User Acceptance of an AI-Powered Medical History-Taking Training System Among Undergraduate Medical Students: Mixed...

AI-powered virtual patient systems provide medical students with repeatable practice environments for history-taking training. However, user acceptance of such systems and the experience dimensions associated with that acceptance lack mixed methods evidence.

This study aimed to (1) examine the associations of system experience and learning experience/intrinsic motivation with overall acceptance among undergraduate medical students using an AI-powered medical history-taking training and evaluation system (AMTES), (2) explore user experience patterns through open-ended questions, and (3) integrate quantitative and qualitative findings to inform system refinement and pedagogical implementation.

A cross-sectional convergent mixed methods design was used. A total of 66 undergraduate medical students at a Chinese medical college completed a postuse questionnaire after AMTES training. The primary outcome was an overall acceptance composite combining use intention, recommendation intention, and overall satisfaction. Associations with system experience and learning experience/intrinsic motivation were examined using linear regression with heteroscedasticity-consistent SEs (type 3) and bootstrap CIs. Sensitivity analyses included covariate adjustment, single-outcome models, a fractional logit model, and content-overlap sensitivity checks for the system-experience composite. Open-ended responses were analyzed using codebook-oriented thematic analysis and integrated with quantitative findings through a joint display.

Both system experience and learning experience/intrinsic motivation were positively associated with overall acceptance (standardized β=.526; P<.001 and standardized β=.377; P=.002, respectively; R²=0.662). Sensitivity analyses supported the robustness of both positive associations. Qualitative analysis showed that the most frequently nominated benefits clustered within the "Practice Accessibility and Feedback Support" theme, particularly self-directed practice (28/66, 42.4%) and immediate feedback (25/66, 37.9%). Refinement priorities clustered around 4 dialogue-quality and assessment dimensions, including semantic understanding, contextual consistency, conversational naturalness, and scoring logic, each cited by 21%-35% of respondents. The mixed methods joint display indicated contextual (same-source) alignment on dialogue and scoring concerns and suggested that perceived scoring-feedback discrepancies may be associated with lower student trust in the feedback function. This cross-dimensional pattern was not apparent from the quantitative model alone.

In this exploratory cohort of undergraduate medical students, both system interaction quality and perceived learning value were positively associated with overall acceptance of AMTES, with system interaction quality showing the stronger association within the study's measurement specification. Dialogue coherence, semantic understanding, and scoring-feedback alignment emerged as the most frequently nominated refinement priorities and are plausible candidate targets for improving acceptance-related perceptions. This study emphasizes implementation-level acceptance rather than solely technical reliability or educational effectiveness. Interaction-quality problems may be associated with less favorable acceptance even when learning value is recognized. These findings may inform system refinement priorities and the curricular integration of AI-powered history-taking training systems, while larger multicenter and longitudinal studies are needed to examine whether improvements in interaction quality translate into gains in learner acceptance and downstream training outcomes.