AI Feedback Literacy in Higher Education: Understanding, Measuring, and Predicting Student Feedback Uptake

Synthesis: Liu and Deris (2025) introduce and validate the construct of AI feedback literacy (AIFL) — students' capacity to critically engage with, evaluate, and apply AI-generated feedback — in the context of English as a foreign language (EFL) writing in higher education. Drawing on Carless and Boud's (2018) feedback literacy framework and motivational models (self-determination theory, feedback-seeking theory, cost-value models), they developed a psychometric scale and tested it with 486 Chinese undergraduate students. Using confirmatory factor analysis, PROCESS-macro mediation modeling, and multiple regression, the authors show that AIFL significantly predicts uptake of AI-generated feedback, both directly and indirectly through motivational appraisals of perceived value and perceived cost. Behavioral engagement (practices) outperformed attitudinal disposition (attitudes) as a predictor of uptake, and AIFL development was driven by frequency of AI tool use rather than demographic background — raising equity concerns about access and opportunity in increasingly AI-mediated assessment environments.

Key Findings

• AIFL is a measurable, two-dimensional construct. The 16-item AIFL scale (Attitudes and Practices subscales) showed strong psychometric properties: Cronbach's alpha 0.78–0.85, composite reliability above 0.70, AVE above 0.50 (0.52–0.60), and factor loadings of 0.58–0.83, supporting convergent validity and construct integrity.

• AIFL predicts feedback uptake both directly and indirectly. The total effect of AIFL on uptake was significant (β = 0.42, p < 0.001), with a still-significant direct effect after adding mediators (β = 0.28, p < 0.001), indicating partial mediation. AIFL raised perceived value and lowered perceived cost, which in turn positively and negatively predicted uptake respectively; both bootstrapped indirect effects were significant.

• Behavioral engagement outweighs attitudes. In a multiple regression explaining 46% of variance in uptake (R² = 0.46), perceived value was the strongest predictor (β = 0.32), followed by AIFL-Practices (β = 0.27), while AIFL-Attitudes was non-significant (β = 0.08, p = 0.091). The authors interpret this as evidence that feedback literacy develops through embodied, situated practice rather than belief alone.

• Value matters more than cost. Perceived value was a stronger predictor of uptake than perceived cost, aligning with self-determination theory accounts in which value appraisals connect to autonomous motivation; the authors argue for designing feedback experiences that raise perceived relevance and learning value.

• AIFL is shaped by use, not demographics. Only frequency of AI tool use significantly predicted AIFL (β = 0.29, p < 0.001); neither academic major nor year of study mattered. This shifts the equity conversation from individual disposition to structural access, warning that integration of AI feedback could widen digital literacy gaps for students with less exposure.

• High-AIFL students differ meaningfully. Median-split comparisons showed high-AIFL students had significantly higher uptake (d = 0.80), higher perceived value (d = 0.64), and lower perceived cost (d = 0.60) than low-AIFL peers.

What this means for practice

• Instructors. Require students to act on AI feedback — revising drafts, comparing their work against rubrics, discussing it with peers — rather than only acknowledging it, because behavioral engagement predicted uptake while attitudes alone did not.
• Instructors. Remain the critical mediator of automated input: model reflective interpretation of AI feedback and protect relational trust instead of ceding assessment to the tool.
• Administrators. Scaffold and broaden access to AI-mediated feedback for students unlikely to engage with it autonomously, since frequency of AI tool use — not major, year of study, or demographics — was what predicted feedback literacy.
• Administrators. Design feedback environments that raise perceived value and lower perceived cost: value was the strongest single predictor of uptake in the regression model.

Limitations

• The design is cross-sectional and entirely self-reported: 486 undergraduate EFL students at four Chinese universities, recruited by convenience and snowball sampling, so no causal claim about AI feedback literacy and uptake follows and generalization beyond this context is limited.
• The AIFL construct covers attitudes and practices only; the authors state it does not address emotional, ethical, or relational dimensions of feedback literacy.
• Only student perspectives were measured, so the teacher-mediation role the paper recommends is untested in the data.

Citation

Liu, K., & Deris, F. D. (2025). AI feedback literacy in higher education: Understanding, measuring, and predicting student feedback uptake. Assessment & Evaluation in Higher Education.