Structured generative AI preparation and classroom belonging in a six-class randomized trial

Abstract Generative artificial intelligence (GenAI) may support independent preparation without necessarily strengthening students’ connections to the classroom. We examined a structured instructional protocol combining AI exploration, confidence reflection, unresolved question generation, and instructor responsiveness. Following baseline assessment, six intact undergraduate class sections at one Chinese university were randomized equally, with three sections and 110 students assigned to each condition. Assessments were conducted in Weeks 1, 6, and 12, and the intervention began in Week 2. Recomputed item means were analyzed using equally weighted section-level contrasts with four degrees of freedom, supplemented by exact randomization tests. At Week 12, the between-condition difference in classroom belonging change was 0.361 (95% CI 0.145 to 0.576; Bonferroni-adjusted p = 0.029). The corresponding differences were 0.282 for emotional engagement, − 0.497 for reactance to teacher guidance, and − 15.24 points for the confidence–performance discrepancy index (CPDI). Exact two-sided randomization p values were 0.100 for all primary contrasts, the minimum attainable with six balanced sections. Exploratory individual-level pathway and effect-modification analyses were treated as hypothesis-generating because only six sections were observed and the pathway estimates were sensitive to measurement timing. Overall, the structured protocol showed promising differences in classroom belonging and related outcomes, but inference remains limited by the small number of randomized sections, parametric model assumptions, and the preliminary discrepancy measure. The findings therefore concern the combined instructional protocol rather than an isolated effect of GenAI or an established psychological mechanism.