Here’s something that’s puzzled educators for years. Feedback is consistently ranked among the most powerful influences on student learning. The research is clear on that. And yet, ask any instructor who’s spent hours writing detailed comments on student work, and they’ll tell you the same thing: most students barely read them. Some never open the file. Others glance at the grade and move on.
Why? Zhan, Boud, Dawson, and Yan (2025) lay out the usual suspects in a paper published in Higher Education Research & Development: large class sizes, limited curriculum time, heavy instructor workloads, and a power dynamic between teachers and students that discourages the kind of back-and-forth that feedback actually requires.
The authors argue that GenAI might help break through some of these barriers. But their framework doesn’t stop at the optimistic pitch. They’re equally interested in what can go wrong, and in the skills students need before any of it works.
The paper takes an ecological approach, which means it treats feedback engagement as the result of two forces interacting: the environment GenAI creates and the individual student’s feedback literacy. One without the other leads nowhere. A perfectly designed AI feedback tool paired with a student who doesn’t know how to prompt it, evaluate its output, or act on what it says? That’s just a more sophisticated version of the same problem we already have.
Seeking, Making Sense, and Acting
Zhan et al. organize the feedback process into three stages from Malecka et al. (2022): eliciting feedback, processing it, and enacting it. At each stage, GenAI brings clear advantages and specific risks.
The eliciting stage is where GenAI looks most promising. Students can seek feedback at any hour. No scheduling conflicts. No fear of looking uninformed in front of a professor. No navigating a top-down classroom culture that discourages questions. Zhan et al. note that studies have found students feel less anxious asking a chatbot for help because they know it’s not a real person judging them (Tai & Chen, 2024, cited in Zhan et al., 2025).
But the quality of what comes back depends entirely on how the student asks. Zhan et al. illustrate this with two examples from a trial at the first author’s university. Student A submits an IELTS writing task to ChatGPT with a vague prompt: “Can you give me feedback on my IELTS writing task?” Generic output. Useless. Student B writes a targeted prompt: act as an IELTS examiner, evaluate grammar diversity and accuracy against the scoring criteria, aim for a Band 7. She gets specific, actionable comments on her grammatical errors.
I’ve been thinking about prompt literacy for a while now, and this example nails why it matters. The gap between Student A and Student B isn’t about intelligence or motivation. It’s about knowing how to talk to the machine in a way that gets useful results back. And most students haven’t been taught that skill.
Zhan et al. also flag something the field doesn’t discuss enough. GenAI’s training data skews heavily toward US standardized English (Bender et al., 2021, cited in Zhan et al., 2025). Students prompting in their first language may get poor feedback simply because the model wasn’t built for them. And access to the best versions of these tools often requires a paid subscription, which creates a digital divide right inside the feedback process (Bozkurt et al., 2023, cited in Zhan et al., 2025). Add data privacy concerns, students worrying about where their submitted assignments end up, and the eliciting stage gets more complicated than it first appears.
At the processing stage, GenAI can reduce cognitive load and emotional friction. Students ask follow-up questions instantly, get jargon broken down into plain language, and don’t have to manage the interpersonal discomfort of face-to-face critique. But GenAI also hallucinates, generates biased responses, and sometimes produces feedback that sounds authoritative but doesn’t hold up under scrutiny. Zhan et al. warn: “If students blindly trust the feedback provided by GenAI, their engagement could be superficial or discouraged” (p. 1295).
Student B in their example handled this well. She cross-referenced GenAI’s feedback with Google, wrote follow-up prompts when something felt off, and kept her trust calibrated. Student A did none of that. He took what ChatGPT gave him and moved on. The difference between those two behaviors is exactly what the whole framework is built around.
At the enacting stage, GenAI supports iterative revision, personalized suggestions, and the ability to synthesize feedback across multiple exchanges. Students can create their own feedback loops, which is something the traditional classroom rarely offers. But Student A copied ChatGPT’s revision directly into his assignment, getting better marks without building better writing skills. Zhan et al. raise an important concern here: “over-reliance on GenAI may inadvertently limit students’ exposure to real-world feedback interactions and discourage them from engaging with humans in classroom learning or workplace communication scenarios” (p. 1295).
If students get all their feedback from a machine, they miss something important. Receiving criticism from a person you respect, working through the discomfort of disagreement, figuring out how to respond, those are relational skills. AI can’t teach them because AI removes the very friction that makes them necessary.
