Fostering feedback literacy by scaffolding self-regulated feedback: a comparative study of GenAI and human peers

Synthesis: In a mixed-methods quasi-experiment, scaffolding self-regulated feedback with GenAI (ChatGPT-4o) produced a small but statistically significant improvement in first-year undergraduates' feedback literacy compared to scaffolding with human peer review (ANCOVA group effect p = 0.049, η²p = 0.03). Gu, Chen, and Yan (2026) argue the advantage operates through self-regulated learning: GenAI supported students' goal setting, planning, critical evaluation, and immediate self-reflection across the forethought, control, and retrospect phases, while peer interaction retained distinct value for evaluative judgment — suggesting the two sources are complementary rather than competing.

Core Finding

The Study

The study compared two parallel English classes (N = 118 first-year undergraduates in China; GenAI group n = 56, Peer group n = 62) over three Self-Assessment cycles across one semester. Students engaged in self-assessment activities requiring interaction with external feedback: the GenAI group used ChatGPT-4o with pre-trained rubrics and prompt guidelines, while the Peer group completed structured peer-review worksheets with a high-quality exemplar. Feedback literacy was measured with a 19-item adaptation of the Feedback Literacy Behavior Scale (Dawson et al., 2024) targeting the receptive/processing dimensions (seek, make sense, use, manage affect), with α = 0.904. Qualitative data came from semi-structured interviews (nine students per group) analyzed thematically through Zhan et al.'s (2025) self-AI Regulation in Education feedback model, which frames feedback interaction across forethought, control, and retrospect phases.

Findings Across the SRL Phases

• Forethought: GenAI students proactively designed and iteratively refined prompts aligned with their learning goals, engaging in a psychologically safe environment. Peer-group students largely selected feedback sources by social convenience (nearby peers, same-major peers, roommates), rarely had specific feedback goals, and faced interpersonal barriers and social anxiety about seeking feedback.
• Control: GenAI students critically and consistently evaluated feedback accuracy and clarity, verified outputs, sometimes challenged and corrected inaccurate GenAI responses, and considered whether feedback fell within their Sociocultural Learning. Peer evaluation was more often distorted by friendship bias and perceived peer proficiency, which could impede impartial evaluative judgment.
• Retrospect: GenAI's immediate, iterative feedback enabled rapid confirmation of improvement and immediate self-reflection, helping students adjust feedback goals in subsequent cycles. Peer-group reflection was often delayed until later exams or assignments. Both groups recognized learner agency as central to effective feedback use.

Connection to Existing Knowledge Base

• Self-Regulated Learning: This paper is a direct empirical test of scaffolding SRL processes to improve feedback literacy, showing how GenAI supports goal setting, planning, critical evaluation, and immediate self-reflection.
• AI Feedback Quality: Provides evidence on students' evaluation of GenAI feedback accuracy and clarity, including instances of verifying and challenging AI output.
• Feedback Loop: GenAI's immediate, iterative feedback loop enabling rapid confirmation of improvement and self-reflection is a central mechanism.
• Peer Assessment: Offers a rigorous comparison against traditional peer review, detailing its interpersonal barriers (friendship bias, social anxiety) and unique value for evaluative judgment.
• Scaffolding / Formative Assessment: Demonstrates concrete scaffolding tools (prompt guidelines, worksheets, training) for embedding GenAI in assessment for learning.

Methodological Notes

Strengths include the mixed-methods design pairing a controlled quasi-experiment with fine-grained qualitative analysis across SRL phases. Limitations acknowledged by the authors: non-random assignment and small effect size may threaten internal validity; limited generalizability beyond the specific context; a short intervention (three cycles) that risks a novelty effect; and reliance on self-reported surveys and retrospective interviews, suggesting a need for more objective measures such as stimulated recall and learning analytics.

What this means for practice

• Instructors. Scaffold GenAI self-assessment rather than assigning it bare: supply pre-trained rubrics, prompt guidelines, and worksheets, since the GenAI group's edge over peers was only 0.17 scale points (η²p = 0.03) and teacher scaffolding is what guards against Cognitive Offloading.
• Instructors. Have the GenAI group set explicit feedback goals and refine prompts iteratively across cycles — goal setting, planning, and immediate self-reflection in the forethought and retrospect phases were the behaviors that distinguished it from the peer group.
• Instructors. Sequence GenAI and peer review in a multi-stage design — ChatGPT-4o for immediate, psychologically safe processing support and peer interaction for the audience awareness and evaluative judgment that only giving feedback builds.
• Instructors. Make peer feedback anonymous where viable: peer-group judgment was distorted by friendship bias and perceived peer proficiency, and peer students often chose partners by social convenience rather than feedback need.

Limitations

• N = 118 first-year undergraduates in two intact parallel English classes at one university in Guangdong, China; groups were not randomly assigned, so class or instructor effects cannot be excluded.
• The intervention ran three self-assessment cycles in a single semester, and the authors flag a possible novelty effect alongside the small effect size (η²p = 0.03).
• Feedback literacy was measured with a 19-item self-report scale (α = 0.904) plus retrospective interviews with only nine students per group — no behavioral, stimulated-recall, or Learning Analytics measures.
• Generalizability is limited to one EFL writing context with Mandarin L1 students.

Citation

Gu, J., Chen, J., & Yan, Z. (2026). Fostering feedback literacy by scaffolding self-regulated feedback: a comparative study of GenAI and human peers. Frontiers in Education, 10, 1736446. DOI: 10.3389/feduc.2025.1736446. (CC BY)