While the merits of dialogic talk for supporting students' learning is well documented, the contribution of specific talk moves on learning outcomes remains less understood. This study was conducted to investigate how student-generated talk moves during whole-classroom discussions, as captured by the Low-Inference Discourse Observation tool, predict subsequent written persuasive essays on the same topic. In this study, 471 written persuasive essays from students in grades 4 through 7, as well as classroom-level discussion features from 42 classrooms are analyzed using a Cumulative Link Mixed Model to account for the ordinal nature of the essay scores and the hierarchical data structure. Findings highlight the infrequency of peer-to-peer interactions, a highly dialogic talk move that was absent in most classroom discussions analyzed. Additionally, controlling for teacher talk moves, topic variations, and student demographics, the analysis revealed that student-generated questions positively predicted persuasive essay scores, whereas minimal responses had a negative effect. These findings underscore the significance of student contributions to dialogic interactions in classroom discussions, providing empirical evidence for their role in supporting argumentative writing skills.