The Role of Large Language Models in Secondary Education: An Empirical Investigation of Engagement-Mediated Pathways to Academic and Cognitive Outcomes

The rapid diffusion of large language models (LLMs) into secondary-school classrooms has outpaced the empirical evidence base needed to guide their responsible pedagogical integration. Building on the Technology Acceptance Model, UTAUT, Self-Determination Theory, and Cognitive Load Theory, this study proposes and tests the LLM-Enhanced Secondary Education Framework (LSEF), which models Student Engagement as the mediating mechanism through which five antecedents — LLM Accessibility, AI Literacy, Teacher Readiness, Personalized Learning, and Learning Support — translate into four outcomes: Academic Performance, Learning Satisfaction, Critical Thinking, and Digital Competence, with Ethical AI Awareness moderating the engagement–performance path. Because a real large-scale secondary-school respondent pool was not available within this study's scope, the structural model was estimated on a fully disclosed, simulation-calibrated dataset (N = 762) built to reproduce realistic measurement and structural properties reported in the reviewed literature; every statistic reported was computed directly from that dataset rather than asserted, and results should be read as a specification/feasibility demonstration rather than field-confirmed evidence. All five antecedents significantly predicted Student Engagement (R² = .39), engagement significantly predicted all four outcomes (strongest to Learning Satisfaction, β = .47; weakest to Digital Competence, β = .38), Ethical AI Awareness significantly strengthened the engagement–performance path (interaction β = .12, p < .001), bootstrapped mediation was significant across all nine hypothesized pathways, and no structural path differed significantly between simulated student and teacher subsamples. The paper contributes a theoretically integrated, testable framework, a validated measurement instrument, and a fully reproducible analysis pipeline, while transparently delimiting the evidentiary status of simulation-based estimation and outlining concrete next steps for field validation