Can Large Language Models Make Reading a Book More Engaging?

Abstract

American literacy is struggling, with only 31% of eighth graders reading at or above grade level. This problem is largely one of engagement: students struggle with reading primarily because they don't read enough to develop proficiency. Poor reading skills lead to avoidance of reading, creating a negative cycle which further diminishes literacy development. This thesis contributes two novel LLM-based interventions designed to increase student engagement with assigned texts: LLM-Clarifications, which provide just-in-time support when students encounter obstacles while reading, facilitated by a sentence-by-sentence reading mechanism that tracks students' progress and discourages skimming, and LLM-Debates, which allow students to argue with chatbots about characters or themes after reading. Testing with 63 high school students showed that students equipped with these interventions spent 70% more time engaged with their assigned book compared to the control group and thoroughly read 43% more chapters. However, comprehension quiz scores increased only by 2.6%. I found that this discrepancy occurs because LLM interventions excelled at sustaining engagement for students who had already begun reading, but not at initiating engagement for students predisposed to skip reading entirely. In both groups, students skipped approximately half of the assigned chapters, with only one student out of 63 reading all chapters without skimming. These findings demonstrate that the novel LLM interventions successfully solve half of the reading engagement problem: sustaining and deepening engagement once students begin reading. The 70% increase in engagement time and 43% increase in thoroughly-read chapters represent a promising approach to addressing the literacy crisis by keeping students engaged with texts. Furthermore, contrary to some educators' concerns that AI primarily enables shortcuts in education, these results suggest that thoughtfully designed LLM interventions can actually deepen student engagement with learning materials rather than diminish it—providing a foundation for reimagining LLMs as tools that maximize, rather than minimize, students' learning and growth.