When AI rules stay vague, staff and students improvise fairness
Updated Jul 05, 2026
student voicefeedbackWhen AI guidance is vague, students and staff do not stop making decisions until policy catches up. They improvise, and the result can be uneven expectations about what counts as fair, acceptable, or risky. That is why Ayesha Afzal, Shahid Rafiq and Martin Oliver's Teaching in Higher Education paper, "Moral improvisation with generative AI in higher education: faculty and student experiences in Pakistan", matters. For universities already reviewing student experiences of GenAI across UK universities, it shows why the next question is not only whether students use AI, but how they experience the rules around it.
Context and research question
Many universities now have central guidance on Generative AI, academic integrity, and disclosure. The harder problem is local translation. A policy may look clear on paper, while students still meet different expectations across modules, assessment types, and staff teams. In practice, that can turn AI use into a judgement call made under pressure rather than a rule applied with confidence.
Afzal, Rafiq and Oliver examine that gap through semi-structured interviews with teachers and students in Pakistani universities, focused on Education and English departments where writing, feedback, and assessment practices are especially exposed to AI-related uncertainty. The paper asks how faculty and students decide what counts as acceptable GenAI use when formal guidance is limited, uneven, or still developing. The context is not UK-based, but the governance problem travels well because UK institutions are also trying to turn broad AI policy into credible course-level practice.
Key findings
The first finding is that unclear guidance does not remove judgement, it relocates it. When formal rules offered limited direction, participants relied on personal judgement to decide what felt appropriate, fair, or defensible. The paper frames that pattern as moral improvisation: situated decision-making under uncertainty, shaped by norms, relationships, and the immediate demands of academic work. For UK universities, that matters because ambiguity does not create neutrality. It pushes governance into local interpretation.
The paper states that point plainly:
"participants relied on moral improvisation, making practical and ethical decisions in the moment when formal rules offered limited direction."
The second finding is that expectations become uneven very quickly. The abstract describes differing expectations across classes and peer groups, which means the same kind of AI assistance may feel acceptable in one setting and risky in another. That matters for universities because students experience AI governance through real assignments, conversations, and peer comparison, not only through a central policy page. Uneven local expectations create fairness problems before any formal misconduct case appears.
The third finding is that staff and students approach the same situation from different responsibilities. The introduction shows faculty focusing on academic standards, assessment fairness, and their responsibilities towards students, while students are more focused on meeting course requirements and managing their learning effectively. That mismatch can make policy feel clearer to institutions than it feels to students. What looks like a simple rule from above may still feel like a live negotiation from below.
The fourth finding is that the impact is emotional as well as procedural. The abstract links GenAI use to new forms of uncertainty, emotional responsibility, and uneven expectations. That is an important point for student voice work. AI questions are not only about tool use or rule knowledge. They affect whether students feel safe asking for help, honest disclosing assistance, or confident drawing their own boundary between support and overreach, a tension that also appears when students disclose AI use only when governance feels fair and trustworthy.
Practical implications
For UK higher education teams, the first implication is to move from central policy to module-level guidance. Students need concrete examples of permitted, discouraged, and prohibited AI use in specific tasks, especially in writing-heavy assessments. If a disclosure expectation exists, it should appear in the brief, marking guidance, or submission instructions, not only in a distant policy document. That reduces avoidable guessing.
Second, universities should collect student feedback on how AI rules land in practice, not only whether students know they exist. Ask where guidance felt clear, inconsistent, unfair, or too vague to use. That is close to the question raised by earlier evidence that AI surveys should ask about competence, authorship, and fairness, rather than settling for a simple usage rate. The benefit is feedback that points to implementation gaps rather than abstract opinion.
Third, institutions should treat AI-related open comments as governance evidence. If students describe mixed messages between modules, uncertainty about acceptable help, or fear that disclosure will be interpreted inconsistently, those comments should be reviewed as policy signals, not dismissed as background noise. This is where Student Voice Analytics fits naturally: it helps universities group themes such as fairness, clarity, trust, disclosure, and inconsistency across open-text feedback at scale. Pair that with a student comment analysis governance checklist and teams have a stronger route from comments to action. The payoff is clearer evidence before uneven local practice hardens into routine.
Finally, universities should compare staff and student interpretations after major assessment points. A short pulse question after an assessment window can test whether students understood the AI guidance in the way staff thought they had communicated it. Where those accounts diverge, the issue is not only compliance. It is a design problem in guidance, communication, or assessment framing. That gives institutions a faster way to correct ambiguity.
FAQ
Q: How should a university gather usable feedback on its AI guidance?
A: Start small and stay close to real assessment decisions. After a major assignment or pilot, ask students whether the rules felt clear, whether disclosure felt safe, and what kinds of AI use still felt ambiguous. Then add one open-text prompt asking where guidance was easiest or hardest to apply. That gives teams operational evidence instead of a generic sentiment score.
Q: What are the methodological limits of this study?
A: This is a qualitative interview study in Pakistani higher education, focused on Education and English departments. Its value lies in explaining mechanisms rather than estimating how common a pattern is across the sector. UK universities should use it as a strong interpretive prompt, then test the same issues in local surveys, AI consultations, and module comments before generalising too far.
Q: What does this change about student voice on AI more broadly?
A: It shifts the focus from tool use to lived governance. Student voice on AI should not only ask whether students use GenAI or approve of it in principle. It should ask whether local rules feel clear, fair, and consistent enough to act on. That makes open comments especially useful, because they show where policy still depends on improvisation.
References
[Paper Source]: Ayesha Afzal, Shahid Rafiq and Martin Oliver "Moral improvisation with generative AI in higher education: faculty and student experiences in Pakistan" DOI: 10.1080/13562517.2026.2695805
Request a walkthrough
Book a free Student Voice Analytics demo
See all-comment coverage, sector benchmarks, and reporting designed for OfS quality and NSS requirements.
- All-comment coverage with HE-tuned taxonomy and sentiment.
- Versioned outputs with TEF-ready reporting.
- Benchmarks and BI-ready exports for boards and Senate.
UK-hosted · No public LLM APIs · Same-day turnaround