Comparative analysis of peer group and AI-generated feedback in peer assessment: Insights into feedback quality and student perceptions in higher education

Abstract

The rapid integration of generative artificial intelligence (AI) into peer assessment is reshaping feedback practices in higher education. However, empirical evidence directly comparing human and AI-generated feedback remains limited. This mixed methods study compared feedback quality and student perceptions between 13 undergraduate peer groups comprising 39 students and ChatGPT-4o in a smart learning environment design course. Quantitative analyses indicated no significant difference during the first assessment stage; however, in the second stage, ChatGPT-4o provided feedback that was significantly higher in constructiveness, accuracy and concreteness. Thematic analysis further revealed that students perceived group feedback as contextually rich but sometimes subjective, whereas ChatGPT-4o feedback was viewed as more objective and well structured, though occasionally lacking contextual sensitivity. Overall, students demonstrated high levels of acceptance towards both feedback sources, but high-acceptance groups made more effective use of AI feedback during revision, resulting in higher accuracy scores. These findings highlight the complementary roles of human and AI feedback in strengthening the formative value of peer assessment and provide implications for the design of AI-augmented feedback systems in higher education.

Implications for practice or policy:

• Clear rubrics and explicit guidance for AI use are essential to improve the quality, consistency,and transparency of peer assessment.
• Combining peer group feedback with AI-generated feedback can enhance formative assessment by integrating contextualised human insights with structured and objective AI support.
• AI-generated feedback should be used as a supplementary tool rather than a standalone assessor, with ongoing human oversight to address potential inaccuracies and ensure pedagogical quality.