Oak Story: Improving Learner Outcomes with LLM-Mediated Interactive Narratives

Abstract

Narrative-based education engages children in learning, but traditional approaches offer limited adaptability to individual preferences. Although large language models (LLMs) offer promising opportunities for interactive narratives, balancing their unpredictability with structured learning objectives remains challenging. To answer this challenge, we designed and built Oak Story, an educational mobile application for 4th-6th graders centered on local oak woodland ecosystems. Oak Story employs a learning-goal-directed LLM architecture that adapts the narrative, as well as multimodal real-world activities, to each individual student while ensuring that learning goals are met. In a between-participants study (N = 47), we find that Oak Story produces statistically significant increases in learning gains, engagement, and perceived agency compared to a control with static sequencing within and between scenes. These findings demonstrate an effective architectural approach for LLM-based educational systems that successfully balances learner agency with pedagogical structure.