For clinical skills leads, simulation leads and heads of dental education weighing how to deliver communication training at scale
It's Thursday afternoon in a dental school's clinical skills suite. Six final-year students sit around a table. In the middle of the room, one of them takes a seat opposite a trained simulated patient playing a 34-year-old who hasn't seen a dentist in six years, ever since a bad experience with a filling. The student has fifteen minutes to build trust, take a history, and explain a treatment plan without the patient bolting for the door.
The other five students watch. They'll get feedback, and some will get a turn of their own later in the term. But for this hour, in this room, only one student is actually practising.
That single detail sums up communication skills training in dentistry. The method with the best track record, trained simulated patients (SPs) delivering realistic, consistent, assessable encounters, is also the hardest method to give every student enough of. The evidence that SPs work is solid. The open question is dose: how much practice each individual student actually gets before they're in front of a real, anxious person in the chair.
This piece looks honestly at three ways dental schools currently teach communication (simulated patients, peer roleplay, and AI roleplay), including where the newest AI evidence genuinely holds up, and where it clearly doesn't.
Communication training isn't a soft add-on to the dental curriculum. It's what stands between a nervous patient and a treatment plan they'll actually follow through on.
In England, the Adult Oral Health Survey 2021 found that 42% of dentate adults scored between 10 and 18 on the Modified Dental Anxiety Scale (moderate anxiety), and a further 12% scored 19 or above, indicating extreme dental anxiety. Women reported markedly higher rates of extreme anxiety than men (16% versus 8%), and anxiety was highest among younger adults, falling from 14% at ages 16–24 to 6% among those 75 and over. That's roughly one in eight adults for whom a dental appointment is not a routine errand but something closer to an ordeal.
Globally, a 2021 meta-analysis in the Journal of Dentistry, pooling data from 72,577 people across 31 studies, put overall dental fear and anxiety prevalence at 15.3%, with 12.4% reporting high dental fear and anxiety and 3.3% reporting severe dental fear and anxiety.
These aren't abstract figures. They describe the patient sitting in the chair whom every graduating dentist has to be ready for, which is exactly why the UK's General Dental Council has elevated communication into its own stand-alone domain, “Interpersonal Skills,” within the new Safe Practitioner Framework, mandatory for cohorts starting from September 2025.
It requires graduates to “communicate with care, compassion, empathy and respect in all professional interactions.” Across Europe, the ADEE's Graduating European Dentist framework does similar work through its patient-centred care outcomes, and is now used by close to 60% of responding dental schools.
The standards have caught up with the stakes. The delivery model hasn't quite caught up with the standards.
Simulated patients are trained laypeople or actors who consistently portray a clinical scenario so students can practise, and be assessed, against defined criteria, typically as part of a structured pedagogical design that sequences preparation, practice and reflection. They're not new, and the evidence that they work in dentistry specifically is genuinely good.
A 2024 prospective cohort study from RWTH Aachen University tracked 34 fourth- and fifth-year dental students through a course combining one theory lecture with two practical SP sessions across four clinical vignettes. Students reported statistically significant improvements in self-rated communication skills, and significantly greater attention to asking open-ended questions. This is real, dental-specific evidence that structured SP training changes behaviour, not just confidence.
But read the same paper's methods section and the scalability problem is right there in black and white: in each small-group session, one student performed the conversation with the simulated patient, while the others observed passively. This is standard practice at most dental schools, not a flaw specific to Aachen's design. Trained actors, faculty time, and timetable slots are all finite, so someone has to watch rather than practise.
“The evidence that simulated patients work has never really been the problem. The problem is that only one student performs per session, while a room full of future dentists watches.”
The cost side backs this up. A widely cited 2009 survey of standardised patient programmes across US and Canadian medical schools, run by the Association of Standardized Patient Educators, found that over half of programmes hired between 51 and 100 SPs annually, paying an average of $15–$16 an hour for training and portrayal time, on top of an average of 4.8 permanent staff per programme just to run it.
A separate cost-effectiveness comparison found that delivering communication OSCEs with peer roleplay took 112 man-hours, against a markedly higher figure for SP-based delivery once actor training, coaching, and assessment-day staffing were included, for broadly comparable outcomes.
