2025/10/18 by Joshua Weidlich, Flurin Gotsch, Kai Schudel +7 · 1 voice · 20 citations
Computer Science · Social Sciences · #Academic integrity and plagiarism #Feedback loop #Higher education #Negative feedback #Online Learning and Analytics #Peer feedback #Perception #Positive feedback #Quality (philosophy) #Student Assessment and Feedback
paper · doi:10.1016/j.caeo.2025.100300
published in Computers and Education Open 9, 100300 (Elsevier BV)
openalex created_date 2025/10/10 · openalex publication_date 2025/10/18 · openalex updated_date 2026/08/04
With the emergence of Large Language Models (LLMs), AI-generated feedback is gaining traction as a scalable feedback source for higher education. To qualify as viable alternatives for higher education classrooms, its relative effectiveness compared to teacher feedback and peer feedback needs to be better understood. To this end, a randomized field experiment ( N = 90) compared the effects of three feedback sources—teacher, peer, and LLM—on students’ feedback perceptions and their achievement. To eliminate potential confounds, we (a) controlled for learning gains that may result from students giving feedback to their peers and (b) blinded feedback sources from feedback recipients. Results showed that students rated teacher feedback as less fair and harder to accept than peer and LLM feedback. Students receiving teacher feedback also indicated less willingness to revise their work based on the feedback. Conversely, teacher feedback produced the strongest improvements in scientific argumentation and formal quality of students’ work. Here, LLM feedback yielded the smallest improvement overall. Lastly, feedback literacy and intrinsic motivation partly moderated feedback effects on perceptions and achievement outcomes. For example, students with more productive attitudes toward feedback achieved higher argumentation quality after receiving teacher feedback than their peers. Findings indicate that the impact of feedback on both student perceptions and performance depends on the interplay between the feedback source and learner dispositions; deliberately aligning these factors could therefore amplify the benefits of feedback interventions. Future research should explore hybrid and adaptive feedback models that integrate human and AI input.