Can AI replace the stars? The effectiveness of celebrity avatars in emotional connections

Purpose This research aims to investigate how consumers respond to celebrity avatars compared to human celebrities in relational communication contexts. It examines whether parasocial bonds formed with real celebrities transfer to their AI-powered avatar representations. Specifically, the research addresses three questions: how consumers respond to celebrity avatars relative to human celebrities, through what psychological process this difference emerges and under what conditions the avatar disadvantage can be mitigated through interaction design. Design/methodology/approach Six experimental studies test the proposed framework. Studies 1A-1C establish the baseline effect using Chinese Generation Y celebrities (Study 1A), Chinese Generation X celebrities (Study 1B) and globally recognized Western celebrities (Study 1C), with samples drawn from both Chinese and Western participants. Studies 2A and 2B test the sequential mediating roles of perceived intimacy and enduring involvement. By using an experimental-causal-chain design, Study 3 manipulates communication-conveyed intimacy through dialogue design to further test the proposed psychological mechanism. Findings Consumers respond less favorably to celebrity avatars than to human celebrities, exhibiting lower commitment, emotional value and celebrity attachment. This negative effect is sequentially mediated by reduced perceived intimacy and lower enduring involvement. However, when communication-conveyed intimacy is experimentally elevated through interaction design, the negative indirect effect through enduring involvement becomes weaker and can shift in a positive direction. Parasocial bonds therefore do not automatically transfer from celebrities to their avatars. Research limitations/implications This research extends the virtual entity effectiveness literature by examining celebrity avatars as a distinctive type of virtual entity tied to real human celebrities. It introduces agency ambiguity as a conceptual lens for understanding why preexisting parasocial bonds do not automatically transfer to avatar representations, and it integrates construal level theory with parasocial relationship theory through perceived intimacy. The studies relied primarily on static stimuli which may limit ecological validity. Future research could use more dynamic and realistic avatar stimuli, test downstream behavioral outcomes and explore additional moderators. Practical implications The findings offer guidance for celebrities, brands and platform designers on when and how to use celebrity avatars. Celebrity avatars should not serve as full replacements for human celebrities, particularly in emotionally important contexts where relational depth matters most. Instead, they are best perceived as selective complements that extend the celebrity’s communicative reach, supporting routine fan interactions and sustaining accessibility when the celebrity is unavailable. To maximize their effectiveness, brands and platform designers should prioritize the quality of communication, incorporating intimacy cues such as personalized responses, behind-the-scenes content and emotionally resonant dialogue. Originality/value This research advances the understanding of celebrity avatars in marketing communication. It contributes to the emerging literature on virtual entities, human-AI interaction and consumer-celebrity relationships. The study introduces agency ambiguity as a conceptual lens and identifies communication-conveyed intimacy as a key boundary condition, offering practical guidance for deploying celebrity avatars effectively.