None of this means SPs aren't worth the money. A classic cost-benefit analysis of introducing SPs into a physical diagnosis programme found the roughly $43,800 outlay was offset by 608 hours of faculty teaching time saved, with no drop in OSCE performance. The economics can work. But they only work at the scale a programme can actually afford, and that scale is rarely every student, every skill, as often as they need it.
There's a second limitation worth naming honestly: even good SP training doesn't automatically stick. A University of Alabama at Birmingham study following students from a single second-year SP simulation through to their fourth-year clinical performance found only limited evidence that one exposure improved long-term communication with real patients. One dose, however well delivered, isn't the same as sustained practice.
If SPs are expensive, the obvious low-cost alternative is students practising on each other. The evidence here is more mixed than either enthusiasts or sceptics tend to admit.
A 2020 systematic review in Simulation in Healthcare, covering 22 studies, found peer roleplay was well received and did improve communication learning, but no more than other simulation methods, and no study in the review actually measured whether it changed real-world clinical behaviour. A separate review in the International Journal of Evaluation and Research in Education found peer roleplay produced OSCE scores comparable to SP training and was consistently preferred by students over straight lectures.
In dentistry specifically, a randomised trial at Leipzig University found that peer feedback improved undergraduate students' communication skills just as effectively as expert feedback. That's a genuinely useful finding, since it suggests structured peer practice belongs in the curriculum rather than being treated as a poor substitute.
The drawbacks are well documented too. Compared with trained actors, peer “patients” produce less emotional depth and more learner nervousness, according to a 2020 study in BMC Medical Education. Other research on peer roleplay in health professions education points to evaluation apprehension and self-consciousness. It's hard to fully commit to playing an anxious, difficult, or upset patient when you know the person across the table will be grading you, or vice versa, next week.
So peer roleplay earns its place. It's cheap, evidence-supported for skill acquisition, and works well for repetition. It is not, on the evidence, a realistic stand-in for what a genuinely anxious stranger behaves like.
The newest and most directly relevant evidence here is a 2026 randomised controlled trial from Chongqing Medical University, published in BMC Medical Education. Thirty-eight fourth-year dental students were randomised to practise with either an AI virtual patient (built on the DeepSeek large language model) or a peer roleplay partner, then assessed using the SEGUE framework, a validated 100-point communication skills scale.
The results favoured the AI group by a clear margin: a median SEGUE score of 70 versus 60 for peer roleplay, a statistically significant difference with a large effect size. Digging into where that gap came from is instructive. The AI group's biggest advantage was in structuring the consultation: setting the stage at the start and closing the encounter properly at the end.
On the domains involving gathering information, explaining information, and understanding the patient's perspective, there was no significant difference between the two groups.
Median SEGUE communication score (out of 100), fourth-year dental students, n=38. AI group scored significantly higher (p<0.001, large effect size).
That pattern is worth sitting with. AI roleplay, in this trial, was genuinely better at the structural, repeatable, “did you do the steps in the right order” parts of a consultation, exactly the kind of skill that benefits from unlimited, low-stakes repetition rather than one carefully choreographed encounter a term. It was not shown to be better, or even different, at the harder relational work of actually understanding what the patient in front of you is feeling.
Effect size (r) by SEGUE domain. Larger bar = bigger AI advantage. Domains covering information-gathering, information-giving and understanding the patient's perspective showed no significant difference.
The same paper is unusually candid about its own limits, which is precisely why it's a good source to build a comparison on rather than a marketing claim. The authors note that the AI's emotional responses “rely on predefined rules rather than adaptive inference,” producing what they describe as formulaic expressions that don't capture the real variability of how anxious patients actually behave.
Usability scored 65.9 out of 100 on a standard scale, just below the conventional 68-point threshold for “acceptable,” with students specifically marking down the AI's contextual flexibility and clinical authenticity. Most tellingly, the authors state directly that this kind of training “is not a substitute for clinical placements but serves as a supplementary instrument.”
Broader reviews of AI-driven virtual patients across health professions education land in the same place. A systematic review of generative-AI virtual patients found they support history-taking and clinical reasoning reasonably well, but fail to replicate non-verbal cues (pain indicators, gaze, culturally specific gestures) that are central to genuine empathy training. A scoping review of large language model-based virtual patients was blunter still: these tools “cannot replace real-world interactions.”
