A student turns off Grammarly before starting her assignment. Not because it’s banned, but because she’s heard that another student got a six-month study deferral after a high AI score on Turnitin. She doesn’t know exactly what counts as a violation, so she eliminates every tool that could possibly trigger a flag. The university’s policy says GenAI use is allowed when “ethical and appropriate,” but nobody has told her what that actually looks like for her specific course, her specific professor, her specific essay.
That kind of fear is at the center of a 2025 qualitative study by Jiahui Luo, who interviewed 11 university students in Hong Kong about how they navigate trust with their teachers now that GenAI is part of the assessment picture. The students also created concept maps to visualize their experiences. The university in question had a relatively liberal AI policy and even provided students with free access to ChatGPT through its intranet. And yet, fear, not confidence, was the dominant emotion students brought to their assignments.
Luo’s central finding is an absence of what she calls “a lack of ‘two-way transparency’ between teachers and students regarding the use of GenAI in assessment” (p. 1001). Students are required to declare their AI use, submit chat records, and sign accountability forms. But teachers rarely reciprocate. Few professors explained how AI-mediated work would actually be graded.
The Turnitin AI score? Only visible to teachers. Students had no way to check whether their own work would be flagged before they submitted it. One student described the process as operating inside a “black box.” Luo frames this directly: “the same level of transparency is not observed from the teacher’s side” and argues that “the one-way transparency may be interpreted as a surveillance mechanism that reinforces top-down control and a power imbalance between students and teachers” (p. 1001).
This connects to a pattern I’ve seen across the research I cover on this blog. When I wrote about Luo’s earlier policy analysis of 20 top universities, the findings were strikingly similar: GenAI policies overwhelmingly framed students as potential cheaters and focused on policing originality. That paper argued for moving past a narrow originality lens. This new study shows what that narrow framing actually does to students on the ground. They don’t just feel policed. They stop engaging with the tools at all, even when the tools are technically permitted.
The declaration problem is especially telling. Students in Luo’s study were reluctant to acknowledge their AI use even when it was allowed. They worried that admitting to using ChatGPT for brainstorming or grammar corrections would lead to lower grades or negative assumptions about their abilities.
One student was direct about it: if you acknowledge AI use, the teacher may think “you are not independent enough, your language skills are poor, or even you’ve used AI to write most parts of the assignment.” The ambiguity around how disclosure factors into grading created enough doubt that many students chose to hide their use entirely.
I’ve seen this same tension in McDonald et al.’s (2025) analysis of GenAI policies across 116 US R1 universities, where most institutional guidelines focused on restrictions and penalties and very few addressed how AI-mediated work should actually be evaluated. The pattern is consistent: institutions create frameworks that demand student transparency without offering any transparency in return.
GenAI and Trust in Education
Luo also found that personal connection played a major role in trust-building. Students in large classes felt anonymous, and several believed their professors would default to suspicion because they didn’t know them as individuals. One student emailed a professor after seeing a low “human originality score” on an open-source detector.
The professor responded reassuringly, but the student still felt uneasy because the professor didn’t really know her. In contrast, when a professor spent class time walking students through prompts, reflecting on AI outputs, and openly acknowledging that AI detection can be inaccurate, students felt significantly safer. The professor’s attitude functioned as a trust signal.
Luo frames this through the concept of “competence trust,” drawing on Reina and Reine (2006), which describes the belief that someone has the necessary skills for their role. What’s new here is that students now expect that competence to extend to AI literacy, not just subject expertise. And the generational dynamics matter too: Chan and Lee (2023), cited in the paper, found that Gen Z university students are more optimistic about GenAI than their Gen X and millennial professors. Students are ready to engage. They’re waiting for their teachers to catch up.
Some students also raised fairness concerns. They felt it was wrong that classmates could submit AI-generated essays and receive decent grades without being caught, and they expected teachers to recognize AI writing patterns and redesign assignments so they couldn’t be easily completed by AI.
One student discovered that a professor’s own course outline appeared to be AI-generated after running it through a detector. She called it hypocrisy. Luo notes that the student didn’t consider the detection tool might be inaccurate, but the perception itself was enough to damage trust. That detail is instructive: accuracy doesn’t matter if the emotional response has already been formed.
Luo argues that the tone of policy communication is a decisive factor. She observes that “when GenAI policies are communicated to students in a surveillance tone, students tend to adopt a self-defensive approach to avoid any perceived risks of wrongdoings” (p. 1002). And under this framing, “teachers could be ‘subjectified less as educators but more as gatekeepers to avoid academic misconduct’” (p. 1002).
I’ve written before about how multicultural students navigate GenAI in academic writing and the vulnerability that comes with unclear expectations. Luo’s findings extend that problem: it’s not just about unclear rules. It’s about a communication culture that makes students feel they’ll be punished for engaging honestly.
The study recommends reframing AI declarations as self-reflection opportunities, making grading criteria for AI-mediated work explicit, co-designing GenAI policies with students, and training teachers to demonstrate AI literacy openly. These are good recommendations. The co-design element is especially important because it shifts the relationship from compliance to collaboration.
Limitations
Now, the limitations. This is 11 students at a single Hong Kong university, mostly from education programs. These students had dual identities as learners and future teachers, which gave them unusually reflective perspectives but also makes the findings hard to generalize.
The data was collected in late 2023 or early 2024, when GenAI policies were still brand new at most institutions. By 2026, many universities have iterated on their frameworks, and student familiarity with AI tools has grown considerably. Some of the fears Luo documents may have eased in institutions that moved toward more supportive approaches. That said, many haven’t, and the core dynamic Luo identifies, the power asymmetry in transparency, is structural. It won’t resolve on its own just because time passes.
Luo concludes that “a trusting relationship between teachers and students is important because trust indicates a mutual willingness to take risks and collaboratively explore this evolving AI landscape” (p. 1004). I’d agree. And the practical lesson from this study is straightforward: if you want students to be open about their AI use, you have to go first. Show your own process. Explain your grading logic. Admit the detection tools are imperfect. Trust is built by demonstration, not by declaration forms.
References
- Chan, C.K.Y., and K.K.W. Lee. 2023. The AI generation gap: Are Gen Z students more interested in adopting generative AI such as ChatGPT in teaching and learning than their Gen X and millennial generation teachers? Smart Learning Environments 10, no. 1: 1–23.
- 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
- Luo, J. (2025). How does GenAI affect trust in teacher-student relationships? Insights from students’ assessment experiences. Teaching in Higher Education, 30(4), 991–1006. https://doi.org/10.1080/13562517.2024.2341005
- Luo, J. (2024). A critical review of GenAI policies in higher education assessment: A call to reconsider the “originality” of students’ work. Assessment & Evaluation in Higher Education, 49(5), 651-664. https://doi.org/10.1080/02602938.2024.2309963
- McDonald, N., Johri, A., Ali, A., & Hingle Collier, A. (2025). Generative artificial intelligence in higher education: Evidence from an analysis of institutional policies and guidelines. Computers in Human Behavior: Artificial Humans, 3, 100121. https://doi.org/10.1016/j.chbah.2025.100121
- Reina, D.S., and M.L. Reine. 2006. Trust and betrayal in the workplace: Building effective relationships in your organization. Oakland: Berrett-Koehler