#377: Learning by Teaching an LLM: A New Approach to Writing Instruction

This preliminary study investigates how different instructional modalities – i.e., video lecture versus learning by teaching (LbT) with a large language model (LLM) – influence students’ understanding and application of rhetorical strategies in argumentative writing. Using a within-subjects design, 46 participants completed both a video condition and a LbT condition wherein they instructed an LLM-powered “student” on rhetorical strategies. Each session covered a distinct set of three strategies, with order and content counterbalanced. Learning was evaluated through pre-tests, immediate post-tests, delayed post-tests, and post-lesson essays. Participants used more rhetorical strategies in essays written after the LbT condition (2.70 vs. 2.43 on average) and made fewer errors on immediate (1 vs. 4) and delayed (7 vs. 15) post-tests. Participants who completed the video condition second showed a greater decline in strategy use (55% vs. 29%). These findings introduce a replicable experimental framework for studying LLMs as interactive tools for writing instruction.