Siu Cho Shing and David Carless, University of Hong Kong
Our small-scale research investigates the kind of feedback that undergraduate students find meaningful, and how generative artificial intelligence (GenAI) fits into the way they learn. We gathered data from 28 students across six of Hong Kong’s publicly funded universities, of whom 17 completed a questionnaire and 11 took part in semi-structured interviews. Students could respond in English or Chinese according to their preference.
One part of the questionnaire asked about something that has quietly become part of many students’ routines. Before handing in a piece of work, they paste it into ChatGPT or a similar GenAI tool and ask for comments. Of the 28 students in our study, 24 had done this at least once, and about a third did so regularly.
The appeal of GenAI feedback is easy to understand. Students can upload a draft at any hour, revise the same paragraph repeatedly, and ask follow-up questions without feeling that they are imposing on anyone. For catching surface problems such as grammar, weak structure, or clumsy phrasing, they found it quite useful. It also helped them get a quick sense of whether a piece of work was heading in the right direction.
Yet the same students were quick to identify limits to GenAI feedback. Students repeatedly observed that the feedback was too general and rather lenient. Some noticed that when they asked GenAI to estimate a grade, its appraisal was often higher than the grade they eventually received, which made them trust its judgement less. Others felt GenAI could not get inside the particular expectations of a course. A fourth-year student noted that even when the marking rubric was fed into the tool, the advice that came back remained vague and generic. Positive comments came easily, but specific criticism was harder to get. GenAI feedback was encouraging, but not always diagnostic.
Students talked about teacher feedback in rather different terms. They frequently recounted that teachers know what they are looking for, because teachers set the criteria in the first place. As a second-year student put it: “Because teachers made the rubric, they know what to look out for. GenAI isn’t too sure about this.” A third-year student elaborated that GenAI was good at flagging repetition, loose structure and minor errors but his teachers understood the demands of the course and could comment on his reasoning in ways the tool could not. He described the most useful teacher comments as the kind that named a precise weakness, such as an example that did not quite fit the local context or a section that needed more support from the course readings. Students felt that this sort of targeted comment could only come from someone who knew the course well.
These findings resonate with large-scale recent research in Australia by Henderson and colleagues which compared students’ views of GenAI and teacher feedback. Students appreciated GenAI feedback for its ready-availability and perceived it as more positive and less anxiety-inducing than teacher feedback. Teacher feedback was considered to be more relevant and contextualized in that teachers generally possess deep contextual knowledge of the assessment, course and learner, and so are able to identify the essence of quality work.
When we asked whether GenAI should take on a more formal role in giving feedback on drafts, students were thoughtful and, on the whole, cautious. Many could see real advantages, particularly in large courses where a teacher cannot realistically offer every student a lot of detailed comments on a first draft. But students also worried about depending on GenAI too heavily. Some feared that frequent acceptance of GenAI suggestions would weaken their own critical judgement over time, resonating with the concept of metacognitive laziness. Others pointed to the risk of inaccurate or invented information, and to feedback that drifts away from what a course rewards. Most felt that GenAI could serve well as a first pass, but should not stand in for a teacher’s professional judgement.
Students do not see these two kinds of feedback as competing but as playing different roles. GenAI is valued for its speed and for being available at any hour. What teachers bring is different: they know the course, they set the criteria, and their feedback reflects that. The two work best together, each covering ground the other does not. Students were not just receiving feedback from two sources. They were choosing when to use each one, turning to GenAI for quick iteration and to teachers for judgement against course expectations. Hybrid approaches to feedback have attracted growing interest in the academic literature, and our study suggests that students may already be practising something like this on their own. If that instinct is already there, the task for educators is to recognize it and help students do it well.
References
Banihashem, S. K., et al., (2025). Pedagogical framework for hybrid intelligent feedback. Innovations in Education and Teaching International. https://www.tandfonline.com/doi/full/10.1080/14703297.2025.2499174
Fan, Y., et al., (2025). Beware of metacognitive laziness: Effects of generative artificial Intelligence on learning motivation, processes, and performance. British Journal of Educational Technology, 56(2), 489-530. https://doi.org/10.1111/bjet.13544
Henderson, M., et al., (2025). Comparing Generative AI and teacher feedback: Student perceptions of usefulness and trustworthiness. Assessment & Evaluation in Higher Education, https://www.tandfonline.com/doi/full/10.1080/02602938.2025.2502582
