rehearsal can therefore complement each other within one learning project.
iatroX’s exam directory provides pathways across the UK, US, Canada, Australia and Italy. That creates options for clinicians whose next assessment or intended country of practice changes. The useful comparison is how well a selected pathway serves the learner’s actual task; a long catalogue cannot establish that on its own.
Human educators add judgement to visible practice
An AI tutor can make repeated individual practice accessible between supervised sessions. Human educators contribute judgement about the learner, the setting and the significance of a recurring problem.
A trainee might bring a selected transcript, explanation or reflection to a trainer and ask why a particular decision keeps causing difficulty. The educator can examine the reasoning, identify a missing contextual issue and suggest a more demanding test. Practice becomes more useful when it supplies specific material for that conversation.
iatroX’s CPD workflow connects targeted questions, Tutor-supported exploration and a learner-reviewed record. The clinician edits and confirms the final reflection. Such a record can support discussion of a learning need; its existence should not be confused with automatic accreditation or independent evidence of competence.
Progress should be visible beyond the conversation
OpenEvidence’s education move makes the quality of AI tutoring a more prominent question. Platforms need to show how a conversation helps the learner do something independently afterwards.
iatroX has practical foundations for that work: attempted-question Tutor, examination pathways, SCA rehearsal and reviewed learning records. Together they offer a useful way to examine decisions and practise applying what a learner discovers.
For the learner, the next step is concrete. Finish a tutorial with a distinction you can explain, apply it to an unfamiliar task, and revisit it later without the original hints. An education platform earns its place by making that progression easier to pursue and assess.