AI evidence extraction

AI assistance in peer feedback provision: Pedagogically sound, but minimally adopted

Stanislav Pozdniakov, Jonathan Brazil, Seyyed Kazem Banihashem, Omid Noroozi, Dragan Gašević, Shazia Sadiq · Computers & Education · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

7/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

148.20

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1016/j.compedu.2026.105591

Methodology & findings

Study design

Observational study analyzing 7,670 instances of AI-assisted peer feedback from 433 undergraduate students in a semester-long course.

Sample

N = 433, 3 groups

Main result

The study found that "AI-A was typically positive, well structured, and focused on pedagogical strengths (79% of all instances)" but "uptake remained relatively low, with just 9% (690) of AI assistance suggestions leading to revisions." Additionally, "AI-A that was less accurate, less complete in describing the original peer feedback, or less positively worded was more likely to prompt revision, with small-to-medium effect sizes."

Reports effect sizes.

Research paradigm

Empirical - mixed methods (quantitative analysis of platform data with qualitative implications)

Author conclusions

The authors conclude that these findings "offer valuable insights for designing AI-A in peer feedback platforms that promote learning, encourage reflection, and preserve student autonomy." They also note the "nuanced role of GenAI: AI-A, which contained only moderately specific and correct suggestions, prompted students to act on it."

Risk of bias

Selection bias: Only students who engaged with the AI assistance platform were included; non-users excluded; Observation bias: Students may have modified behavior knowing their feedback interactions were being analyzed; Confounding variables: Student motivation, prior feedback experience, and disciplinary knowledge not controlled; Selection bias: Single large undergraduate course may not be representative of broader student populations; Measurement bias: Uptake measured as revisions only; students may have engaged with AI-A in other ways not captured; Confounding: No control group mentioned; cannot isolate effect of AI-A from other pedagogical factors; Selection bias: Single course context may not represent broader student populations; Sampling: Students self-selected into using AI assistance platform; Confounders: Course-specific factors, instructor practices, and student motivation not controlled; Generalizability: Results from one undergraduate course may not transfer to other disciplines or educational contexts

Open questions raised

Limited evidence exists on the effectiveness of GenAI for peer feedback provision; further research needed on designing AI assistance that balances pedagogical support with student autonomy and engagement
Limited evidence on GenAI effectiveness for peer feedback provision. The study identifies the need for better understanding of how to design AI assistance that balances pedagogical quality with encouraging student engagement and reflection rather than passive acceptance.
The paper identifies the need for further research on designing AI assistance in peer feedback platforms that can increase adoption rates while maintaining pedagogical quality and student autonomy.

Data: not_statedCode: not_statedExtracted from: pdfAgreement 57%

Explore related topics

AI Peer Review Peer Review Empirical Study