Invisible Impact of Empathy on Behavioral Change: Isolating the Effect of Empathy in Long-term Physical Activity Coaching Chatbot Interactions

This study investigates whether empathy—a cornerstone of human counseling—actually drives behavioral change in digital physical activity (PA) coaching. While many LLM-powered chatbots are designed with empathy as a core feature, its specific impact on long-term user motivation and adherence remains under-researched. The authors conducted a six-week, within-subject study with 13 participants, comparing three versions of a WhatsApp-based PA coaching chatbot: Non-Empathetic, Standard (relying on the LLM's innate capabilities), and Empathetic (utilizing a specific clinical empathy framework to identify and respond to user negative self-disclosure).

The results reveal a complex, "invisible" role for empathy in digital coaching. Interestingly, participants often rated the Non-Empathetic chatbot as more useful and engaging, suggesting that users may prefer direct, task-oriented communication over perceived emotional labor. However, quantitative data told a different story: the Empathetic variant significantly outperformed the others in accelerating user self-efficacy and strengthening the intention to follow the chatbot's advice. Furthermore, the Empathetic condition was associated with higher trust and lower attrition rates over the six-week period.

These findings suggest that empathy in AI coaching acts as a subtle, peripheral influence rather than a direct, consciously recognized feature. For designers, this implies that while users may not explicitly praise empathetic responses, these interactions are crucial for building the underlying rapport and motivation necessary for sustained behavioral change. The study cautions against relying solely on user satisfaction surveys to evaluate chatbot effectiveness, as users' conscious preferences may not align with the interventions that actually drive their long-term success.