This explanatory sequential mixed-methods study gave qualitative evidence interpretive priority while using the quantitative strand as supplementary and contextualizing evidence. It examined how family, school, pedagogical, and learner-use conditions were associated with perceived AI-supported learning capability among students in six rural lower-secondary schools in China. The quantitative strand drew on 486 valid student questionnaires, supplemented by contextual questionnaires from 24 teachers, and described sample patterns and selected associations. The qualitative strand comprised 32 semi-structured interviews with students, teachers, parents, and school leaders and provided the primary basis for interpreting how AI-supported learning was experienced and organized. Integrating Bourdieu’s theory of capital with Sen’s capability approach, the study treats mediated conversion as a conceptual account of how AI resources are socially and pedagogically organized, not as a statistically established indirect effect. AI-use frequency and teacher mediation varied across family-resource and home-connectivity conditions. In the supplementary school-fixed-effect regression model, teacher mediation, planned academic AI use, and output-evaluation confidence showed the strongest focal associations with perceived learning capability. Interviews indicated that attempt-feedback-explanation-revision routines supported explanation, verification, revision, and persistence, whereas weakly guided use was often oriented toward rapid completion. The findings distinguish nominal access from perceived capability and identify possible compensatory patterns alongside risks of reproducing existing inequalities. The cross-sectional, self-report design does not establish causal conversion, statistical mediation, school-level inference, or objective achievement gains.