Human–AI Collaboration in Academic Writing Instruction: Calibrated Feedback Model for EAP

Abstract

This study adopts a mixed-methods design to compare the feedback generated by a human instructor and ChatGPT-3.5 on student essays. It analyzes scoring consistency, students’ perceptions of feedback usefulness, and the extent to which each form of feedback aligns with instructional objectives. The dataset comprises 30 first-year students enrolled in a Canadian English for Academic Purposes (EAP) course. Each student submitted two argumentative essays, which were evaluated independently by both the instructor and ChatGPT-3.5 across three calibration stages. The findings suggest that while ChatGPT-3.5 consistently identified surface-level issues and demonstrated close alignment with human scoring, it was less effective in addressing rhetorical structure and contextual nuance. Student responses indicated a clear preference for a hybrid feedback model, which combines the immediacy of AI-generated comments with the depth and pedagogical value of instructor feedback. Through calibrated prompts and pedagogical (re-)framing, GenAI can supplement human feedback effectively but will not replace it. This study puts forward a hybrid model of feedback provision in which language instructors can leverage GenAI to support formative assessment while maintaining pedagogical oversight. The findings offer guidance for integrating GenAI into academic writing instruction in ways that preserve instructional quality and student agency.