Twenty-four undergraduate students walked into focus groups at a UAE university between April and August 2023. They came from Vietnam, Pakistan, India, Albania, Egypt, Russia, China, Bangladesh, Spain, Australia, and Canada. None of them spoke English as a first language. All of them could tell you what plagiarism is and why it’s wrong. And most of them were using GenAI to do their academic writing anyway.
That contradiction drives a study by Hysaj, Dean, and Freeman (2025), published in Higher Education Research & Development. The paper ran four focus groups during a period when GenAI tools were everywhere on campus but institutional policy on their use was nowhere.
Students had unrestricted access to ChatGPT, Grammarly, Google Translate, and paraphrasing software, but no meaningful guidance on responsible use. What Hysaj et al. found complicates just about every assumption in the AI-in-education debate about what students know, what they do, and why those two things don’t match.
The first thing that came through clearly is that these students didn’t just find academic writing difficult. They found it alienating. Hysaj et al. report students calling it confusing, boring, abstract, disconnected from their disciplines. One participant said, “I want to write personally and not academically … since I cannot mention ‘I’ or ‘my’ in my writing I feel my writing is not mine anyways” (Participant 23, cited in Hysaj et al., 2025, p. 1693).
Read that again. The student isn’t complaining about workload. They’re saying the format has removed their voice, and once that happened, they stopped feeling ownership of anything they produced. Hysaj et al. citing Lee and Canagarajah (2019) refer to it as the “hidden curriculum” of academic writing, the unspoken rules that native English speakers absorb over years of schooling but that multilingual students have to decode on the fly with almost no support.
What struck me about the GenAI adoption Hysaj et al. document isn’t the tools. Those are predictable. What’s striking is the emotional weight students attached to them. One participant said, “I used to be so embarrassed because I didn’t understand so many words … Now I can try on my own with GenAI, first translate, then paraphrase and I feel smart” (Participant 17, cited in Hysaj et al., 2025, p. 1694).
That’s not a student talking about productivity. That’s someone talking about dignity. Another said, “Before I used to ask friends who knew English better to help me, now I don’t need anybody’s help and don’t have to thank anyone” (Participant 15, cited in Hysaj et al., 2025, p. 1694). GenAI didn’t just fix grammar for these students. It erased a power dynamic that had been following them around campus.
GenAI and Academic Writing
Hysaj et al. also found students using GenAI for critical analysis, not just grammar. One participant admitted, “GenAI can understand the ideas of journal articles and simplify them for me. I just cannot trust myself to do it on my own” (Participant 13, cited in Hysaj et al., 2025, p. 1693). That’s a different level of offloading. The student isn’t polishing sentences. They’re outsourcing comprehension.
I’ve covered a related pattern in my post on metacognitive laziness, where Fan et al. (2025) found students forming tight loops between prompting ChatGPT and revising without ever pausing to evaluate the output. Hysaj et al. add an emotional layer that study didn’t surface: students aren’t just cognitively offloading. They’re emotionally attached to these tools because the tools removed the shame and the feeling of being permanently behind.
Hysaj et al. name that complexity directly: “an expression of gratitude for GenAI was echoed, yet often entwined with a painful shame for not developing academic writing skills to the expectations required” (p. 1694). Relief and guilt, braided together. “I feel sorry that I cannot write good but I am happy that tools can help me” (Participant 11, cited in Hysaj et al., 2025, p. 1694).
And for some, the relationship had become a dependency. Hysaj et al. report that “the addictive need to rely on writing tools has also caused some distress” (p. 1694). One student said, “It is really bad because now I depend on GenAI even when I know what to write” (Participant 10, cited in Hysaj et al., 2025, p. 1694). Students also disclosed using GenAI for daily communication beyond academics, simple emails, routine messages.
The disconnect between ethical knowledge and ethical action is the most important finding here. Hysaj et al. write that “the findings demonstrate a disconnect between attitudes towards plagiarism and the unethical use of GenAI tools for academic writing” (p. 1686). Students could define plagiarism and explain why it’s unethical. And then they explained, with equal clarity, why they did it anyway. “We know plagiarism is wrong, but can’t finish assignments otherwise” (Participant 1, cited in Hysaj et al., 2025, p. 1695).
Hysaj et al. also report that “multicultural students, despite recognising the unethical nature of plagiarism, often justify its use due to external pressures” (p. 1695). High-stakes writing tasks carrying large assessment weightings made the pressure worse. And a small number of students explicitly asked for help navigating GenAI ethically. “We need to know how to use these tools in our benefit but try to be ethical too. But how?” (Participant 17, cited in Hysaj et al., 2025, p. 1695).
I wrote about a related argument in my coverage of Vosen’s (2008) Bloom’s taxonomy approach to teaching plagiarism. Vosen’s core argument was that telling students “don’t plagiarize” doesn’t count as instruction. You have to build understanding in layers, from recall through analysis to evaluation. Hysaj et al.’s findings confirm that from the student side. These students had the recall. They knew plagiarism was wrong. Nobody built the layers above that. Nobody gave them the skills or the confidence to see a path forward that didn’t involve GenAI.
Limitations
The study has real limitations. Twenty-four students at one institution is a small sample, and the findings can’t travel far across cultural contexts. The paper is qualitative and descriptive, not interventional. It tells us what students feel but doesn’t test any response. The recommendations are also broad: updated policies, professional development for educators, targeted resources for multicultural students, reconsidering whether grammar should weigh so heavily in assessment. All reasonable. But what do these look like in a multilingual classroom with thirty students from twelve countries? That question stays open.
And yet the paper does something that most policy documents don’t. It gives voice to students who are usually on the receiving end of academic integrity systems, not shaping them. Their frustration, their logic, their complicated emotional attachment to GenAI, all of it makes the GenAI-in-education conversation messier. And messier is exactly what we need. The gap between knowing and doing is a signal that institutions are still treating plagiarism as only a behavior problem when it’s also a teaching problem, and GenAI just made the consequences of that mistake impossible to ignore.
References
- 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
- Hysaj, A., Dean, B. A., & Freeman, M. (2025). Exploring the purposes and uses of generative artificial intelligence tools in academic writing for multicultural students. Higher Education Research & Development, 44(7), 1686-1700. https://doi.org/10.1080/07294360.2025.2488862
- Lee, E., & Canagarajah, S. (2019). The connection between transcultural dispositions and translingual practices in academic writing. Journal of Multicultural Discourses, 14(1), 14–28. https://doi.org/10.1080/17447143.2018.1501375
- Vosen, M. A. (2008). Using Bloom’s taxonomy to teach students about plagiarism. The English Journal, 97(6), 43-46. https://www.jstor.org/stable/40503410