Enhancing Physical Activity Among University Students in China: A Parallel-Group Quasi-Randomized Controlled Trial of an AI Chatbot Intervention Assessing Engagement, Usability, and Efficacy (Preprint)

BACKGROUND Physical inactivity among university students is a prevalent global health concern with long-term implications. Emerging technologies, such as wearable devices and AI-driven chatbots, offer promising avenues to promote physical activity (PA). However, empirical studies assessing the engagement, usability, and efficacy of next-generation large language model-based AI chatbots remain limited. OBJECTIVE This study aimed to evaluate the engagement, usability, and efficacy of an AI Coach intervention, based on the Doubao large language model, in enhancing PA levels among university students in China. METHODS A parallel-group quasi-randomized controlled trial was conducted over 8 weeks with 60 eligible university student participants (intervention group, n=30; control group, n=30). The intervention group used both their wearable devices and the AI Coach delivered via WeChat, while the control group used wearable devices only. Outcome measures included system-recorded interaction metrics to assess engagement, the System Usability Scale (SUS) and user experience questionnaires for usability, and changes in average daily step counts and weekly moderate-to-vigorous physical activity (MVPA) duration for efficacy. Data were analyzed using mixed-design ANOVA and appropriate non-parametric tests. RESULTS Participants engaged actively with the AI Coach, exchanging an average of 17.20 conversation rounds (SD = 8.72), with 60.27% of conversations initiated by participants. The average conversation duration was approximately 3 minutes. The mean SUS score was 85.75 (SD = 3.99), indicating excellent usability. Compared to the control group, the intervention group showed significant increases in average daily steps (mean increase 557 steps; p < 0.001) and weekly MVPA duration (mean increase 8.73 minutes; p < 0.001). No adverse events were reported. CONCLUSIONS The Doubao-based AI Coach is a safe, practical, and efficacious tool to enhance PA among university students. Its high engagement and usability suggest potential for scalable digital health interventions. This study highlights the value of combining AI chatbots with wearable devices to deliver personalized, real-time coaching. Future research should focus on automating data collection, expanding to diverse populations, and exploring user experiences to improve the intervention’s effectiveness and broader public health impact. CLINICALTRIAL Ethical approval was obtained from the Research Ethics Committee of Jiangxi University of Technology (JXUT.TNC2/JXUTREC2409).