Addicted to AI? Prevalence and psychological drivers of generative AI dependency among pre-service teachers

Generative artificial intelligence (GenAI) has become embedded in higher education, raising concerns about students' behavioral dependency on these tools. Focusing on pre-service teachers-who will shape how a future generation learns to use AI-we asked how prevalent GenAI dependency is, what drives it, and whether perceived AI literacy protects against it.

In a mixed-methods design, 565 education-major students in Chinese teacher-preparation programs completed an online survey, and five were interviewed.

Dependency markers were already widespread: 79% called AI indispensable to their learning, 78% felt unaccustomed without it, 54% "asked AI first," and 31% used its answers without verification. In the structural model (R ² = 0.60), the strongest drivers were impulsivity (β = 0.40) and anthropomorphic attachment (β = 0.28), with a smaller contribution from AI anxiety (β = 0.15); trust was non-significant. Necessary condition analysis identified social norms as a necessary-in-degree condition. Perceived AI literacy was modestly protective (β = -0.19), stronger in the full model than its zero-order correlation (r = -0.14)-a suppression effect, because more AI-literate students also trusted AI more and sat within more AI-reliant peer groups. Interviews converged with the survey model, most strikingly on anthropomorphic attachment, while qualifying the smaller anxiety pathway.

Dependency is driven chiefly by dispositional impulsivity and affective attachment rather than by low competence; interventions-for policymakers, teacher educators, and pre-service teachers-should prioritize self-regulation and awareness of parasocial engagement, treating AI-literacy training as a secondary lever.