Leveraging generative AI to facilitate peer feedback in collaborative argumentation learning

Synthesis: Chang et al. (2026) compare three peer-feedback conditions in collaborative argumentation learning — peer feedback (PF), peer feedback with GenAI (GenAI-PF), and peer feedback with GenAI supported by prompt scaffolding (PS-GenAI-PF) — with 45 student teachers in 12 groups. The GenAI-supported groups achieved better argumentation performance, and the prompt-scaffolded group performed best on advanced elements like "rebuttal data and warrant" and "addressing the opposing view." GenAI-supported feedback contained more explanations, suggestions, and neutral/negative feedback, and epistemic network analysis showed the scaffolded group paired negative emotions with higher-order feedback content (identification-suggestion, explanation-suggestion). The study offers concrete implications for designing GenAI-supported peer feedback.

Core Finding

Prompt-scaffolded GenAI support improves both argumentation performance and the higher-order quality of peer feedback in collaborative argumentation learning. The GenAI-supported peer-feedback groups (GenAI-PF and PS-GenAI-PF) outperformed the plain peer-feedback group on argumentation performance (Kruskal-Wallis H test), with the prompt-scaffolded group excelling on advanced argumentation elements — "rebuttal data and warrant" and "addressing the opposing view." The GenAI-supported groups also produced higher-quality feedback (more explanations and suggestions) and more neutral/negative (as opposed to purely positive) feedback. Critically, prompt scaffolding appears to convert students' interaction with GenAI from passive reliance into a more critical, higher-order evaluative process.

The Role of Prompt Scaffolding

The study's key design innovation is prompt scaffolding — structured prompts (context, objective, role) that guide students' dialogue with GenAI during peer feedback. Because students often experience over-reliance on GenAI-generated content and struggle to phrase questions, scaffolding helps them engage in cognitive and metacognitive processes. The PS-GenAI-PF group showed strong associations between negative emotions and higher-order feedback content (identification-suggestion and explanation-suggestion), suggesting scaffolded students critically evaluated rather than merely accepted AI output. This directly engages the knowledge base's Scaffolding and Prompt Engineering concepts and shows how structured prompts can transform GenAI from answer-provider into learning facilitator.

Methodology

• Participants: 45 student teachers (16 male, 29 female), third-year undergraduates in Educational Technology at a central China university.
• 12 groups of 3–4 students, randomly assigned to three conditions (PF, GenAI-PF, PS-GenAI-PF), four groups each; four rounds of collaborative argumentation across four topics in a 16-week course.
• Tools: Tencent QQ (organization), Tencent Meeting (discussion/feedback), Boardmix (collaborative editing), and ERNIE Bot (GenAI conversational partner).
• GenAI served as a cognitive/evaluative scaffold (evaluating structure/coherence, evidence sufficiency, counterarguments, and generating suggestions). GenAI-PF groups generated their own prompts; PS-GenAI-PF received prompt scaffolding.
• Analysis: Kruskal-Wallis H tests for argumentation performance and feedback-quality differences; epistemic network analysis (ENA) for associations between feedback content and emotion.
• Feedback quality assessed on elements (descriptive comments, identification, explanations, suggestions) and affective tone (positive, neutral, negative).

Relevance to the Knowledge Base

This article contributes empirical evidence to the knowledge base's Feedback, Collaborative Learning, and Scaffolding clusters. It connects Prompt Engineering to peer feedback and argumentation, and links GenAI support to Self-Regulated Learning, Metacognition, and Critical Thinking. It also provides a useful model of Human AI Collaboration where GenAI acts as an evaluative scaffold within a student-centered, collaborative Pedagogies and Teaching Strategies, relevant to Learning Design and Assessment in Higher Education.

What this means for practice

• Instructors. Supply structured prompt scaffolding (context, objective, role) rather than open GenAI access for peer feedback; only the scaffolded group excelled on rebuttal data and warrant and on addressing the opposing view.
• Instructional designers. Require students to state which GenAI suggestions they adopt, modify, or reject and why, converting reliance into the critical judgment the scaffold is meant to build.
• Prepare students for the neutral and negative feedback GenAI surfaces rather than avoiding it; in the scaffolded group those signals were productively paired with higher-order feedback content.
• Teacher educators. Deploy GenAI as an evaluative scaffold in argumentation, where students otherwise struggle to produce rebuttals or treat criticism as a personal attack.

Limitations

• 45 third-year educational-technology undergraduates at a single university in central China; the authors state generalizability is constrained by sample composition.
• The group-level experimental design assigned only four groups to each condition, which the authors note may constrain statistical power.
• Cognitive load was not measured or controlled, which the authors flag as a confound influencing motivation, engagement, feedback quality, and argumentation performance.
• Prompt scaffolding was static and uniform rather than adaptive to students' differing abilities and cognitive levels.

Citation

Chang, Y., Liu, Q., Lu, Y., & Miao, E. (2026). Leveraging generative AI to facilitate peer feedback in collaborative argumentation learning. International Journal of Educational Technology in Higher Education.