Related Experiment Video
Updated: Oct 2, 2026
Virtual Agent for Real-Time Motivational Interviewing by Integrating Adaptive Nonverbal Behavior and Language Models
Published on: December 23, 2025
Doctor of physical therapy students' perceptions of artificial intelligence chatbots versus peer role play for
Kelby B Kaplan1, Alisa M Pravdo1, Rachel Reoli1
1Department of Physical Therapy and Rehabilitation Science, University of Maryland Baltimore, Baltimore, MD, USA.
Introduction:
Therapeutic interviewing supports rapport building, diagnostic accuracy, and person-centered care. Peer role play (PRP) is used in Doctor of Physical Therapy (DPT) education for rehearsing patient-provider interactions; however, there are limitations regarding patient portrayal. Artificial intelligence chatbots (AI) may offer an alternative. This study examined DPT students' perceptions and quality of information using AI for therapeutic interviewing compared with PRP.
Methods:
This retrospective mixed methods cohort study analyzed data from two simulated interviewing activities involving AI and PRP cases. Quantitative data were analyzed descriptively and with Mann-Whitney U tests; qualitative responses underwent thematic analysis.
Results:
68 DPT students participated in Activity 1, and 58 students (85.3%) completed an anonymous survey. Seventy students completed Activity 2 and documentation checklist. Four themes highlighted AI advantages: realistic rapport and conversation, active listening and reflective responses, reduced performance anxiety, unlimited independent practice. Six themes identified AI disadvantages: robotic/disingenuous interactions, technological glitches, lack of non-verbal communication, impediments to conversational flow, environmental/social ethical concerns, lack of real-world pressure. PRP consistently demonstrated higher accuracy and fewer omissions compared to AI. AI led to higher rates of incorrect data regarding prior level of function (65.7% vs. 9.7% PRP), medical/surgical history (54.3% vs. 0% PRP), and current level of function (80% vs. 57.2% PRP). Medication was more accurate in PRP (87.5% correct) than AI (67.1% correct). Mann-Whitney U tests showed a significant difference for demographics (U = 1424, p = <0.001, alpha = 0.05), prior level of function (U = 672, p=<0.001, alpha = 0.05), patient goals (U = 2048, p = .044, alpha = 0.05), social history (U = 1756, p=<0.001, alpha = 0.05) general health, medical/surgical history (U = 1486, p=<0.001, alpha = 0.05), and medications (U = 1980, p = .002, alpha = 0.05).
Conclusions:
While AI mitigates student performance anxiety via a safe space to practice interviews, it also reduces documentation accuracy. DPT programs may benefit from utilizing AI as a supplemental tool; however, technological limitations and formatting inconsistencies prevent it from replacing PRP.
