Writing-to-learn (WTL) pedagogy involves implementing writing assignments in STEM courses to engage students in conceptual learning and disciplinary thinking. Studies of WTL in organic chemistry indicate its value for supporting meaningful learning and eliciting students’ reasoning about reaction mechanisms. Our prior research investigated supervised machine learning (ML) to analyze students’ WTL responses, which involved training models using human annotations to identify features of mechanistic reasoning. These models automatically capture the features of mechanistic reasoning present within students’ responses at the sentence level. The present study explores the output of this automated analysis by using latent class analysis, an unsupervised ML method, which can identify patterns to provide insights about student responses that may be difficult to identify through human analysis alone. Grounded in cognitive theories of writing which view writing as an external representation of student reasoning, the unsupervised ML analysis provided additional insight regarding students’ mechanistic reasoning in the context of a second-semester organic chemistry laboratory course. Findings indicate the utility of ML for holistically analyzing student writing, with potential applications for research and classroom practice. For example, the automated analysis could offer instructors an overview of classroom responses or provide students specific feedback to support their mechanistic reasoning.