Constructing Replaceability: Authenticity Regimes and the Automation of Emotional Labor

Abstract

Why does the same AI technology replace identical emotional labor in one context but not another? Existing explanations focus on the nature of the task or the capability of the technology. This paper shows instead that the automation of emotional labor is fundamentally a problem of valuation. I develop the concept of authenticity regimes to describe the evaluative infrastructure through which organizations define what counts as authentic emotional performance. Based on 15 months of ethnographic fieldwork and 141 interviews in China's live-commerce industry, I compare two business models that adopted the same AI avatars yet reached opposite conclusions. In influencer streaming, the authenticity regime evaluates streamers through engagement metrics. AI avatars failed these benchmarks and were relegated to backstage roles. In brand streaming, the authenticity regime evaluates streamers through transaction metrics that capture sales outcomes but not the emotional labor behind them. AI avatars met these benchmarks, and their adequate performance gave managers the basis to treat that invisible labor as dispensable. As a result, AI replaced one-third of human streamers. This paper argues that replaceability is neither an inherent property of the work nor a product of technological capability. It is constructed when AI enters an evaluative infrastructure that was never built to see the labor it displaces.