Virtual Coaching for Mental Health: Effects of Coach Ethnicity Match, Coach Type Framing, Warmth, and Competence in a VR Setting

With the increasing integration of virtual coaching technologies and AI-driven interventions into psychological well-being applications, it has become essential to understand how users perceive and emotionally respond to virtual coaches. Drawing on Social Identity Theory (SIT), the Computers Are Social Actors (CASA) model, the Stereotype Content Model (SCM), and Parasocial Interaction (PSI) theory, this study examined the effect of coach ethnicity match (human vs. AI) and coach type framing (human vs. AI) on participants' mental health state (anxiety and positive emotion) in a VR context. This study employed a 2 (coach ethnicity match: matched vs. unmatched) × 2 (coach type framing: human vs. AI) between-subjects design, and 151 participants were randomly assigned to one of four VR coaching conditions. Participants in all conditions received an identical standardized coaching session delivered by a VR coach whose ethnicity and role framing varied according to the experimental condition. The participants then completed questionnaires that measured their perceived competence and warmth of the coach, as well as their mental health state (anxiety and positive emotion). The results revealed a significant interaction effect between coach ethnicity match and coach type framing on anxiety: participants experienced less anxiety when the coach matched their ethnicity and was framed as human. However, neither experimental manipulation showed a significant main effect when analyzed independently. Additionally, a general reduction in anxiety overtime was observed. Although coach ethnicity match and coach type framing did not significantly affect perception towards VR coach(perceived warmth and competence), perception towards VR coach were found to strongly predict positive emotion, yet failed to reduce anxiety alone significantly. These findings partially support the SCM and suggest that people are emotionally sensitive to interpersonal traits, particularly in the development of trust and positive effects. Interestingly, although identity cues such as coach ethnicity and coach type were not strong predictors on their own, the participants were still socially and emotionally reacting to the VR coach, supporting PSI theory. Users appeared to form parasocial bonds with the coach based on relational cues, with warmth and competence being the most important factors in creating emotional interest, even without direct human interaction. This study contributes to the human-agent interaction literature by illuminating when and how social identity cues and relational qualities influence user experience in a VR environment. This study offers practical insights into how emotionally intelligent virtual agents can be designed, based on the conclusion that relational qualities are more effective than surface-level identity signals in eliciting emotional support, particularly in task-oriented and digital therapeutic contexts.