GAI Pedagogical Agents and Teacher-Guided Prompt Use in EFL Debate Training

Abstract

This study investigated the pedagogical functions and learner uptake associated with two generative AI approaches for training EFL students’ debating skills. Conducted within the context of a Model United Nations (MUN) activity in two university EFL classes, 20 students from class A engaged with a Generative AI Pedagogical Agent (GAI-PA), while another 20 from class B used teacher-guided ChatGPT prompts to enhance their debating over a period of 8 weeks. The instructions embedded in the two approaches were analyzed using a constructivist grounded theory approach to determine the specific pedagogical functions emerging from each system. After 8 weeks of exposure, the learners’ uptake of these functions was examined through a deductive analysis of their responses recorded in their conversation sessions. The results show that the GAI-PA embedded a structured, scaffolded, and role-driven instructional design that produced consistent learner uptake across most functions. In contrast, the teacher-guided prompt system relied on greater learner initiative and yielded more variable uptake, particularly in cognitively demanding areas such as revision and discourse construction.