The mediating role of self-efficacy in the relationship between prior AI exposure and adoption intentions of AI-driven objects-to-think-with among preservice mathematics teachers

Interest in artificial intelligence in education is growing, but limited empirical evidence explains how prior AI exposure relates to pedagogically meaningful adoption among preservice mathematics teachers. This study examines adoption intentions regarding AI-driven objects-to-think-with in South African teacher education. Guided by an integrated Social Cognitive Theory and Technology Acceptance Model, the study examined prior AI exposure, technical and pedagogical self-efficacy, perceived usefulness, and an adoption-intention composite, together with exploratory patterns by gender and year of study. A cross-sectional survey was conducted with 79 participants. Data were analysed using Partial Least Squares Structural Equation Modelling with bootstrapped indirect-effect estimation and gender- and year-stratified models. Pedagogical self-efficacy showed the largest and most consistent associations with perceived usefulness and the adoption-intention composite. Prior AI exposure showed small, statistically imprecise total effects under the binary exposure measure, and the sequential pathway through perceived usefulness added little explanatory value beyond the self-efficacy association. Gender-stratified estimates suggested a clearer positive exposure-to-self-efficacy association among women than among men, but the reported within-group estimates did not establish a between-group difference. Estimates were statistically imprecise across year-of-study groups. These findings identify pedagogical self-efficacy as the belief most closely associated with adoption-related responses in this sample and motivate prospective evaluation of mastery-oriented, pedagogically grounded AI experiences in teacher education.