Virtual streamers, real purchases: the roles of anthropomorphism, interactivity, and presence in e-commerce live streaming

Purpose This study aims to investigate how anthropomorphic design (appearance and behavior) and interactivity (viewer–streamer and viewer–viewer) among artificial intelligence (AI)-driven virtual streamers influence consumer purchase intention in tourism and hospitality e-commerce live streaming (THCLS), which is mediated by social presence, telepresence and continuous watching intention, while accounting for demographic and behavioral heterogeneities. Design/methodology/approach An integrated model was developed and tested using survey data from 747 consumers experienced in watching virtual streamers in the THCLS context. A hybrid approach was used, combining partial least squares structural equation modeling (PLS-SEM) for linear relationships, multigroup analysis (MGA) for subgroup differences and artificial neural networks (ANNs) for nonlinear effects. Findings Anthropomorphic cues and interactivity significantly enhance social presence and telepresence, which in turn positively influence continuous watching intention and purchase intention. MGA reveals meaningful moderating effects of gender, age and viewing frequency, indicating heterogeneous responses to anthropomorphism and interactivity. ANN analysis further reveals nonlinear relationships, identifying social presence as the most influential predictor of purchase intention beyond linear rankings. Practical implications Practitioners should prioritize anthropomorphic appearance and behavior alongside real-time interaction to strengthen social presence, the primary driver of purchase intention. The development of community-oriented features that foster viewer–viewer engagement and the tailoring of strategies based on age and viewing frequency can enhance effectiveness. Continuous updates to content and interaction design are essential to mitigate novelty decay and sustain conversion performance. Originality/value This research extends presence-based theories and the stimulus–organism–response framework to AI-powered THCLS, uncovering dynamic linear, heterogeneous and nonlinear mechanisms in consumer decision-making, thereby addressing gaps in the virtual streamer literature and offering a novel hybrid analytical lens for THCLS.