This study investigates the linguistic and rhetorical challenges EFL learners encounter in academic writing and examines how engagement with AI tools influences their drafting and revision processes. Anchored in cognitive, noticing-based, and sociocultural perspectives, the study combines a corpus-based analysis of 152 argumentative essay drafts with qualitative insights from semi-structured interviews. Quantitative findings show consistent reductions in grammatical, lexical, and cohesion-related errors after AI-supported revision, with improvements becoming more evident over the semester. Learners’ initial drafts in the second writing cycle also contained fewer micro-level errors, suggesting partial internalization of patterns highlighted through repeated AI feedback. However, progress in higher-order features, particularly argument clarity and discourse organization, remained limited, indicating that current AI systems provide strong surface-level scaffolding but insufficient guidance for complex rhetorical development. Qualitative analyses further reveal that learners increasingly adopted selective and critical engagement with AI suggestions, balancing appreciation for enhanced accuracy and confidence with concerns about overreliance, authorship, and ethical boundaries. Together, these findings demonstrate that AI tools function as effective mediators of local linguistic accuracy but offer only partial support for deeper argumentative competence. The study underscores the need for pedagogical frameworks that integrate AI strategically with explicit instruction in rhetorical reasoning and responsible tool use to promote sustainable writing development.