AI Tutor-Based Language Learning: Linking Service Quality to Learners’ Continuance Intention Through A Dual-Pathway Model

Aim/Purpose: This study employs the Stimulus-Organism-Response (SOR) framework to investigate the psychological drivers of learners’ continuance intention toward AI tutors. The study explores how five core service quality dimensions (aesthetics, control, personalization, responsiveness, and reliability) shape learners’ internal states, conceptualized as a dual system consisting of rational attitude and experiential engagement. The study further examines how these internal states foster satisfaction and subsequently influence continuance intention, while also analyzing the asymmetrical moderating role of perceived risk on these psychological pathways. Background: AI tutors are increasingly adopted in language education due to their capacity to deliver scalable, adaptive, and personalized instruction, an especially relevant development in settings such as Vietnam, where language proficiency gaps persist. Despite their rapid integration into everyday learning, the mechanisms that sustain learners’ long-term use remain insufficiently understood. Existing research largely relies on broad technology acceptance models that conceptualize AI systems as uniform delivery tools, overlooking their highly interactive and adaptive nature. Consequently, limited attention has been given to which specific service quality dimensions most strongly shape learners’ evaluations and sustained engagement. Moreover, prior studies often assume a direct relationship between system features and continuance intention, neglecting the cognitive and experiential states that emerge as learners interact with AI tutors. These gaps are particularly salient in Vietnam’s fast-growing digital learning environment, where high demand for effective language education converges with increasing public interest in AI-based solutions. Yet, empirical evidence on how learners assess the quality of AI tutor services remains scarce. This underscores the need for a more fine-grained examination of how distinct service characteristics influence learners’ internal evaluations and their intention to continue using AI tutors, providing both theoretical precision and practical guidance for enhancing AI-supported language learning. Methodology: This study utilized a quantitative, cross-sectional design via a structured online questionnaire. Data were collected through purposive sampling from 771 experienced users of AI language tutors (including high school students, university students, and working professionals) in Vietnam. The data was analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM) with SmartPLS 3.0 software to test the complex research model, including the proposed moderation effects. Contribution: This study advances technology acceptance theory by applying a nuanced SOR framework to the AI tutor domain, deconstructing service quality into five specific, actionable dimensions. Its primary theoretical contribution is the validation of a “dual-engine” retention model (distinguishing between a rational attitude path and an experiential engagement path) and the introduction of the resilience of flow phenomenon, demonstrating that the engagement-based pathway is uniquely resilient to the negative impacts of perceived risk. The study also provides crucial empirical evidence from the under-researched context of Vietnam’s emerging market. Findings: The results reveal that all five service quality dimensions (aesthetics, control, personalization, responsiveness, and reliability) significantly and positively influence both learner attitude and engagement. These two internal states, in turn, cultivate satisfaction and drive continuance intention. Crucially, perceived risk demonstrates an asymmetrical moderating effect: it significantly weakens the influence of the rational pathway (attitude → continuance intention) and the summative judgment (satisfaction → continuance intention), but it does not weaken the influence of the “hot” experiential pathway (engagement → continuance intention). Recommendations for Practitioners: Developers and educators should adopt a dual defensive-offensive strategy. Defensively, they must mitigate perceived risk by ensuring pedagogical accuracy (reliability) and data transparency. Offensively, they should prioritize investments in features that drive deep, resilient engagement, as this pathway is robust to risk. Enhancing aesthetics, learner control, personalization, and responsiveness are all critical levers for building both positive attitudes and strong engagement. Recommendation for Researchers: Researchers should validate this model across specific linguistic cohorts to test for variance. Studies should also test the model’s generalizability in other cultural or educational contexts beyond the specific emerging market of Vietnam. Impact on Society: By providing a strategic blueprint for designing more effective and engaging AI tutors, this study helps educational technology providers create tools that foster long-term learner commitment. This can help close persistent language proficiency gaps in emerging economies, support economic globalization, and enable transformative, scalable language acquisition for learners at all stages. Future Research: Future research should employ longitudinal designs to track the evolution of engagement and satisfaction beyond the initial novelty period. Subsequent studies must also connect these service quality factors and psychological constructs to objective, tangible learning outcomes, such as measured proficiency gains, to ascertain the AI tutor’s true pedagogical efficacy.