Artificial Intelligence in Education: A Mixed-Methods Case Study of AI-Mediated Adaptive Teaching Using Large Language Models

Abstract

The integration of large language models (LLMs) into secondary education presents both significant opportunities and pedagogical challenges. This mixed-methods case study examines the effectiveness of ChatGPT's adaptive learning modes - AI-Tutor, AI-Student, and AI-Simulator - in augmenting instruction for 11 advanced sixth-form students across STEM and humanities disciplines. Over a 6-week pilot intervention, we collected quantitative data through weekly Likert-scale ratings (n=242 observations), pre/post self-efficacy surveys, and learning analytics, supplemented by qualitative focus groups and open-ended reflections. Quantitative findings revealed that the AI-Simulator mode achieved the highest mean effectiveness rating (M = 4.45, SD = 0.52), particularly for conceptual understanding and self-regulated learning. The AI-Tutor mode showed strong engagement benefits (M = 4.27, SD = 0.65) but raised concerns about passive knowledge reception. Pre/post paired analysis demonstrated significant improvements in self-efficacy for problem-solving (t(10) = 3.84, p = .003, Cohen's d = 1.16) and intellectual risk-taking (t(10) = 2.91, p = .016, d = 0.88). Thematic analysis identified four key emergent themes: (1) Agency and ownership (40% of comments), (2) Scaffolding depth and adaptivity (35%), (3) Affective dimensions - reduced anxiety and increased confidence (22%), and (4) Limitations of artificial feedback (18%). Our findings suggest that AI modes emphasizing learner agency and peer-like dialogue produce superior cognitive and affective outcomes aligned with Bloom's 2-sigma hypothesis. However, structural scaffolding and pedagogical intentionality remain essential; LLMs are not pedagogical panaceas but tools requiring expert teacher design. We propose a framework for "pedagogically purposeful AI integration" emphasizing active learning, metacognitive monitoring, and human-AI collaboration rather than direct instruction substitution.