Abstract
INTRODUCTION: Virtual agents with realism have been widely promoted as a potential solution to the pervasive global problem of loneliness by providing companionship. However, prior research and practical applications have produced inconsistent and sometimes contradictory findings. To elucidate the mechanisms underlying this discrepancy, this study further explores how virtual assistants with behavioral realism influence the usage willingness of lonely consumers and the underlying mechanisms of this influence. METHODS: Using a scenario-based experimental approach, this research conducted three single-factor between-subjects experiments manipulating response type (scripted vs. natural). Data were analyzed using SPSS 25.0 and PROCESS 4.1. Study 1 examined the main effect of response type on willingness to use. Study 2 tested the mediating roles of cognitive trust and affective trust. Study 3 investigated the moderating role of loneliness type in the relationship between response type and willingness to use, as well as the conditional mediation pathways through which this moderation operates. RESULTS: The results reveal distinct preference patterns across different types of loneliness. Consumers with high state loneliness show greater willingness to use virtual agents with natural responses due to enhanced social connection and affective trust. In contrast, consumers with high trait loneliness tend to prefer virtual agents with scripted responses, as natural responses often elicit social avoidance and reduce affective trust. DISCUSSION: By jointly examining the types of loneliness and virtual agent characteristics, this research helps reconcile previously inconsistent findings in the human-AI interaction literature. The results offer important implications for the design and deployment of virtual agents, suggesting that tailoring behavioral realism to users' loneliness profiles can enhance willingness and effectiveness among different types of lonely consumers.