The AI dilemma in L2 writing: Is AI literacy associated with willingness to write through self-efficacy, anxiety, or both? A serial mediation investigation

This study investigates the psychological mechanisms through which artificial intelligence (AI) literacy relates to L2 learners’ willingness to write (WTW). It proposes a serial mediation model wherein L2 writing self-efficacy (WSE) and L2 writing anxiety (WA) sequentially mediate this relationship. Grounded in social cognitive theory and control-value theory, the research employed a sequential explanatory mixed-methods design with 303 Chinese EFL learners. Quantitative results from structural equation modeling indicate that: (1) L2 WSE fully mediates the relationship between AI literacy and L2 WTW; (2) AI literacy is indirectly linked to L2 WTW through a sequential pathway: by first positively predicting L2 WSE, which in turn negatively predicts L2 WA; and (3) the direct effects of AI literacy on L2 WA and L2 WTW are non-significant. To explain these quantitative pathways, qualitative interviews with 16 participants further elucidated how AI tool usage corresponds to higher L2 WSE through strategic partnership, reduced startup burden, and task scaffolding support. Furthermore, the qualitative narratives revealed that while AI usage can initially be associated with anxiety (e.g., fear of exposure, tool-related frustration), the heightened L2 WSE linked to AI literacy effectively serves as a protective factor against these affective barriers. The study concludes that AI literacy positively correlates with L2 WTW through a fully indirect mechanism, in which heightened L2 WSE serves as a vital prerequisite for lowering L2 WA. These findings contribute to theory by integrating two major psychological frameworks and offer practical implications for leveraging AI to cultivate more willing L2 writers.