Does Environment Matter in the Conversational Training «XARA»?

While conversation training has come to increasingly rely on virtual technologies, there are still few direct comparisons between immersive and less immersive virtual environments. The present study examines how an immersive virtual environment (IVE) and a low-immersive one (L-IVE) affect experiential, behavioral, and evaluative outcomes in multi-session conversation training. Using the AI-based conversational training application XARA and a within-subjects design, 49 human resources professionals and managers completed almost identical conversation modules (on recruiting, conflict, and resignation) across both environments. Spatial presence, conversational competence, conversational experience, ecological validity, and user preference were assessed after each module, and objective conversational behavior was assessed based on the conversation logs. The results showed consistently higher spatial presence in the IVE compared to the L-IVE. In contrast, conversational competence and conversational behavior did not differ between environments and varied systematically across modules. Conversational experience showed a differentiated pattern, with higher ratings for several dimensions in the IVE, although the effects were not consistent. Ecological validity, operationalized as perceived verisimilitude, did not differ between environments and was rated as moderate overall. Despite comparable conversational outcomes, the participants preferred the IVE. Limitations should also be acknowledged, including as the absence of a comparison with a real-life control condition and the restricted conversational scope. This study contributes to improving our understanding of human behavior in virtual extended environments and human–AI interaction. Looking ahead, the future development of virtual extended environment-based conversation training will require closer examination of the role of ecological validity alongside technological immersion.