Virtual patient tools improved clinical reasoning in 11 of 19 experimental studies.[7]
Plackett R, et al.. BMC Med Educ 2022.
Voice-based simulated patient encounters
MedListo is a simulated patient encounter, by voice, in real time. You present out loud. The attending interrupts and probes. Afterwards you get structured written feedback on exactly what you said. At 2am. In private. As many times as you need.
One free encounter. No account required.
Oral presentation is one of the most assessed activities in medical school and one of the least explicitly taught. Students learn it by trial and error, mostly in public, mostly in front of the people writing their evaluations. The feedback that would fix it, frequent and specific narrative feedback, is the one thing nobody has time to give.
So practice where it is safe to be bad.
Real attendings do not wait politely. Neither does this one. Present and get probed, the way rounds actually feel.
Every score links to the exact words you said that produced it. Nothing is a black box.
Spaced repetition picks your next case from your weakest areas, not a random deck.
Bedside presentations, surgery rounds in ninety seconds, USMLE-style vignettes, and residency and fellowship interview prep.
Switch languages before the encounter starts.
Every encounter is scored and saved, so you can watch your trajectory instead of guessing at it.
We did not invent virtual patient practice. We made it available on demand. Here is what the published literature says about the method:
Virtual patient tools improved clinical reasoning in 11 of 19 experimental studies.[7]
Plackett R, et al.. BMC Med Educ 2022.
Generative-AI virtual patients beat control conditions in every controlled comparison in a 2026 review of 15 studies.[8]
Jiang J, et al.. J Med Internet Res 2026.
In a 2026 pilot, AI-generated feedback covered the case's key learning points in 20 of 20 instances on history and physical, against 39 of 60 for faculty feedback. Four students, five cases, so treat it as a signal, not proof.[9]
Fruitman H, et al.. Neurol Educ 2026.
When a Korean medical school built a voice-based AI patient for licensure prep, 97.7 percent of eligible students used it voluntarily. Nobody made them.[11]
Song JW, et al.. NPJ Digit Med 2026.
None of these studies is about MedListo. Ours has not been run yet, and we say so below. That is the difference between evidence and marketing.
Every rating carries the verbatim transcript segment that produced it, so a judgment can be checked against its source. MedListo rates observed readiness against the AAMC Core EPAs and produces the structured evidence an entrustment decision draws on. The decision itself belongs to a supervising physician. It stays there.[4][5] On the R2C2 model, the platform automates the content phase of feedback; the relationship and coaching phases stay with faculty.[6]
The closest published system, CPX-MATE at Yonsei, established that a voice-based virtual patient with automated evaluation can agree with expert raters at a Gwet’s AC1 of 0.916, at 12 to 78 cents a session.[10] What it did not examine is whether the instrument functions equivalently across learners. That question, measurement invariance with a pre-registered subgroup analysis, is the study we are looking for an academic partner to run.
If you run a clerkship or a simulation program, we would rather agree on what to measure before a pilot starts than report usage numbers afterwards.
Start the conversationAAMC, Core Entrustable Professional Activities for Entering Residency
Stated plainly, because you would find these limits anyway and they bear on how the tool should be used.
Scores are formative. Single encounters should inform coaching conversations, not grades. Aggregated across a rotation, the same data is more defensible than any one encounter.