AI Co‐Host Adoption and Traffic Governance in Live Commerce: Human Effort Incentives and Architecture Switching

ABSTRACT Artificial intelligence (AI) virtual streamers are making live‐commerce selling increasingly scalable, yet human explanation and trust remain central to purchase conversion. This paper develops a Stackelberg model of AI co‐host adoption under algorithmic traffic bias. The platform first chooses a transaction commission and an AI‐oriented traffic bias, after which the brand selects AI‐only selling, human‐only selling, or AI‐human co‐host selling. When a human streamer participates, the brand offers a performance commission, and the streamer chooses costly explanation effort. Algorithmic traffic allocation shapes both market exposure and the marginal return to human effort. Greater AI‐oriented traffic bias weakens effort under human‐only selling but strengthens it under AI‐human co‐hosting by enlarging the revenue base for human explanation. Once AI traffic support makes AI‐only selling more profitable than co‐hosting, human effort drops discontinuously to zero. The model identifies when AI co‐hosting complements human streamers, when traffic support instead induces substitution, and why a commission‐driven platform may favor more AI‐oriented exposure than a welfare‐aware benchmark. Extensions with return risk and two‐part contracts clarify how product fit and compensation design affect the value of co‐hosting. Robustness analyses show that the main qualitative results continue to hold under bounded nonlinear traffic allocation, nonlinear demand, and binding contract constraints.