Generative AI use in higher education is often discussed as though lecturers can see what students are doing with these tools. In most courses, they see the submitted work. The conversations students have with AI while reading, planning, or writing happen elsewhere.
A 2026 study by Peter Kahn and colleagues asks what staff and students at one UK university thought about AI use. Their answers reveal a gap in perception, though the survey cannot tell us exactly how students used the tools.
The view from the finished assignment
Imagine receiving an essay that reads more smoothly than a student’s earlier work. Was AI used to correct grammar, translate a passage, suggest an argument, or write the text? The finished essay rarely settles that question. Staff may form an impression from the work they see, while students may have quite different ideas about what counts as ordinary help.
That uncertainty matters when a university sets expectations for AI use. A rule about disclosure can specify what students should report. It cannot, by itself, tell an instructor what students believe the tool is useful for or where they draw the line between assistance and substitution. I have written elsewhere about acceptable AI use in assessment; this study adds the relationship between staff perceptions and student practices to that discussion.

What the study found about generative AI use in higher education
Kahn and colleagues (2026) surveyed academic staff and students in a social sciences school at a research-intensive university in the UK. The questionnaires asked about perceived benefits, conditions supporting use, attitudes, and perceptions of students’ academic uses of generative AI.
The researchers collected 45 complete staff responses, plus some partial responses, and 86 complete student responses. Reported staff use referred to the six months leading to May 2024, when the tools and institutional practices were still changing.
Student respondents were generally more positive about the potential benefits for study and felt more supported in using AI than staff did. Staff responses varied substantially, including in their intention to use AI themselves.
Staff also perceived students to be using AI for academic work to a greater extent than student respondents reported. The student uses that emerged most consistently involved language support and producing text, such as improving grammar, finding answers for assignments, and generating written work.
These are perceptions gathered through a survey at one point in time, not observations of students working with AI or measures of learning. The response rates were low, and students in taught postgraduate programmes made up most of the student respondents.
People interested in AI may have been more likely to participate. We therefore cannot treat these results as a reliable estimate of use across the university, much less across higher education. The staff and student questionnaires also used different items for some topics, so every apparent difference cannot be read as a direct comparison.
What this could mean for teaching
The authors use the idea of imagined affordances: the uses people believe a technology makes possible. A student who mainly sees AI as a way to improve wording will approach it differently from one who uses it to test an interpretation or question an explanation.
An instructor who has mostly encountered AI-generated essays may imagine students’ uses differently again. These beliefs affect the opportunities teachers offer and the decisions students make.
I would start by asking students about specific stages of a task. When do they turn to AI? What do they ask it to do? Which suggestions do they reject? What work do they want to keep for themselves?
These questions can be discussed without requiring students to disclose every private interaction or treating them as suspects. The aim is to understand the practices that an assignment alone leaves hidden.
Then I would make the learning purpose visible. If the goal is to develop an argument from sources, show students the difference between asking AI to generate that argument and asking it to challenge one they have already drafted.
If the goal is to communicate an idea clearly, discuss when language editing helps a student express their own thinking and when generated prose takes over the work being assessed. The appropriate boundary depends on the task and its intended learning, a concern that also runs through my review of assessment validity in the AI age.
Staff need room to try the tools as well. Kahn and colleagues point out that educators who have little practical experience with AI may struggle to guide students toward educationally useful applications.
Trying a tool on a real teaching task, inspecting where it fails, and talking with students about their uses would give a department a more grounded starting point than a policy written around a single imagined student behaviour.
For researchers, the next step is to get closer to actual practice. Interviews, task observations, and accounts of how students decide to accept or reject AI output could test the explanations suggested by this survey.
Studies that follow use across a semester would also show whether perceptions change as tools, assignments, and local norms change. The broader AI and teaching practice conversation needs this kind of detail if advice to educators is to reflect what happens while students work, not only what appears in their final submissions. Med Kharbach
Reference
Kahn, P., Carrigan, M., Wyman, I., Liu, R., Smith, P., Romero Andonegui, A., Murtagh, L., & Song, F. (2026). Still emerging: Understanding generative AI use in higher education. Frontiers in Education, 11, Article 1885253. https://doi.org/10.3389/feduc.2026.1885253




