JMIR Publications

ABSTRACT

Background:

Physical activity plays a crucial role in managing obesity, while achieving sustained behavioral change is challenging when support is confined to clinic-based interactions. AI-driven conversational agents have potential to provide scalable and continuous support; however, there is a paucity of evidence regarding theory-informed conversational agents specifically designed to promote physical activity in adults with obesity.

Objective:

This study aimed to translate the situation-specific theory of Physical Activity in Adults with Obesity (PAAO) into an AI-driven conversational agent designed to promote physical activity, and to evaluate its feasibility, acceptability, and preliminary effects.

Methods:

A four-stage mixed-methods development and evaluation process was employed, wherein evidence was mapped onto PAAO constructs (stage 1), refined through expert consultation (stage 2), and translated into a WeChat-delivered conversational agent (stage 3). The resulting intervention was then evaluated and refined (stage 4); 16 adults with obesity were randomized in a 1:1 ratio to either a 6-week intervention supported by the conversational agent or to usual care. The feasibility and acceptability of the AI agent were evaluated through the use of implementation indicators, satisfaction ratings, the System Usability Scale (SUS), and semistructured interviews.

Results:

The development process translated the constructs of PAAO theory into the functions of a conversational agent, incorporating elements such as episodic future thinking prompts, habit formation reminders, adaptive plan adjustments, and performance monitoring with personalized feedback. In the evaluation phase, recruitment rate was 76.2% (16/21), retention within the intervention group was 100%, and the task response rate was 77.4%; no intervention-related adverse events were reported. Mean satisfaction score was 4.41 (SD = 0.47) out of 5, and the mean SUS score was 71.25 (SD = 10.44). Qualitative analysis identified three primary themes: perceived behavioral change and motivational factors, appropriateness of conversational agent’s design and human–agent collaboration, and recommendations for content and functional improvement.

Conclusions:

This study proposed and initially validated an obesity-specific theory-to-agent development pathway. The agent, delivered via human-AI collaboration, holds the potential to offer a scalable tool for continuous support in physical activity and obesity management, and may serve as a replicable model for the development of conversational agents for other chronic conditions.