Al-driven coaching for stress management: exploring the role of empathic design

AbstractThis study examined whether empathy influences the effectiveness of a chatbot-based CBSM coaching intervention (Traut-Mattausch et al., 2021). Seventy adults (M = 24.9 years; 64% women) were randomly assigned to an empathic or standardized version of a GPT-based chatbot that conducted three coaching sessions over two weeks. The primary outcome was perceived stress, which was measured at six time points. Secondary outcomes included perceived empathy, social presence, adherence and goal attainment. The participants rated the empathic chatbot as significantly more empathic (d = 1.96, p < .001) and more socially present (ß = .63; p < .001). Contrary to the primary hypothesis, no significant difference in stress reduction was found between groups, F(5, 340) = 1.95, p = .086. Both conditions showed significant within-group improvements in goal attainment (empathic: ΔM = 3.53, d = 1.07; standardized: ΔM = 2.76, d = 1.11), but the difference between groups was not significant, t(64,50) = -1.10, p = 0.280. A postulated serial mediation model supported an indirect effect of empathic cues on stress reduction via social presence and adherence (ß = .10; p = .036; R² = .16). The empathic chatbot led to higher perceived empathy, social presence, and adherence. However, this model does not consistently lead to better short-term stress outcomes for brief, scripted interventions.Keywords: artificial intelligence, chatbot, empathy, social presence, stress management, coaching, adherence, digital health