Being precise about this matters more than the sales pitch either way. Based on the current evidence, AI roleplay does not replicate touch or physical examination. Nothing about a conversational AI system substitutes for the intra-oral examination, palpation, or hands-on assessment that sits at the centre of a dental consultation; VR haptic simulators exist for procedural skills like cavity preparation, but that's a separate use case from communication practice, not an overlapping one.
Nor does it reliably capture emotional nuance. The Chongqing trial's own authors flag this as the current system's weakest point: scripted emotional responses rather than the adaptive, sometimes contradictory way real anxious patients behave. And it is not, on current evidence, a stand-in for high-stakes summative assessment.
With small trial samples, a single assessor, and usability scores hovering around the acceptability line, the evidence so far supports AI for formative, low-stakes rehearsal, not for OSCE-grade judgement calls about whether a student is safe to progress. Nor does it read non-verbal cues reliably: eye contact, gaze, subtle pain signals, and culturally specific body language remain largely outside what current systems can produce or interpret.
None of this argues for choosing one method and discarding the others. It argues for matching each method to what it's actually good at, and being honest about the gap that's currently unaddressed: volume.
Keep simulated patients where only a human will do. Summative OSCE-style assessment, emotionally complex scenarios like breaking bad news or managing an angry patient, and any encounter that depends on reading and returning genuine non-verbal cues should stay with trained SPs. Budget for this properly, using the published per-hour and per-programme figures as a starting point for local costings, rather than treating SP delivery as infinitely scalable if you just add more sessions.
Use AI and peer roleplay to close the practice gap. The Aachen finding, one active performer per SP session, is the clearest evidence that most cohorts are structurally under-practised in communication skills. AI virtual patients, like those in Bodyswaps' AI coaching, can absorb unlimited repetition of history-taking, consultation structure, and motivational interviewing-style conversations drawn from a scenario library, at whatever hour a student is actually free to practise. Peer roleplay, backed by the Leipzig findings on peer feedback, remains a legitimate and cheap way to build in-class repetition alongside it.
Map the blend explicitly to your standards, too. In the UK, that means showing how each method contributes to the GDC Safe Practitioner Framework's Interpersonal Skills domain. In Europe, it means the ADEE GED's patient-centred care outcomes. In the US and Canada, it means CODA and CDAC accreditation language. A blended model is easier to defend to an accreditor, and to your own quality assurance processes, when the mapping is explicit rather than assumed.
Finally, set a threshold for expanding AI's role rather than an open-ended one. Keep AI formative-only until local usability data clears the acceptability benchmark and students report genuine authenticity, not the marginal, formulaic reception flagged in the Chongqing trial. Only expand toward higher-stakes use once multi-rater OSCE data show AI-supplemented cohorts matching SP-trained cohorts on real assessment outcomes, not just self-reported confidence.
Simulated patients remain the gold standard for a reason, and nothing in the current evidence suggests AI roleplay is close to replacing them for the encounters that matter most: the emotionally complex, physically hands-on, high-stakes ones.
What the evidence does show is that AI roleplay is a credible, evidence-backed way to solve a problem SP programmes were never designed to solve on their own: giving every student in a cohort enough repetition to walk into a real consultation room having actually practised, not just watched someone else practise five times a term.
The question for clinical skills leads isn't simulated patients or AI. It's how to use each one for what it does best, and be clear-eyed about what still needs a trained human sitting across the table.
See how Bodyswaps supports dental communication training →
→ Bodyswaps for higher education
→ How Bodyswaps' AI coaching works
→ The pedagogy behind Bodyswaps
→ Bodyswaps scenario library
No. Current evidence shows AI roleplay is good for repeated, low-stakes practice, especially structuring a consultation. It does not reliably capture emotional nuance, touch, or physical examination, so simulated patients still matter for high-stakes and emotionally complex training.
Because trained actors, faculty time, and timetable slots are limited. Studies show that in a typical small-group session, only one student actively performs with the simulated patient while the rest of the group observes.
Research suggests peer roleplay produces similar skill gains and comparable OSCE scores to simulated patients, and is often preferred over lectures. However, it lacks the emotional realism and non-verbal cues of a trained actor.
In England, 12% of adults have extreme dental anxiety and 42% have moderate anxiety, according to the 2021 Adult Oral Health Survey. Globally, a 2021 meta-analysis found dental fear and anxiety affects around 15% of adults.