Abstract
INTRODUCTION: Generative AI is increasingly embedded in foreign-language learning, yet its effects may depend on how students use it. We examined how distinct AI usage behaviors relate to learning/career anxiety and learning performance among Spanish majors in China. METHODS: We surveyed 733 Spanish majors from 59 universities (July-August 2025). Instruments captured AI behaviors, learning anxiety, career anxiety, and learning performance. Exploratory and confirmatory factor analyses identified usage dimensions; Pearson correlations and multiple regressions (controlling for gender, age, grade, and program type) tested associations. Mediation/suppression was examined via Bootstrap procedures. RESULTS: A three-factor structure emerged: (1) substitutive use (AI replaces learners' cognitive work), (2) after-class auxiliary use, and (3) in-class auxiliary use. Substitutive use was positively associated with learning anxiety (β = 0.331) and career anxiety (β = 0.189) and negatively predicted learning performance (β = -0.178). After-class auxiliary use positively predicted performance (β = 0.271) but slightly increased anxiety; both learning and career anxiety exerted small suppressor effects on this positive pathway. In-class auxiliary use showed no significant effects on anxiety or performance. Both learning anxiety (β = -0.221) and career anxiety (β = -0.092) negatively predicted performance. DISCUSSION: The educational impact of AI hinges on usage patterns and scenarios. Substitutive behaviors elevate anxiety and undermine performance, whereas well-scaffolded after-class auxiliary use benefits performance despite minor anxiety-related suppression. Guiding students to curb substitutive use and optimize auxiliary strategies-especially outside class-may improve outcomes.