How Generative AI Is Reshaping Student Writing: A Data-Driven Perspective for Writing Instructors

Abstract

Generative AI has rapidly entered college writing classrooms, raising practical questions about how student texts are changing and what that means for instruction. This study analyzes 255 final-draft analytical essays written in first-year writing classes across three instructional contexts—pre-Gen-AI (Winter/Spring 2022), AI-prohibited, and AI-permitted with specified uses (Winter/Spring 2024). We combined holistic quality ratings of essays with Coh-Metrix indices of writing volume, lexicality, referential cohesion, and syntax. Analytically, we estimated a regression of essay quality on class type and demographics, and MANCOVAs (with essay score and demographics as covariates) for the four linguistic constructs. Essay quality did not differ by AI policy. However, compared to 2022, essays of AI-permitted classes were organized into fewer but shorter paragraphs; displayed greater lexical diversity and used less frequent, less familiar vocabulary; showed lower local and global anaphor overlap (other cohesion indices were stable); and exhibited lower verb-phrase, passive, and negation densities but higher gerund density. We interpret these as selective redistributions of linguistic resources rather than uniform gains or losses. For instructors, the actionable implication is two-fold: leverage AI-era gains in lexical precision while explicitly teaching referential continuity and clause-level strategies that sustain argumentative coherence.