Evaluating Artificial Intelligence, Peer, and Instructor Feedback Across Scientific Communication Assignments in...

Generative artificial intelligence (AI) has emerged as a potential source of formative feedback for student writing. However, relatively little is known about how AI-generated feedback compares with instructor and peer feedback across different Writing-to-Learn (WTL) assignments implemented in authentic undergraduate biology classrooms. This exploratory study compared AI (specifically ChatGPT-5), peer, and instructor feedback across infographic and academic explanatory article assignments implemented in three undergraduate biology courses representing diverse student populations. Agreement among feedback sources was evaluated using intraclass correlation coefficients (ICC), while student perceptions of feedback usefulness, trust, engagement, and effort were assessed using Likert-scale survey data. Agreement between AI and instructor feedback varied according to assignment modality, disciplinary context, and revision stage. AI demonstrated moderate agreement with instructor feedback during the draft stage, but agreement frequently diverged following student revision, particularly for higher-order rubric criteria such as content accuracy and disciplinary reasoning. Agreement was consistently stronger for academic explanatory articles than for infographics. Students perceived instructor feedback as more useful and trustworthy than either peer or AI feedback, whereas the effort required to apply feedback did not differ across feedback sources. AI feedback was generally perceived similarly to peer feedback. Collectively, these findings suggest that AI is best used as a formative, early-stage scaffold within WTL assignments rather than as a summative evaluation tool. When integrated into structured instructional workflows alongside instructor and peer feedback, large language models can support student revision while preserving the central role of instructor expertise in undergraduate biology education.