Research on student learning in organic chemistry indicates that students tend to focus on surface level features of molecules with less consideration of implicit properties when engaging in mechanistic reasoning. Writing-to-learn (WTL) is one approach for supporting students' mechanistic reasoning. A variation of WTL incorporates peer review and revision to provide opportunities for students to interact with and learn from their peers, as well as revisit and reflect on their own knowledge and reasoning. However, research indicates that the rhetorical features included in WTL assignments may influence the language students use in their responses. This study utilizes machine learning to characterize the mechanistic features present in second-semester undergraduate organic chemistry students' responses to two versions of a WTL assignment with different rhetorical features. Furthermore, we examine the role of peer review on the mechanistic reasoning captured in students' revised drafts. Our analysis indicates that students include both surface level and implicit features of mechanistic reasoning in their drafts and in the feedback to their peers, with slight differences depending on the rhetorical features present in the assignment. However, students' revisions appeared to be primarily connected to the peer review process "via" the presence of surface features in the drafts students read (as opposed to the feedback received). These findings indicate that further scaffolding focused on how to utilize information gained from the peer review process (i.e., both feedback received and drafts read) and emphasizing implicit properties could help support the utility of WTL for developing students' mechanistic reasoning in organic chemistry.