Medicine already knows how a consultation skill is built. A student practises under close watch. An intern works with a supervisor nearby. A registrar works more independently, and a specialist works alone. Each step is the same method at a different distance: practice, watched by someone who knows the job, followed by feedback.
The profession trusts that method enough to use it when things go wrong. When a doctor's competence falls below the standard, the answer is more supervised practice.
So nobody needs convincing that practice with feedback works. The hard question is where enough of it comes from. Real patients are rarely watched, actors are paid by the hour and peer practice is uneven. Artificial intelligence (AI) is the first way to make repetitions cheap and available at any hour. It doesn't replace the others. It carries the volume.
Medicine builds skill through graded supervision
In New Zealand, interns practise under supervision as part of the Medical Council's prevocational training programme. So do doctors who trained overseas when they first register, so that their performance can be assessed while they learn the New Zealand system (Medical Council of New Zealand). Registrars train in specialist programmes under supervisors, as we described in why postgraduate training has a different problem from medical school.
The supervision steps back as competence grows. Independence is earned through practice that someone has watched and commented on.
The evidence on communication training points the same way. An overview of 12 systematic reviews of physician communication training found the most effective strategies were "role-play, feedback, and small group discussions". Lectures, modelling and written information worked best as support (Berkhof et al., 2011, Patient Education and Counseling). The skill is built by doing it.
When a doctor struggles, supervision comes back
The same logic runs right through a career. When the Medical Council of New Zealand finds that a doctor's competence has fallen below the required standard, it must make one or more orders. One is an educational programme, typically 12 months, with an educational supervisor in the doctor's area of medicine who reports on their progress (Medical Council of New Zealand).
That tells you what the profession believes. Skill is built, and rebuilt, through supervised practice with feedback. The method isn't in question. The supply is.
Every source of practice has a limit
Real patients are where the skill is used, but they are seldom watched. Of 71 students surveyed on paediatric clerkships at six institutions in the United States, about a third said they had never been observed taking a history by an inpatient attending physician (Alex et al., 2024, Academic Medicine).
Simulated patients, played by trained actors, are realistic and consistent, and can be physically examined. They are also costly. One Australian medical programme notes that simulated patient programmes "require significant administrative and financial investment", with trained professional actors paid more than A$80 an hour (Taylor et al., 2019, Journal of Educational Evaluation for Health Professions).
Peers cost little and work better than their reputation. In a randomised trial in Germany, students trained with peer role-play and students trained with simulated patients both did better on an Objective Structured Clinical Examination (OSCE) than a control group (Bosse et al., 2012, Patient Education and Counseling). But peer practice depends on someone organising it, and a classmate is never quite a patient.
None of these is a flaw. Each does a real job. Together, they still don't add up to the number of watched, corrected consultations a learner needs.
AI makes repetitions cheap, within limits
AI changes the supply. A learner can practise a consultation at any hour, as often as they like, and get feedback each time.
The early evidence is encouraging, and small. In a trial in Vietnam, 34 medical students practised history-taking with AI simulated patients and 33 practised with standardised patients played by people. Their OSCE scores were similar (6.5 and 6.3), and the authors found no significant difference in learning outcomes (Dang et al., 2026, BMC Medical Education). One small trial doesn't settle the question, but it points the right way.
The limits are real too. An AI patient can't be physically examined. Its feedback isn't always right. In a study of 106 history-taking conversations, feedback from a large language model agreed closely with a human rater overall, but agreement was weaker in 8 of the feedback categories (Holderried et al., 2024, JMIR Medical Education). Feedback needs checking against the standard it claims to apply.
Each source of practice has a job
The answer is to give each the job it does best.
| Best at | Main limit | What drives the cost | |
|---|---|---|---|
| Real patients | The stakes and unpredictability of real care | Seldom observed, so feedback is often missing | Supervisor time |
| Simulated patients | Realistic, consistent encounters, including examination | Limited by bookings and budget | Actors, paid by the hour |
| Peers | Frequent, low-stakes practice | Depends on organisation; less realistic | Student time and the structure to run it |
| AI simulated patients | Volume, at any hour, for history-taking and communication | No physical examination; feedback needs checking | Low per attempt |
AI carries the volume, so that the costly, human parts of training can go where they matter most: examination, real patients and the supervisor's judgement. That's the pattern we set out in what pro-human AI means in medical education.
Where Gestalt fits
Gestalt builds consultation practice with formative feedback. A learner works through a realistic consultation and gets specific feedback on how they reasoned and communicated.
That makes it one source of repetitions among several, and the one that can be there whenever the learner is. Practice stays private unless the learner chooses to share it, and an institution sees a cohort-level picture. Feedback is formative. A human examiner makes every summative decision.
Medicine has always built skill through watched, corrected practice. The opportunity now is to make sure every learner gets enough of it.