Working Papers

Influencer marketing has become an essential component for brand strategies, and the adoption of AI-generated virtual influencers is accelerating with advancements in generative AI. However, these emerging practices raise challenges related to transparency and ethical concerns. By conducting a large field experiment involving over 1.8 million consumers, we examine the interplay between virtual influencers’ anthropomorphism levels and the disclosure of their AI identity. The results show that virtual influencers with higher anthropomorphism levels enhance engagement metrics, while AI identity disclosure reduces link clicks and three-second video plays, particularly for highly realistic virtual influencers. Our underlying mechanism analysis based on an online experiment reveals that this reduction is driven by “expectation violation effect”, where disclosure violates consumers’ expectations and evokes negative feelings. Importantly, the negative effects reverse for consumers with prior experience interacting with virtual influencers—among this group, highly realistic virtual influencers with disclosure actually lead to greater engagement than their less anthropomorphic counterparts. Our findings provide theoretical and practical insights into the design and development of AI agents, emphasizing the need to strategically manage anthropomorphism and transparency to optimize consumer engagement.

Influencer marketing has become a pivotal e-commerce strategy, revolutionizing how brands interact with consumers. This strategy’s reliance on high-profile streamers, or "superstars," presents opportunities and challenges. While these influential figures drive significant platform value, their unexpected exits pose strategic challenges for ecosystem stakeholders. This study investigates a critical yet underexplored question: how does a superstar streamer's sudden departure reshape market dynamics within LSE platforms? Using a natural experiment, the unexpected exit of Viya, China's leading live streamer from Taobao Live, we conduct an empirical case study to examine the effects on 3,102 peer streamers across 165,004 streaming sessions. Our findings reveal that peer streamers in similar content categories experience significant benefits following a superstar's exit. Notably, this improvement stems from demand redistribution rather than changes in content supply, as peer streamers maintain consistent productivity and promotional strategies post-exit. Our analysis further identifies two key moderating factors: channel similarity and streamer type. Peer streamers more similar to the departed superstar experience amplified sales gains, particularly when similarity exists in hedonic attributes like content category and popularity. Interestingly, brand-employed streamers, who represent specific brands, gained greater sales and number of views than influencer streamers. This study extends the literature on LSE, the superstar effect, and brand halo effects while offering practical implications for platform managers, brands, streamers, and consumers.

As digital platforms integrate prosocial features into their monetization strategies, creators face a critical trade-off between sustaining income and engaging in altruistic initiatives. We study this tension in the context of Twitch’s built-in charity tool, using a quasi-experimental design based on the 2023 “Together for Good” campaign. Drawing on multi-level panel data comprising 10,704 streamer-day and 672,452 viewer–streamer–day observations, we implement difference-in-differences estimation with propensity score matching. We find that charity streams trigger short-term substitution: direct tipping declines on charity streams. However, this is followed by a long-term generosity spillover, as tipping increases in subsequent non-charity streams. Viewer-level analyses uncover a dual mechanism: active tippers reallocate spending away from tipping during charity events, while new or previously inactive viewers are more likely to initiate tipping to a streamer after exposure to the streamer’s charity events. Our study advances understanding of how platform-integrated prosocial tools reshape financial support patterns in the creator economy, offering design implications for balancing monetization and social impact.

The rapid advancement of generative AI has fueled the rise of hyper-realistic AI agents—virtual entities capable of delivering highly personalized, human-like interactions at scale. While these agents present exciting opportunities for enhancing consumer engagement, they also raise complex challenges related to trust, transparency, and data privacy. To investigate these tensions, we propose a large-scale field experiment involving 10,000 consumers of a startup. We manipulate two key design factors in the AI agents’ promotional videos: the level of service personalization and whether the agent’s artificial identity is disclosed. Drawing on dual-process, expectation-violation, and privacy calculus theories, we hypothesize that while personalization increases engagement, identity disclosure may undermine it. By exploring the interaction between these factors, we aim to uncover the mechanisms underlying consumer cognitive perception. Our findings contribute to the literature on AI agents and the personalization-privacy paradox, providing practical guidance for brands seeking to deploy hyper-realistic AI agents responsibly and effectively.

Live Streaming E-commerce (LSE) is a new format that embeds live streaming into e-commerce, where streamers sell products and interact with viewers in synchrony. Many stores have launched LSE channels to attract traffic and increase sales. As competition for effective human streamers is intense, platform owners have developed artificial intelligence (AI) streamers as an alternative. However, it is unclear whether human or AI streamers are more effective at engaging viewers and selling products. Drawing on media synchronicity theory, we develop a research model that examines the differences between human and AI streamers. We report tests of our model using observational data from Taobao Live. We conducted an online experiment to evaluate the underlying mechanisms that explain the performance differences between human and AI streamers. Our study provides insights for platforms and store owners seeking to better utilize AI technology to sell products, and for designers interested in developing more effective AI streamers.