Students value AI feedback most when teacher judgement stays in the loop
Updated May 18, 2026
At Student Voice AI, we increasingly see universities ask whether AI feedback is genuinely helping students, or simply speeding up a process they still do not fully trust. That is why Yağmur Öztürk and Ayça Çebi's Assessment & Evaluation in Higher Education paper, "The potential of AI-generated feedback from the students’ perspective: a systematic review", matters. For UK universities collecting feedback on assessment, digital support, and GenAI pilots, the review offers a useful correction to simple pro-AI or anti-AI narratives. Students do see value in AI feedback, but much like earlier evidence that teacher feedback still carries more trust when the stakes are high, they care about context, credibility, and whether the feedback feels usable.
Context and research question
Generative AI has made it much easier for institutions to imagine faster feedback at scale. But speed is only one dimension of feedback quality. If students experience AI-generated comments as generic, misaligned, or emotionally thin, universities may create a faster process without creating a better one.
Öztürk and Çebi examine that problem through a systematic review focused on higher education students' perspectives on AI-generated feedback in online learning. The review is organised around three practical questions: how AI feedback contributes to students' cognitive and emotional development, how students compare AI feedback with teacher feedback, and what students expect AI systems to provide. That makes the paper useful for UK Student Experience, Quality, and Market Insights teams because it synthesises what students say they want before an institution locks AI feedback into policy or workflow.
Key findings
Students valued AI feedback for more than speed alone. The abstract reports that students saw both cognitive and emotional benefits in AI-generated feedback. That matters because it suggests AI feedback is not only a labour-saving device from the institution's perspective. For some students, it can also reduce uncertainty, help them move forward, and make feedback feel more continuously available between taught interactions.
Students did not converge on a simple replacement model. Some preferred teacher feedback, others preferred AI, and many wanted a combination of both. That is the most useful finding for UK teams. The practical choice is not usually human or machine. It is which parts of the feedback process can be supported by AI without weakening the parts students still want from teachers. That blended preference also fits wider evidence that students often turn to GenAI when support feels private and instant, especially when they want help outside normal teaching contact.
The clearest expectation in the abstract is that students wanted feedback that was:
"detailed, constructive, and context-sensitive."
That short phrase is more useful than a generic statement that students "like quality feedback". It tells universities exactly where weak AI feedback is likely to fail. If comments are quick but vague, supportive but non-specific, or technically correct but poorly matched to the task, students are unlikely to treat them as strong feedback even if they welcome the convenience.
The review therefore points towards design quality, not just technical novelty. Students were not asking whether AI feedback was impressive in the abstract. They were asking whether it helped them understand what to do next, whether it felt relevant to their work, and whether it could sit alongside teacher judgement rather than pretending to replace it. For UK higher education teams, that is a useful design rule: the more consequential the task, the more important human context and academic judgement remain.
Practical implications
For UK universities, the first implication is to evaluate AI feedback with more than a usage rate or satisfaction score. Ask whether students found the feedback specific, actionable, trustworthy, and worth using in later work. Ask when they used it, draft-stage support, clarification, revision, or something closer to judgement. That produces sharper evidence for deciding whether an AI feedback pilot is actually improving the student experience.
Second, institutions should design AI feedback as a layer in the support system, not as a wholesale substitute for teachers. The review suggests many students are open to blended models, where AI helps with speed, iteration, and early clarification, while staff provide disciplinary judgement, reassurance, and nuance. That is a more credible route than treating automation as the main point of innovation. The benefit is simple: students get faster support without losing the human judgement they still value.
Third, universities should collect and analyse open comments on what makes AI feedback feel useful or risky. A yes-or-no question about whether students used AI feedback will miss the real signal. Teams need to know whether students found the comments generic, helpful, overconfident, impersonal, or genuinely developmental. If those comments are being gathered across pilots or modules, they need a defensible workflow for analysing open-text feedback at scale rather than ad hoc interpretation. That gives quality teams evidence they can actually use.
Finally, institutions should treat AI feedback as a governance issue as well as a pedagogic one. Comments about trust, fairness, inconsistency, or unclear boundaries can become sensitive quickly, especially when students are discussing assessed work. A governance checklist for student comment analysis is a sensible starting point. That helps teams separate experimentation from evidence use, and reduces the risk of scaling a process students experience as fast but thin.
FAQ
Q: How should universities test AI-generated feedback before scaling it?
A: Start with a narrow use case and gather targeted feedback on speed, specificity, trust, and actionability. Compare responses across assessment types and stakes. If students describe AI comments as generic or hard to act on, keep the pilot small and redesign it before broader rollout.
Q: What are the methodological limits of this paper?
A: This is a systematic review rather than a single UK institutional dataset. Its value lies in surfacing recurring patterns in how higher education students perceive AI-generated feedback, but universities still need local evidence because expectations will vary by discipline, assessment design, and the role AI is being asked to play.
Q: What does this change about student voice on AI more broadly?
A: It shifts the question from "Did students use AI feedback?" to "What kind of feedback do students trust, act on, and want repeated?" That makes student voice more diagnostic. It tells universities not only whether AI is present, but whether it is improving learning support or just accelerating generic responses.
References
[Paper Source]: Yağmur Öztürk and Ayça Çebi "The potential of AI-generated feedback from the students’ perspective: a systematic review" DOI: 10.1080/02602938.2025.2588385
Request a walkthrough
Book a free Student Voice Analytics demo
See all-comment coverage, sector benchmarks, and reporting designed for OfS quality and NSS requirements.
- All-comment coverage with HE-tuned taxonomy and sentiment.
- Versioned outputs with TEF-ready reporting.
- Benchmarks and BI-ready exports for boards and Senate.
UK-hosted · No public LLM APIs · Same-day turnaround