Thinking, Feeling, and Acting with Generative AI: Affective, Behavioral, and Cognitive Engagement and Perceived...

Generative AI requires students not only to adopt technology but also to engage critically, behaviorally, cognitively, and emotionally with its outputs. This study develops and tests the AI Engagement ABC+ Model, a domain-specific adaptation of the affective-behavioral-cognitive engagement framework. The model differentiates positive and negative AI affect, three behavioral functions, creation, verification, and community participation, and two cognitive regulatory functions, strategic and critical cognition. A cross-sectional online survey collected data from 1250 students representing 15 nationalities at eight universities in China. Partial least squares structural equation modeling was used to examine 15 hypothesized associations between seven first-order engagement facets and four self-reported AI-supported outcomes: perceived Output Quality, Creativity, Integrity, and Trust Calibration. Fourteen hypotheses were supported. Verification Behavior had the largest estimated coefficient among the included predictors of Integrity (β = 0.303) and perceived Output Quality (β = 0.269) and was also associated with perceived Trust Calibration (β = 0.191). Critical Cognition had the largest estimated coefficient among the included predictors of perceived Trust Calibration (β = 0.270), and Community Behavior was positively associated with Integrity (β = 0.290) and Creativity (β = 0.240). Positive AI Affect was associated with perceived Output Quality (β = 0.214) and Creativity (β = 0.139), whereas Negative AI Affect was negatively associated with Trust Calibration (β = -0.137) but was unrelated to Integrity. The model explained 25.7% of the variance in Creativity, 25.2% in Integrity, 20.8% in perceived Output Quality, and 17.6% in Trust Calibration. Positive Q²_predict values provided additional evidence of out-of-sample predictive relevance. The findings support a differentiated view of GenAI engagement in which specific affective, behavioral, and cognitive facets are associated with distinct perceived outcomes. The model offers a theoretically structured basis for further research on responsible GenAI engagement, while its cross-sectional self-report design precludes causal or objective-performance conclusions.