GUIDE-PATH: An AI-Driven Framework for Personalized English Education via Explicit Pedagogical Strategy Guidance

Abstract

The opacity of large language model (LLM)-based English tutors constrains learners’ awareness of pedagogical decisions, diminishing metacognitive awareness and trust. We present GUIDE-PATH (Guided Instruction via Decision-Explicit Pedagogical Actions for Tailored Help), an AI-powered framework that explicitly communicates teaching strategies during the tutoring process. In contrast to black-box LLM tutors, GUIDE-PATH separates pedagogical reasoning from content generation through three core components: a dynamic student profiling engine that tracks CEFR level, error-specific mastery, and learning style proxies; a formal strategy library comprising eight evidence-based pedagogical moves (such as explicit correction, recast, metalinguistic feedback, and scaffolding prompt) grounded in second-language acquisition theory; and a weighted reasoning module that selects the optimal strategy based on the learner's state. The selected strategy is communicated to the learner through non-intrusive pedagogical notes. A randomized controlled trial involving 172 intermediate EFL learners compared GUIDE-PATH with an LLM-only tutor and a no-AI control. Results indicated that GUIDE-PATH produced significantly greater learning gains in grammar (Cohen’s d = 0.82), vocabulary (d = 0.91), and speaking confidence (d = 0.76). Transparency ratings strongly favoured GUIDE-PATH (d = 2.31), and learners correctly identified the pedagogical strategy in 73% of cases, a skill correlated with learning gains. These results show that explicit support of pedagogical strategies improves both learning outcomes and learner trust. The magnitude of learning gains observed in this study is consistent with findings from adjacent domains of educational AI. Explainable machine learning models for student dropout prediction [121] and LSTM-SHAP-based academic performance prediction [120] have similarly reported that transparency-enhanced systems produce measurably better outcomes than opaque alternatives. The positive correlation between strategy recognition accuracy and learning gains (r = 0.44 for grammar; r = 0.39 for vocabulary) aligns with research on the role of attention in learning [48], which established that conscious awareness of instructional stimuli is a prerequisite for deep encoding. Engagement detection studies [57], [94], [103] corroborate the finding that learners who are more actively engaged with instructional content as evidenced by their ability to identify pedagogical strategies achieve greater learning gains. The strategy library and the reasoning module are released as open-source tools to support transparent educational AI.