Students want AI feedback chatbots to clarify comments, not replace judgement
Updated Jul 29, 2026
Students often receive feedback, understand part of it, and still do not know what to do next. At Student Voice AI, that matters because open comments about feedback rarely stop at "more detail please". They often point to a deeper problem: students do not always feel able to ask follow-up questions, challenge inconsistencies, or turn written comments into action. Noorhan Abbas's Assessment & Evaluation in Higher Education paper, "Investigating student perceptions of an AI-powered chatbot to support feedback interpretation and uptake in higher education", is useful because it treats that gap as a student experience problem, not just a marking-process problem.
Context and research question
Feedback remains one of the weaker parts of the UK student experience story. The paper notes that the National Student Survey item on whether feedback helps students improve their work reached 75.2% positive responses in 2025, which is better than many institutions once expected but still leaves a large minority unconvinced. For universities already using surveys, module evaluations, and student voice in assessment and feedback to understand what is going wrong, that matters because dissatisfaction with feedback is often about interpretation as much as turnaround.
Abbas examines that problem in an online postgraduate context at a UK research-intensive university. The study focused on MSc computing students and explored their perceptions of a proposed AI chatbot designed to help them engage with feedback already given by markers. Importantly, this was not a live trial of a fully deployed tool. It was a mixed-methods design using four focus groups and a structured survey to test how students imagined such a chatbot should work, what they would trust it to do, and where they would draw the line.
That makes the paper practical for UK higher education teams. It sits at the intersection of feedback literacy, AI governance, and student experience evidence. The core question is not whether AI can produce more feedback. It is whether AI can help students make better sense of feedback they already have, without weakening academic judgement or institutional trust.
Key findings
The first finding is that feedback disengagement was driven by structural and affective barriers, not simple apathy. Students described delayed feedback, inconsistent standards, and the awkwardness of asking for clarification after comments were released. In other words, the barrier was often not a lack of interest in improving. It was the friction involved in working out what the feedback meant and whether it was worth challenging or discussing.
The paper captures that friction starkly:
"69% of respondents wanted to ask questions but did not."
That figure matters because it shows how much potentially useful dialogue never happens. A feedback system can look complete on paper while still leaving most students to interpret complex comments alone.
Students were interested in AI support, but not as an answer machine. According to the abstract, they valued Socratic and dialogic interaction more than direct answer provision. That is an important distinction. Students did not simply want a tool that told them what to write next time. They wanted help unpacking comments, understanding criteria, and thinking through the logic of the marker's response. For UK institutions, that is a much more credible use case than generic AI tutoring bolted on after the fact.
Trust depended on contextual grounding. Students wanted any chatbot to work from their own submission and the actual comments they had received, not from general study advice. That is a strong design signal for universities now exploring AI in feedback processes. If a tool cannot show that it is anchored in the real assessment context, students are much less likely to treat it as reliable or fair.
Human judgement remained the non-negotiable boundary. The abstract says students maintained human judgement as essential, even while rating the proposed tool positively overall. That is one of the paper's most useful conclusions. Students were not rejecting AI outright. They were accepting it on clear terms: the tool could mediate, clarify, and prompt reflection, but it could not replace the marker's judgement or become the final authority on what the feedback meant.
Taken together, these findings suggest that AI support for feedback is most acceptable when it reduces friction without pretending to replace academic expertise. That is a narrower claim than many technology pitches make, but it is a much more useful one for institutions trying to design something students will actually trust.
Practical implications
For UK universities, the first implication is to treat feedback interpretation as a measurable part of the student experience. Many institutions ask whether feedback was timely or useful, but fewer ask what made it hard to apply. This paper suggests that questions about hesitation, confusion, and follow-up matter just as much. If students want clarification but do not seek it, the feedback process is weaker than headline satisfaction figures imply. That gives Student Experience and quality teams a clearer starting point for improvement.
Second, institutions piloting AI support should define the use case tightly. The paper points towards a tool that explains, prompts, and guides, not one that generates replacement answers or overrules staff. In practice, that means grounding any tool in the student's own assessment context, setting clear boundaries on what it can and cannot do, and avoiding the false confidence that comes with generic outputs from large language models used without HE-specific controls. The benefit is higher trust and lower governance risk.
Third, universities should analyse feedback comments for recurring interpretation problems, not just general sentiment about feedback quality. Students in this study surfaced a pattern of delay, inconsistency, uncertainty, and social friction around seeking clarification. Those themes are unlikely to appear cleanly in one score. They show up in free text. Using a reproducible method for NSS and module-evaluation comment analysis makes it easier to compare where those issues cluster by programme, mode, or cohort. That gives teams a more actionable improvement queue.
Finally, institutions should keep the human route visible. Students may welcome AI support when it helps them prepare better questions or understand marker comments more clearly, but the legitimacy of the process still rests on academic staff being available to make and explain judgements. If universities keep that escalation path explicit, they improve the odds that AI support will feel assistive rather than evasive. That is a practical safeguard as well as a trust benefit.
FAQ
Q: How should a university pilot an AI feedback chatbot without creating more confusion?
A: Start with one tightly defined use case: clarifying existing feedback, not generating new feedback or grading advice. Anchor the tool in the rubric, the student's own submission, and the marker's comments; keep a clear escalation route to a human academic; and review the resulting student comments with a documented process. If the pilot also collects open-text reactions, a student comment analysis governance checklist helps keep the evidence trail clear.
Q: What are the methodological limits of this study?
A: This was one mixed-methods study in an online MSc computing context at a UK research-intensive university, and it examined perceptions of a proposed tool rather than measured outcomes from a live deployment. That means the paper is strongest as design guidance, not proof that chatbot support will improve feedback uptake everywhere. UK universities should use it to shape local pilots, then test whether trust, usefulness, and actual student action improve in their own setting.
Q: What does this change about student voice around feedback more broadly?
A: It suggests that student voice on feedback should move beyond asking whether comments were "helpful" in the abstract. Universities also need to know where interpretation breaks down, when students avoid asking for clarification, and what kind of support feels legitimate. That is a useful shift because it turns feedback evidence from a broad satisfaction measure into a clearer account of where action is being blocked.
References
[Paper Source]: Noorhan Abbas "Investigating student perceptions of an AI-powered chatbot to support feedback interpretation and uptake in higher education" DOI: 10.1080/02602938.2026.2706016
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