GenAI avatars in VR: Role of presence, health, and technological factors

In higher education, VR environments increasingly include Generative AI (GenAI) avatars, but it is unclear how user expectations, presence, and health relate to continued engagement. We examined these relationships in an AI literacy activity ( N = 30) in which students interacted with two avatars of GPT (one RAG, one non-RAG) using the Educational VR platform Classlet. Using surveys, system logs, and coded dialogues, we observed that participants reporting higher technological expectations were less likely to proceed to the second avatar when interaction issues occurred. In regression models, presence was more closely associated with perceived functionality and self-reported health than with visual realism in this sample ( R 2 = 0.538 for presence; R 2 = 0.782 for intent). Dialogue coding showed a modest difference in contextual relevance. Taken together, the results suggest that expectation management and interaction fluency may be important for continuation decisions, and that usability and comfort may support presence alongside immersion. We discuss implications for VR adoption (expectancy violations), presence as tied to operational factors, and the role of adaptive dialogue, noting that these patterns warrant confirmation in larger, diverse samples.