Abstract
Artificial Intelligence (AI) is increasingly deployed in English-as-a-foreign-language (EFL) education, offering adaptive feedback, automated evaluation, and personalized learning pathways. However, existing research overwhelmingly emphasizes AI adoption and performance benefits, while largely overlooking what happens when AI systems fail to meet learner expectations and how learners recover from such failures. As a result, the cognitive-affective processes through which expectation violations translate into disengagement-or are mitigated through recovery-remain under-theorized and empirically unexplored. Addressing this gap, this study proposes and tests a cognitive-affective recovery model of learner engagement in AI-supported EFL contexts. Drawing on Expectation Violation Theory (EVT), Cognitive Appraisal Theory (CAT), and Digital Divide/Resilience Theory, the model explains how expectation violations influence engagement and how cognitive reappraisal and trust recovery mediate this relationship, while digital grit conditions learners' ability to persist following setbacks. A two-wave survey of 298 Chinese EFL learners from urban and rural settings, including both university students and private institute learners, was analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM). Results show that expectation violations significantly reduce learner engagement, but perceived AI adaptivity narrows the adaptation gap and activates recovery processes. Cognitive reappraisal and trust recovery emerged as key mediating mechanisms, while digital grit moderated critical pathways by sustaining engagement under adverse conditions. By shifting the focus from AI success narratives to failure-and-recovery dynamics, this study advances theory on AI-learner interaction and offers practical guidance for designing resilient, trust-sensitive, and equity-oriented AI systems in language education.