AI-generated text in academic writing: Balancing structural proficiency and intellectual autonomy

Abstract

As generative AI tools like ChatGPT-4 become more common in higher education, their impact on student writing demands critical attention. This study investigates how AI-assisted composition influences undergraduate academic writing, focusing on structure, evidence use, and critical thinking. Drawing on Vygotsky’s Zone of Proximal Development and Sweller’s Cognitive Load Theory, a randomized controlled trial was conducted with 50 first-year students. Participants were divided into two groups: one using ChatGPT-4 during essay writing and another working independently. Essays were evaluated using Toulmin’s model of argumentation. Results showed that AI-assisted essays scored higher in organization (M = 4.3 vs. 3.2) and thesis clarity (M = 4.5 vs. 3.0), but lower in critical engagement (M = 2.9 vs. 3.6). Qualitative findings revealed that while AI users appreciated its efficiency, 74% expressed concerns about overreliance. Non-AI writers reported stronger conceptual understanding developed through personal effort. Although 75% of AI-assisted essays included peer-reviewed sources, only 37% of users verified them. These results suggest that AI enhances surface-level fluency but may weaken independent reasoning. The study calls for pedagogical strategies that position AI as a writing aid—not a replacement—while fostering critical thinking and academic integrity.