As I was refining the AI policy on my syllabus this term, I naturally asked AI for feedback on my syllabus 😅. One of the questions it raised was whether I wanted students to disclose their AI use.

I’ve been hesitant to require disclosure in the past because, frankly, I’m not always great about disclosing my own AI use and because I’d heard others raise concerns about requiring it. Wanting to read more about it, I asked AI to find me five articles that discussed reasons to be cautious about disclosure requirements or other caveats to consider. I knew AI could impose it’s own biased search criteria to the task (though possibly less biased than my own cherry-picking), but I also knew that I probably would put off doing my own research in favor of a more pressing task. Thus I entered my prompt.

Naturally, it also gave me summaries of the articles and a synthesis of the themes across them. I exercised my human free will and read (or skimmed) the articles myself before looking at its summaries 😇. But the articles and synthesis were useful enough that I wanted to share them here in case others are thinking through the same issue.

From ChatGPT:

Yes. There are some quite strong student-centered reasons to be cautious about blanket AI-disclosure requirements. The research doesn’t suggest “don’t ask students to disclose AI use”; rather, it suggests that how disclosure is framed matters a great deal. A reflective, low-stakes disclosure can support learning, while a compliance-oriented declaration can unintentionally produce anxiety, concealment, or mistrust.

The source I’d start with is Chahna Gonsalves, “Addressing student non-compliance in AI use declarations” in Assessment & Evaluation in Higher Education. This is unusually relevant because it examines an actual mandatory AI-use declaration at King’s Business School. Up to 74% of students left the declaration blank; interviews suggested several reasons: students were unclear about what counted as AI use, worried that disclosure would make instructors suspicious or affect their grade, experienced inconsistent expectations across courses, and sometimes saw declaring every small use as invasive or burdensome. One particularly important finding is that students could interpret disclosure as essentially “admitting to something wrong” even when the use was permitted.

A second excellent source is Jiahui Jess Luo, “How does GenAI affect trust in teacher-student relationships?” This study is explicitly centered on students’ assessment experiences. Students described a lack of “two-way transparency”: they might be required to declare AI use or even submit ChatGPT records, while being unsure how instructors themselves were using AI or how an AI disclosure would influence grading. Luo argues that this asymmetry can reinforce the instructor-student power imbalance and make disclosure feel like surveillance rather than transparency; some students reported avoiding even permitted AI uses because they feared how instructors would interpret them.

There is also a very new 2026 study, “Should I tell my teacher? Student AI disclosure practices, stigma, and self-regulated learning in higher education.” It found that fear of teacher judgment and social stigma tended to occur together and was associated with students concealing AI use. Importantly, the authors specifically caution that mandatory AI-use declarations aimed only at compliance may be insufficient; disclosure is also an emotional/interpersonal decision. They recommend reducing the perceived cost of honesty—for example, making clear that disclosing permitted AI use does not automatically mean a penalty or suspicion. It’s a small single-course study (78 students), so I’d treat it as emerging evidence rather than definitive.

Oregon State Ecampus published a useful teaching-oriented synthesis this month. It raises a deceptively basic issue: what exactly counts as “AI use”? Does an AI-generated Google summary count? Grammarly? Grammar correction? Brainstorming? At what threshold does disclosure become necessary? It also points out a potential relational problem when instructors require students to disclose AI use but do not disclose their own use of AI in teaching or grading, and notes that extensive documentation requirements can add considerable work for students.

And Jisc’s 2025 research on student perceptions gives useful broader context. Even when institutions technically have AI guidance, students continue to report uncertainty about what is permitted, anxiety about being perceived as cheating, concern that AI use may make their work seem less original or effortful, and frustration with inconsistent rules across instructors and courses. Students also raise equity, privacy, and accessibility concerns and say they want to be involved in shaping AI practices rather than simply having policies imposed on them.

What I think the caveats boil down to

From these sources, I’d identify five student-experience questions worth asking before requiring disclosure:

• Will disclosure feel informational or incriminating? If students think “yes, I used AI” puts a target on their work, the policy may actually encourage concealment.
• Can students reasonably know what they are supposed to disclose? AI is increasingly embedded in search engines, writing tools, browsers, Office products, etc.; “disclose any AI use” is becoming surprisingly hard to interpret.
• Could disclosure bias how I evaluate the work? Students themselves worry that instructors may see an identical paper differently after learning that AI was involved.
• Is the burden proportionate to the purpose? A two-sentence explanation of substantive AI assistance is very different from requiring prompts, outputs, chat histories, and declarations on every assignment.
• Is transparency reciprocal? Requiring detailed student transparency while faculty use AI without disclosure can undermine the trust that the disclosure policy is intended to create.

If the goal is to make disclosure useful without those downsides, a synthesis of these sources would point toward something like: require disclosure only for meaningful AI involvement; clearly define what does and doesn’t need disclosure; explain why you’re asking; explicitly say that permitted use will not lower the grade; and don’t require extensive chat logs unless there’s a specific pedagogical reason. That seems much more defensible from the student-experience literature than “declare all AI use.”

Disclosure statement: I used to ChatGPT to find the articles using the prompt “I feel like there are pros and cons to having a GenAI disclosure policy in one’s class. Could you find me sources that discuss why one should be cautious, or other caveats? Centered on student experience.” Then I pasted ChatGPT’s response above. I also used ChatGPT to refine my blog post with the prompt “how can I improve the clarity of this opening to a blog post”.