Abstract
Purpose: Physical inactivity remains a major public health issue globally, highlighting the need for effective, scalable, and cost-efficient interventions. In recent years, artificial intelligence (AI)-powered tools, such as chatbots and virtual coaches, have emerged as innovative approaches to support physical activity (PA) practice. However, their effectiveness compared to traditional strategies is still not well investigated. This systematic review aimed at exploring the characteristics, behavioural strategies, and effectiveness of AI-based conversational agents in promoting PA and to compare their outcomes with traditional interventions.
Methods: A systematic search was conducted in Scopus, Web of Science, PubMed, and Cochrane databases up to May 2025. Eight interventional studies were included, all of which used AI-powered chatbots or virtual agents with the goal of increasing PA levels. Risk of bias was assessed using ROBINS-I and RoB 2 tools. For each study, intervention characteristics, applied behavioural techniques, PA-related outcomes, and user engagement were extracted and interventions.
Results: AI-mediated interventions commonly integrated evidencebased behaviour change techniques including goal setting, feedback, and motivational support. Several studies reported positive effects on PA indicators such as step count and minutes of moderate-to-vigorous physical activity, although findings varied considerably. Overall, user engagement and system usability were rated positively, especially in interventions that included relational or empathic features (e.g., social dialogue and humour). Compared to traditional PA interventions, AI-driven solutions offered advantages in terms of scalability and user autonomy, but often lacked methodological rigor and long-term outcome assessments.
Conclusions: AI-powered conversational agents and chatbots appear to be promising tools for promoting PA, particularly in in reducing dependence on human resources and enabling broader, more accessible implementation. Nevertheless, future research should focus on methodologically robust extended follow-up periods, and the development of hybrid models that combine human support and AI-driven components to enhance effectiveness and user experience.