Students' Perceptions of Generative AI-Powered Learning Analytics in the Feedback Process: A Feedback Literacy Perspective

Synthesis: Jin, Maheshi, Lai, Li, Gasevic, Chen, Charwat, Chan, Martinez-Maldonado, Gašević & Tsai (2025) examine how generative AI (GenAI) can support students' engagement with feedback through a feedback literacy lens. They piloted PolyFeed, a student-facing learning analytics feedback tool whose two GenAI features (both running ChatGPT 3.5) were: a ChatGPT explanation function that reformats or explains specific educator feedback in simpler, bullet-pointed language (capped at two attempts per item), and a GenAI-powered dashboard visualizing common strengths and weaknesses across assessments and units, derived from students' own feedback annotations and inductive thematic coding. Data came from 18 higher-education students across information technology, education, business/economics, and engineering, spanning three phases: introductory lab sessions, in-semester trace-data use, and post hoc interviews.

The central finding is a discrepancy between initial perceptions and actual usage. During the introductory lab sessions, students reacted overwhelmingly positively — all 18 reacted positively to ChatGPT explanations and 16 to the visualizations, seeing them as aids to sense-making, reflection, and acting on feedback. Yet in-semester trace data showed only modest engagement: just half used the ChatGPT explanation function (mean rating 3.23/4), and Visualization access ranged 50–67% per graph. Post hoc interviews revealed three drivers of the drop: a mismatch between students' expectations and GenAI outputs, a lack of relevance across varied disciplines, and a sense that the features were redundant when educator feedback was already clear. The authors argue GenAI can help close the feedback loop and shift students from passive recipients to active participants, but call for adaptive, discipline-specific, explainable designs that set realistic expectations and give learners control over AI interactions.

Key Findings

• Strong initial enthusiasm that did not fully translate into use. In lab sessions, 100% of students reacted positively to ChatGPT explanations and 89% to GenAI visualizations; half said explanations helped them make sense of vague feedback, and 14 valued visualizations for using feedback and reflecting. Yet in-semester use was modest (50% used explanations; visualization access 50–67%), and positive reports declined in post hoc interviews.
• Feedback literacy mapping. Students connected the features to several feedback literacy components (from Carless & Boud and Molloy et al.): making sense of feedback, seeking feedback information, using feedback information (reflection, monitoring, action planning), and — notably — managing emotions, as visualizations and explanations reduced the overwhelm of dense, "jumbled" feedback and offered a low-stakes intermediary before approaching tutors (especially valued where cultural norms made direct tutor contact uncomfortable).
• Redundancy and relevance limits engagement. Some students found the functions unnecessary when feedback was already clear ("not needing to use it"), and criticized explanations as repetitive or unable to handle subject-specific content such as mathematics. These findings echo broader questions about AI feedback quality and when AI adds value versus when it duplicates what educators already provide.
• Trust and expectation-setting. Students raised concerns about the reliability of the older ChatGPT version and about privacy (whether their details were sent to ChatGPT). The authors recommend transparency about GenAI capabilities, explainable-AI principles, and greater user control over outputs to build trust and foster sustained self-regulated learning.

What this means for practice

• Instructors. Set expectations before deploying GenAI feedback features: a mismatch between what students expected and what the tool produced was one of three reasons that 100% positive lab reactions shrank to 50% in-semester use.
• Designers. Make explanation features discipline-specific and adaptive: students criticized the ChatGPT explanations as repetitive and unable to handle subject-specific content such as mathematics.
• Instructors. Check what a feature adds over your own comments before offering it — redundancy with already-clear educator feedback was a stated reason for non-use ("not needing to use it").
• Learners. Use explanations and visualizations as a low-stakes first pass: students reported they made sense of dense "jumbled" feedback, supported reflection, and reduced the emotional overwhelm of comments before approaching a tutor.
• Researchers. Measure engagement with trace data rather than intentions: all 18 lab-session students reacted positively to the explanations, yet only 9 used the function during the semester (mean satisfaction 3.23/4).

Limitations

• The study rests on a small convenience sample: 18 students completed the lab sessions and in-semester phase, so reported percentages often rest on counts of 9 or 11 students.
• All findings come from one institution and four disciplines, and the features ran on ChatGPT 3.5 — an older model whose reliability students themselves questioned — so the design conclusions may not carry to current models.
• The study captures perceptions and access, not learning: there was no comparison condition and no measure of feedback quality, so it cannot show that the features improved feedback use or outcomes.
• Trace data recorded access counts (visualization engagement of 50–67% per graph), which cannot distinguish a brief glance from sustained use.

Citation

Jin, F. J.-Y., Maheshi, B., Lai, W., Li, Y., Gasevic, D., Chen, G., Charwat, N., Chan, P. W. K., Martinez-Maldonado, R., Gašević, D., & Tsai, Y.-S. (2025). Students' Perceptions of Generative AI-Powered Learning Analytics in the Feedback Process: A Feedback Literacy Perspective. Journal of Learning Analytics, 12(1), 152–168.