Abstract
<title>Abstract</title> <p>Artificial intelligence (AI) is increasingly embedded in the everyday academic lives of undergraduate students, yet the evidence on how AI-based tools shape learning and social engagement remains scattered across disciplinary and geographic boundaries. This paper presents a PRISMA 2020-guided systematic review, using narrative thematic synthesis, of empirical and review studies examining the application of AI technologies, including intelligent tutoring systems, adaptive learning platforms, and generative AI (GenAI) chatbots, in promoting undergraduate learning outcomes and social engagement. A structured search of academic databases and publisher platforms, supplemented by hand-searching and a subsequent broadened search of adjacent literatures, identified 235 records, of which 189 were screened after duplicate removal, 58 were assessed in full text, and 29 met all eligibility criteria and were synthesised. Because the heterogeneity of designs, populations, and outcome measures precluded meta-analytic pooling, data were synthesised narratively and thematically following Whittemore and Knafl (2005). Findings indicate that AI applications, particularly intelligent tutoring systems and GenAI chatbots, are associated with measurable but inconsistent gains in cognitive, behavioural, and emotional engagement, and that chatbot-mediated social presence can partially offset the isolation associated with online and blended learning, while in some cases coexisting with disengagement and academic-integrity risk. These benefits are moderated by pedagogical design, learner self-regulation, and institutional context, and are accompanied by well-documented risks to critical thinking and equitable access, especially within resource-constrained settings such as those in sub-Saharan Africa and Latin America. Building on this synthesis, the review proposes a strengthened integrative framework, the AI-Mediated Engagement Contingency Model, which reframes AI's effects on engagement as jointly contingent on task-level scaffolding, learner self-regulatory capacity, and institutional readiness, rather than as a fixed property of the technology itself. Practical implications for students, educational psychologists, school counsellors, and policymakers, together with directions for future research, are discussed.</p>