Generative AI is now commonly used by students in UK higher education. Evidence from a national student survey suggests that reported AI use increased within a short period of time, with many respondents indicating that they used AI to support learning activities as well as assessed coursework (Freeman, 2025). This level of uptake matters because it complicates the assumption that AI use occurs only in marginal or exceptional situations.

Policy responses have developed quickly, but they have not yet resulted in a consistent shared approach. Within the UK sector, the Quality Assurance Agency has published guidance and resources that encourage institutions to safeguard academic standards while responding to generative AI (The Quality Assurance Agency [QAA], 2024). However, a global analysis of academic integrity policies indicates that references to AI-related risks vary considerably across institutions and often follow, rather than anticipate, technological developments (Perkin and Roe, 2024). As a result, a gap can emerge between the formal language of policy and how academic practices are experienced in everyday study.

Much of the evidence from the perspective of students concentrates on attitudes and perceptions rather than on decision-making in context. Some studies show that students often regard generative AI as helpful, while also expressing concerns about reliability, ethical implications, and learning outcomes (Chan and Hu, 2023). While this work provides important insight into general views, it offers a limited understanding of how students respond when they encounter concrete academic tasks that require judgment about acceptable use.

This uncertainty may be particularly pronounced for international students. Research on academic literacies suggests that expectations in higher education are frequently implicit, contested, and shaped by disciplinary norms rather than by a single transferable set of skills (Lea and Street, 1998). For students who move across educational systems, uncertainty should not be interpreted simply as an individual shortcoming. Instead, it reflects how academic conventions are communicated and evaluated. In this context, generative AI can function both as a form of assistance and as a potential source of risk. The same use of AI may be interpreted as legitimate learning support in one course and as inappropriate practice in another.

This study focuses on Chinese international students in UK higher education and asks how students exercise agency when deciding whether, when, and how to use generative AI in academic tasks. Agency is treated here as something enacted through practice, rather than as a fixed characteristic that students possess to varying degrees. This perspective draws on research into generative AI tools supported learning, which suggests that digital tools can both support and constrain agency depending on how they are taken up in practice (Darvishi et al., 2024). It is also informed by recent reviews of the emerging literature on generative AI and agency, which point to ongoing tensions around control, access, and changing understandings of what agency entails (Roe and Perkins, 2024).

The theoretical framework brings together academic literacies with sociomaterial and posthuman perspectives. Sociomaterial research emphasises that digital literacy practices emerge through relationships among learners, technologies, texts, and institutional contexts, rather than through individual capability alone (Gourlay and Oliver, 2013). Similarly, posthuman scholarship in digital education questions accounts of learning and authorship that locate responsibility solely within the individual, especially in digitally mediated environments (Bayne, 2016). While these perspectives offer valuable insights, they can remain abstract if they are not connected to concrete academic activity. For this reason, the study pays particular attention to moments during academic work when students pause, reconsider, accept, or reject AI input, as these moments offer a way to examine how agency and academic integrity are negotiated in practice.

The study adopts a design that links broad patterns of use with observed practices and participants’ own interpretations. The first stage consists of an online survey that explores how students use generative AI across a range of common academic tasks. It also gathers information about students’ perceptions of institutional guidance and their own understandings of acceptable boundaries. This stage provides an overview of patterns and variation, but it does not capture how decisions unfold during actual study activity. The second stage involves a writing task completed with screen recording. Participants carry out a short academic writing activity using their usual digital tools and practices, with the option to consult generative AI if they wish. This approach is adopted because reported perceptions tend to overlook hesitation and uncertainty. Research on writing processes suggests that screen recording can make visible aspects of decision making, such as pauses, revisions, and attention shifts that writers may not recall in retrospect (Chan, 2017). The resulting data are treated as evidence of practical reasoning rather than as an evaluation of writing ability. The final stage consists of interviews structured around specific moments from the recorded writing task. Participants are invited to reflect on selected episodes and explain what they were trying to achieve and how they understood the limits of acceptable AI use. Anchoring the discussion in concrete activity helps to move beyond general statements and allows closer examination of how institutional messages are interpreted in practice. It also enables exploration of how students relate AI use to learning aims, language support needs, and their perceptions of fairness. The study draws on academic literacies theory to situate integrity as negotiated rather than fixed categories. The design aims to connect students’ reported concerns with observed judgment practices and interpretations of institutional ambiguity.

As data collection is ongoing, this study does not claim to present findings at this stage. The discussion offered here is exploratory in nature and draws only on a small amount of preliminary material alongside relevant literature. It is intended to outline possible analytical directions rather than to support conclusions about students’ practices or decision-making. First, the study is expected to indicate that students’ judgements about the use of generative AI differ across tasks and depend on how clearly expectations are communicated. Although reported levels of use are relatively high, this does not appear to correspond to widely shared understandings of what counts as acceptable practice. Instead, existing research suggests that students often make judgments in response to specific situations rather than relying on consistent or clearly defined rules (Freeman et al., 2025). Second, the study is expected to suggest that uncertainty in institutional guidance plays a role in shaping students’ decisions. Where policies differ across modules or provide limited detail, students may turn to peer practices, previous experiences, or personal interpretations of academic integrity. This can result in a range of approaches to AI use rather than a single common pattern. Third, the research seeks to develop a more detailed understanding of how agency is expressed during academic work that involves AI tools. Instead of treating agency as a fixed personal quality, the study focuses on how decisions take shape through the interaction of tools, task requirements, institutional expectations, and students’ own goals. This view is consistent with academic literacies research, which understands writing and judgement as practices that are socially situated and shaped by context. These insights may contribute to ongoing discussions about generative AI, academic integrity, and student agency in higher education. They may also offer practical value for institutions seeking to communicate guidance more clearly.

Bayne, S. (2016). Posthumanism and research in digital education. SAGE Handbook of E-learning Research, 82-100. Chan, C. K. Y., & Hu, W. (2023). Students’ voices on generative AI: Perceptions, benefits, and challenges in higher education. International Journal of Educational Technology in Higher Education, 20(1), 43. Chan, S. (2017). Using keystroke logging to understand writers’ processes on a reading-into-writing test. Language Testing in Asia, 7(1), 10. Darvishi, A., Khosravi, H., Sadiq, S., Gašević, D., & Siemens, G. (2024). Impact of AI assistance on student agency. Computers & Education, 210, 104967. Freeman, J. (2025). Student generative ai survey 2025. Higher Education Policy Institute: London, UK. Gourlay, L., & Oliver, M. (2013). Beyond ‘the social': digital literacies as sociomaterial practice. In Literacy in the Digital University (pp. 79-94). Routledge. Lea, M. R., & Street, B. V. (1998). Student writing in higher education: An academic literacies approach. Studies in higher education, 23(2), 157-172. Perkins, M., & Roe, J. (2024). Decoding academic integrity policies: A corpus linguistics investigation of AI and other technological threats. Higher Education Policy, 37(3), 633-653. Roe, J., & Perkins, M. (2024). Generative AI and agency in Education: A critical scoping review and thematic analysis. arXiv preprint arXiv:2411.00631. The Quality Assurance Agency [QAA]. (2024). Advice and resources on Generative AI. https://www.qaa.ac.uk/sector-resources/generative-artificial-intelligence/qaa-advice-and-resources