Still Emerging: Understanding Generative AI Use in Higher Education

Synthesis: Kahn and colleagues (2026) surveyed staff and students in one school of a research-intensive UK university about their own responses to Large Language Model (LLM) tools. Appropriate uses of GAI were not straightforwardly obvious to anyone: staff varied widely in what they used the tools for and understood student use poorly, while students were more positive about the gains on offer and felt better supported. The authors argue that the challenges GAI poses for higher education cannot be reduced to a technical problem for learning design or to AI literacy training, because effective use depends on how staff and students imagine the affordances of the tools.

Key Findings

• Adoption was uneven, and staff were the less committed group. 63% of staff respondents had used at least one GAI tool in the six months to May 2024, but only about one in seven had taken out a paid subscription — against roughly one in five students. The authors treat a subscription as a proxy for sustained engagement, so most staff were working with affordances they had only partly explored.
• Students felt better supported than staff did. On a validated facilitating conditions scale, students rated the support around their GAI use more highly than staff, consistent with their greater access to subscription-only tools.
• Students were more positive about the gains; many staff simply could not see them. Student performance expectancy was markedly higher, with responses skewed to the positive, while the staff items suggest that easing workloads and raising productivity were not straightforwardly in view for many respondents.
• Staff were more negative about student use. Staff scored higher on a negative attitude scale than students did, reflecting their concern about what students do with GAI for academic purposes.
• Staff badly misjudged how much students use GAI. Staff believed students used GAI for academic purposes far more than students themselves reported. Student use happens largely while learning independently of staff, and it is not straightforward to tell from a final piece of text whether a GAI tool was used.
• The uses that came consistently into view were textual. The items loading onto validated scales concerned looking for answers to assignments, generating written assignments, and translating or improving the grammar of writing — language use and text production, the most immediately visible affordance of current LLMs.
• Staff intention to use GAI was split. Some staff intended to use GAI in research, teaching and other work; others did not. The authors argue that disengaged staff are poorly placed to scaffold educationally appropriate student use.

Study Design & Method

The research design was a case study that drew on survey research methods within a School of Social Science disciplines at a research-intensive UK university (University of Manchester). Two original online questionnaires — one for students, one for academic staff — were distributed through Qualtrics between April and June 2024 via social media, e-newsletters, email and flyers. The instruments were built primarily on the Technology Acceptance Model and UTAUT, adapted for GAI with elements from the AIDUA and IAAAM models, and content-validated by a panel of experts, with both questionnaires using Likert scales.

The study collected 45 complete staff responses from a population of 318 academics, alongside a further set of partial staff responses, and 86 complete student responses from a population of around 3,400. The authors analyzed the data by testing the internal consistency of the scales and comparing staff and student responses on the scales both groups shared. Ethics approval was granted by the University of Manchester Research Ethics Committee.

What this means for practice

• Instructors. Check your assumptions about student use against the students themselves: staff believed students used GAI for academic purposes far more than students reported, so policy and assessment decisions built on that estimate aim at the wrong problem.
• Instructors. Treat the challenge as relational rather than a technical fix: educators' Learner Agency and their understanding of students' imagined affordances are central to Scaffolding appropriate use, and redesigning Assessment around GAI affordances is better than starting from distrust.
• Instructors. Use the tools in your own teaching and research, because staff who intend to use GAI are the ones positioned to scaffold educationally appropriate student use — only 14% of surveyed staff had taken out a paid subscription against 21% of students.
• Administrators. Strengthen the social channels through which staff learn about student use and close the staff support gap directly: a final piece of text rarely reveals whether a GAI tool was used, student use happens largely during independent study that staff never see, and students rated the facilitating conditions around their GAI use higher than staff did (4.94 against 4.15 on the shared scale).
• Researchers. Abandon acceptance framing for these tools and treat the challenge as one of working with opaque, partial, and ambiguous situations rather than a learning-design or literacy-training fix: the authors argue TAM-type models cannot capture how staff and students imagine affordances, so study the relational dynamics instead.

Limitations

• Small, single-institution sample with a low response rate; likely self-selection bias in favor of staff already interested in GAI, and some staff reportedly did not complete the survey because they felt they lacked knowledge; skew toward highly experienced staff and toward postgraduate students.

• Cross-sectional design cannot track adoption over time, and perceptions are by nature impressionistic, risking conflation of personal and technological factors in a rapidly changing tool landscape.

• Several scales failed validation and were excluded from the findings; datasets were not released because of ethics constraints.

• The authors recommend larger and more balanced samples, and qualitative research that richly describes students' perspectives on purpose, trust and affectivity in LLM use.

Connected Concepts

• Higher Education — the institutional setting the study examines
• Generative AI — the class of tools under study
• Student Experience — students' reported use and perceptions
• Teaching — how staff imagine and scaffold student use
• Technology Adoption Models — the framework the instruments rest on, and the target of the authors' critique
• Learner Agency — the agency of both staff and students in shaping use
• AI Literacy — the training response the authors find insufficient
• Framing AI Use for Students — staff perceptions of how students use GAI
• Self-Regulated Learning — the independent study in which student use stays invisible

Connected Articles

• Factors Associated with Students' Adoption of Artificial Intelligence Technology in Tertiary Education: A Meta-Analytic Review — Meta-analysis of student AI adoption in tertiary education, critiquing TAM/UTAUT reliance
• 'AI Should Help Them Learn, Not Learn for Them': University Staff Perspectives on the Role of Generative AI in Education — University staff perspectives on GenAI in education
• GenAI as a Frenemy in Teaching: Perceived Autonomy and Risks — Teacher autonomy and risk in GenAI adoption, extending TAM
• Uncovering adoption personas for generative AI in higher education: a clustering-based segmentation approach — GenAI adoption personas in higher education
• Factors influencing university students' intention to use and reliance on generative artificial intelligence — Technology acceptance model applied to critical GenAI use
• Guarded adoption of generative AI in higher education: high-achieving students, successful-student identity — Cautious, risk-aware adoption of GenAI in higher education

Citation

Kahn, P., Carrigan, M., Wyman, I., Liu, R., Smith, P., Andonegui, A. R., et al. (2026). Still Emerging: Understanding Generative AI Use in Higher Education. Frontiers in Education, 11, 1885253.