Feedback Literacy in the Age of AI
The central claim of the paper is that GenAI’s potential depends on what the student brings. As Zhan et al. write, “the extent to which students are engaged with feedback depends on their degree of feedback literacy as orchestrated in the GenAI context” (p. 1289). In this context, feedback literacy means prompt engineering, evaluative judgment of AI outputs, recognition of hallucinations and bias, self-regulation during revision, and ethical awareness around academic integrity.
That’s a serious list. And the paper is upfront that most students aren’t there yet. Zhan et al. propose a cyclical self-regulation model drawn from Zimmerman (2000): feedback forethought (setting goals and planning), feedback control (monitoring the interaction as it unfolds), and feedback retrospect (reflecting on the whole process afterward). Clean model and maps well onto what we know about effective learning.
My concern is practical. The model assumes students can self-regulate in an environment the paper has just spent pages explaining is full of traps. Fan et al. (2025) found exactly this problem in their study on metacognitive laziness. Students using ChatGPT for revision formed tight loops, bouncing between prompting and revising without ever stopping to evaluate whether the AI’s suggestions actually improved their work. The self-regulation model Zhan et al. propose is designed to prevent that. I’m not convinced it can do so without real scaffolding from instructors, structured practice opportunities, and a curriculum that builds these skills deliberately across the semester.
What’s Missing
This is a conceptual paper, and it’s open about that. No empirical validation of the framework. No data on how it performs across disciplines or cultural contexts. The Student A and Student B examples are illustrative, not evidence.
I also think the paper underweights the emotional dimension. It acknowledges that GenAI reduces anxiety, which is a genuine benefit. But it doesn’t fully reckon with what’s lost when feedback becomes frictionless. A critical comment from a respected teacher can sting, and that sting is often what motivates real change. A polite AI response might feel easier to process, but easier isn’t always better when the goal is growth.
Still, Zhan et al. do something valuable. They refuse to treat GenAI as salvation or threat. Their conclusion sums it up well: “GenAI can be a new enabler of student feedback engagement if it is orchestrated with what students bring, especially their feedback literacy” (p. 1300). That “if” carries enormous weight. The field needs empirical research to test whether the conditions behind it can be met at scale.
For educators reading this, the practical message is clear. Giving students access to AI feedback tools won’t solve the engagement problem on its own. You need to teach them how to ask good questions (see Cheng et al. 2025 research on the importance of students learning to ask AI the right questions), how to evaluate what comes back, and when to stop relying on the machine and do the thinking themselves. The tool opens a door. The student still has to walk through it.
References
- Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM conference on fairness, accountability, and transparency (pp. 610–623). Association for Computing Machinery. https://doi.org/10.1145/3442188.3445922
- Bozkurt, A., et al. (2023). Speculative futures on ChatGPT and generative artificial intelligence (AI): A collective reflection from the educational landscape. Asian
Journal of Distance Education, 18(1), 53–130. https://doi.org/10.5281/zenodo.7636568 - Cheng, Y., Fan, Y., Li, X., Chen, G., Gašević, D., & Swiecki, Z. (2025). Asking generative artificial intelligence the right questions improves writing performance. Computers and Education: Artificial Intelligence, 8, 100374. https://doi.org/10.1016/j.caeai.2025.100374
- Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (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
- Malecka, B., Boud, D., & Carless, D. (2022). Eliciting, processing and enacting feedback: Mechanisms for embedding student feedback literacy within the curriculum. Teaching in Higher Education, 27(7), 908–922. https://doi.org/10.1080/13562517.2020.1754784
- Tai, T. Y., & Chen, H. H. J. (2024). The impact of intelligent personal assistants on adolescent EFL learners’ listening comprehension. Computer Assisted Language Learning, 37(3), 433–460. https://doi.org/10.1080/09588221.2022.2040536
- Zhan, Y., Boud, D., Dawson, P., & Yan, Z. (2025). Generative artificial intelligence as an enabler of student feedback engagement: A framework. Higher Education Research & Development, 44(5), 1289-1304. https://doi.org/10.1080/07294360.2025.2476513
- Zimmerman, B. J. (2000). Attaining self-regulation: A social cognitive perspective. In M. Boekaerts, P. R. Pintrich, & M. Zeidner (Eds.), Handbook of self-regulation (pp. 13–39). Academic Press. https://doi.org/10.1016/B978-012109890-2/50031-7