Background: Shared decision-making (SDM) is a core competency in patient-centered care, yet it is challenging for physicians. Simulation-based education with standardized patients (SPs) offers valuable practice opportunities for undergraduate medical education (UME), but is limited by cost, scalability, and variable feedback quality. Advances in large language model (LLM) technology and virtual patient (VP) design present a promising pathway toward accessible, realistic, and high-quality SDM skill training. Objective: This study aimed to develop LLM-powered, real-time interactive 3D avatars for SDM training in UME and to evaluate their acceptance among undergraduate medical students. Methods: We conducted a descriptive, cross-sectional mixed methods study at KU Leuven, Belgium (November-December 2025). The avatar system was developed through an iterative, interdisciplinary collaboration between medical educators, SDM experts, software developers, user experience designers, and AI engineers, grounded in prior work on physicians’ educational needs and SDM learning mechanisms. Development comprised prototype case authoring, #usability testing with medical students, and iterative refinement. The final system is a web-based platform featuring real-time, LLM-powered spoken dialogue with diverse, photorealistic 3D patient avatars; layered prompts configuring avatar personality, clinical scenario, and difficulty; an educator panel for case authoring; and automated AI-generated formative feedback structured around 6 phases of SDM. After using the system for 1 month, a convenience sample of UME students completed a 9-item, 5-point Likert scale questionnaire grounded in the technology acceptance model (TAM), assessing perceived usefulness, perceived ease of use, and intention to use. Optional open-text responses were narratively analyzed to triangulate these findings. Results: Of 25 recruited participants, 24 were included in the analysis. Internal consistency was acceptable for perceived usefulness (α=0.76) and intention to use (α=0.78), but not for perceived ease of use (α=0.69). All construct means exceeded the scale midpoint of 3, suggesting high overall acceptance. Students rated perceived usefulness highest (mean 4.43, SD 0.50), followed by intention to use (mean 4.42, SD 0.59), and perceived ease of use was the lowest (mean 3.81, SD 0.68). Fourteen participants provided open-text responses. These highlighted the value of a safe, low-pressure environment for practicing communication without patients or peers present; reported mixed experiences with avatar realism, with reservations directed at behavioral rather than physical realism; and identified technical difficulties (microphone issues, freezing, and crashes) as the most frequent area for improvement. Conclusions: LLM-powered, real-time interactive 3D avatars are feasible and well-accepted tools for SDM training in UME. Students found the system educationally valuable and intended to use it again. Technical reliability and behavioral realism should be prioritized in future development. The modular architecture and automated feedback pipeline support scalable deployment across disciplines, institutions, and languages. Larger, multi-institutional studies with objective learning outcome measures are needed to establish efficacy. Trial Registration:

Background: Effective history taking helps clinicians identify key symptoms and form accurate hypotheses. Generative artificial intelligence (GenAI)–based virtual patients (VPs) are increasingly used…

Background: Standardized patients (SPs) prepare medical students for difficult conversations with patients. Despite their value, SP-based simulation training is constrained by available resources and competing clinical demands. Researchers are turning to artificial intelligence (#AI) and large language models, such as generative pretrained transformers, to create communication training that incorporates virtual simulated patients (VSPs). GPT-4 is a large language model advance allowing developers to design virtual simulation scenarios using text-based prompts instead of relying on branching path simulations with prescripted dialogue. These nascent developmental practices have not taken root in the literature to guide other researchers in developing their own simulations. Objective: This study aims to describe our developmental process and lessons learned for creating a GPT-4–driven VSP. We designed the VSP to help medical student learners rehearse discussing abnormal mammography results with a patient as a primary care physician (PCP). We aimed to assess GPT-4’s ability to generate appropriate VSP responses to learners during spoken conversations and provide appropriate feedback on learner performance. Methods: A research team comprised of physicians, a medical student, an educator, an SP program director, a learning experience designer, and a health care researcher conducted the study. A formative phase with in-depth knowledge user interviews informed development, followed by a development phase to create the virtual training module. The team conducted interviews with 5 medical students, 5 PCPs, and 5 breast #Cancer survivors. They then developed a VSP using simulation authoring software and provided the GPT-4–enabled VSP with an initial prompt consisting of a scenario description, emotional state, and expectations for learner dialogue. It was iteratively refined through an agile design process involving repeated cycles of testing, documenting issues, and revising the prompt. As an exploratory feature, the simulation used GPT-4 to provide written feedback to learners about their performance communicating with the VSP and their adherence to guidelines for difficult conversations. Results: In-depth interviews helped establish the appropriate timing, mode of communication, and protocol for conversations between PCPs and patients during the breast #Cancer screening process. The scenario simulated a telephone call between a physician and patient to discuss the abnormal results of a diagnostic mammogram that that indicated a need for a biopsy. Preliminary testing was promising. The VSP asked sensible questions about their mammography results and responded to learner inquiries using a voice replete with appropriate emotional inflections. GPT-4 generated performance feedback that successfully identified strengths and areas for improvement using relevant quotes from the learner-VSP conversation, but it occasionally misidentified learner adherence to communication protocols. Conclusions: GPT-4 streamlined development and facilitated more dynamic, humanlike interactions between learners and the VSP compared to branching path simulations. For the next steps, we will pilot-test the VSP with medical students to evaluate its #feasibility and acceptability.