RAG in Self-Regulated Personalised Learning in Digital Education

Large language models (LLMs) generate fluent text, but they hallucinate and work from knowledge that ages quickly. Retrieval-Augmented Generation (RAG) answers both challenges by grounding generation in external sources a reader can check. At issue is whether generative artificial intelligence (AI) can be used trustworthily in high-stakes domains such as education. Self-regulated learning (SRL) research has argued for decades that students who set goals, track their own progress, pick strategies deliberately and reflect on the results learn more deeply and retain more, and that these matter most in the loosely structured, learner-directed conditions typical of digital education platforms. This review examines where the two lines of work meet, and what evidence exists on how RAG systems are being designed, deployed and evaluated as support for self-regulated, personalised learning. We first trace RAG from classical information-retrieval models through neural dense retrieval to the current generation of large models. We then map its applications onto the phases of self-regulation: goal setting and planning, retrieval-grounded tutoring and conversational scaffolding, personalised content retrieval and adaptive sequencing, automated assessment and feedback, learning-analytics-based monitoring, curriculum-grounded content creation, and support for diverse learners. Thereafter we examine the challenges that qualify this promise, including retrieval-conditioned hallucination, data privacy and security risks specific to vector-based retrieval, algorithmic and retrieval fairness, the interpretability of retrieved evidence, infrastructural and implementation barriers, and disruptions to academic integrity. In this review conclusion, it is noted that RAG offers a feasible and increasingly evidenced pathway to a personalised, scalable and pedagogically grounded support for self-regulated learning, but that realising this potential responsibly requires deliberate attention to the quality of retrieval, along with rigour in terms of evaluation as well as learner agency rather than technical capability alone.