Abstract
This study examines how English-as-a-foreign-language (EFL) teachers experience and regulate emotions when integrating generative artificial intelligence (AI) into high-stakes test development. Drawing on control–value theory, and using a qualitative case study of a university-wide examination team in China, we analysed interviews, interaction logs and participant observation to show how emotional experiences diverge systematically across organisational roles. The findings identify a structural pattern—termed the emotional funnelling phenomenon—where AI-related risks and uncertainty concentrate in leadership positions, producing sustained anxiety and, at times, hopelessness. In contrast, technically or creatively oriented roles, structurally insulated from exam-related accountability, experienced curiosity, enjoyment and a strong sense of control. A different emotional trajectory emerged for the practitioner responsible for reading items, who reported boredom and grievance stemming from the repetitive verification of unreliable AI outputs, compounded by the sense that her labour was undervalued by colleagues and under-recognised within the institution. Across roles, teachers employed AI-specific emotion regulation strategies, such as reframing, anthropomorphisation and competence-oriented role reversal. The study concludes that emotional outcomes in AI adoption are shaped not by the technology itself but by how institutions distribute control, risk and recognition within AI-mediated workflows.