From shortcut to study partner: Student approaches to GenAI

During the many AI training sessions that we run at EDLAB, we meet teaching staff with very different views. Some want students to develop their skills independently without turning to Generative AI (GenAI). Others want students to use these tools more thoughtfully, as a scaffold to support learning rather than as a crutch. What most colleagues share, however, is a concern that students may be relying on GenAI too much, while lacking a clear sense of why students turn to it or what drives less desirable forms of use. 

To give teaching staff more concrete guidance, I wanted to better understand what motivates students’ unsupervised use of GenAI. Last year, I conducted a short survey. The 199 responses from UM students offered a surprisingly honest picture of what happens when students turn to GenAI without guidance and why some end up outsourcing their work to these tools in ways that are not beneficial for learning. 

When I looked at the survey responses, the first thing that stood out was that almost half of the respondents admitted to outsourcing their academic work to GenAI at times, with “little to no thought”. Students gave concrete examples, such as generating answers to tutorial learning goals or asking GenAI to summarise readings in preparation for tutorials. Their main reasons for delegating work to these tools were a lack of time, a lack of perceived value in the task, or just not understanding the material or knowing the answers.   

What was striking was that a third of students who admitted to outsourcing their work to GenAI in this way, also believed it constituted cheating.  This illustrates in some ways the rationalisation element of the fraud triangle: students know it is cheating, but they still justify it to themselves. At the same time, reasons such as time pressure, limited understanding of the material or a perceived lack of value in the task, also shows how quickly academic pressure (another pillar of the fraud triangle) can push students toward overly convenient use of GenAI. In contrast, over 80% of those who said they did not use GenAI in this way considered it cheating and avoided delegating their work to GenAI mainly due to fears of detection, concerns about the limitations of GenAI, a sense that it was wrong or a desire to learn and achieve the outcome for themselves. 

Worryingly, very few respondents expressed concern about the accuracy or reliability of GenAI output. Several appeared to treat GenAI as interchangeable with a search engine, not recognising how differently and often unreliably they operate. This gap highlights the ongoing need to explicitly teach AI literacy and critical thinking, rather than assuming students already have the skills to assess AI-generated information. 

These insights suggest that students occupy a spectrum of relationships with AI, ranging from self-reliant learners to full outsourcers, with most falling somewhere in between, depending on the context. To reduce over-reliance on GenAI, the three components of the fraud triangle provide a helpful framework for designing interventions, for example: 

Rationalisation – what matters most leads the way 

Students are most likely to outsource work to GenAI when a task feels meaningless. When students understand why a task matters, they are far less likely to bypass it with AI.  This can be addressed by: 

• Explaining the purpose of tasks and how they support key learning outcomes or relate to skills students will need in the future. This means ensuring courses are constructively aligned and relevant to real world contexts. Consider your students’ perspectives and ask yourself why the related skills matter to them. 

• Being clear about when and how AI can be used and talking openly with students about effective and ethical AI use, distinguishing collaboration from delegation. Discuss what it means for your students to know or be able to do something. Being explicit about your expectations reduces the temptation for students to interpret the rules for themselves. 

• Discussing academic integrity as a matter of professional identity and a way of acknowledging their effort and creativity. 

• Embedding reflection on GenAI use into assignments to increase awareness of its potential, its limitations and shortcomings. 

• Creating space in assignments for personal creativity and expression which makes GenAI less useful as students still need to think, create and engage using their own perspectives and interests.  This can also make assignments more appealing. Fun is an important but often overlooked part of learning. 

Pressure – so much to do, so little time… 

Many students turn to GenAI because they feel time-pressured and unsure how to begin, do not fully understand the task (or material) or feel unable to complete the assignment.  This pressure can be reduced by: 

• Scaffolding complex assignments into different phases, such as outlines, drafts and revisions. 

• Creating opportunities for early feedback and being available for support and follow-up questions to prevent students from feeling stranded, or showing students how to use AI effectively for these purposes. 

• Checking workload and deadline clustering across courses and providing realistic deadlines. 

• Providing worked examples or clear starting points to help students build initial confidence. 

• Supporting the development of students’ self-regulated study skills so that they can better manage their work (and/or guiding students on how to use AI effectively for these purposes). 

Opportunity – makes the thief, or sparks creativity 

The opportunity to outsource work to GenAI is strongly shaped by assessment design. By varying assessment types across a course or curriculum, teachers can reduce the likelihood that learning can be fully delegated to AI. The Digital Education Council³distinguishes between three types of assignments: AI-free, AI-assisted (where tasks clearly define when and how AI may be used to support learning) and AI-integrated assessments (where AI-literacy is a central part of the learning and assessment process).  Within a course, teaching staff could therefore consider a mix of the following strategies: 

• Using controlled conditions when appropriate, especially for foundational skills that require independent mastery. 

• Assessing specific in-class components, such as presentations, discussions, debates, teamwork or other in-class assignments. 

• Revisiting learning outcomes to ensure they capture higher-order thinking, creativity and critical judgment competences rather than simple content reproduction and remain relevant in an AI-driven world. 

• Focusing on the process, not (just) the product, such as drafts, reasoning steps, logbooks, or reflections on teamwork, AI-use or the learning process. 

• Designing personalised or applied tasks that require creativity and reflection on the part of the learner and are harder to outsource to GenAI, such as personal reflections, work based on data students collect themselves or tasks that students have to carry out in the “real-world”. 

• For AI-assisted and AI-integrated tasks, requiring reflection and transparency by asking students to record, submit, reflect on and critically evaluate their use of GenAI and what they learned from it.  Offer training where necessary. 

Understanding how our students tend to use GenAI offers valuable insight into what drives them, how they think and learn and how they navigate academic pressure. As educators, we can build on these insights to design learning experiences that strengthen students’ autonomy, while embracing the potential of AI as a learning partner. I hope the practical tips that I’ve gleaned from this study (also summarised in this poster on page 5 of the Use Cases: Use of Large Language Models in Problem-Based Learning document) as well as the link I’ve made between student perspectives and the academic fraud model are useful in supporting a more considered approach to course and assessment design. In fact, these thoughts and observations about the use of AI strongly reflect the UM vision on assessment, which continues to be as relevant today as when it was first published in 2020. 

The most important takeaway is that students’ over-reliance on GenAI is not fundamentally different from more traditional forms of academic fraud, such as copying from course mates or googling the answers. Students who are determined to cheat will usually find a way, with or without new technologies. The real issue, therefore, is not the tool itself, but the conditions that make students feel willing or unwilling, able or unable to invest time and effort in a task. 

This is where course and assessment design become critical. As GenAI increases the opportunities to shortcut learning, the most effective response is not tighter control, but more thoughtful task design. For me, this means designing for curiosity, creativity and personal expression, by creating assignments that are intrinsically engaging and relevant, leave room for students’ own interests, perspectives and offer choice and customisation. When assignments reward understanding, reflection, process and individuality, GenAI becomes less attractive as a shortcut and learning becomes more meaningful. We shouldn’t forget that both fun and perceived value, though often overlooked, play an essential role in sustaining genuine student engagement.   

By Donna Carroll, Senior Coordinator of Professional Development for Teaching & Learning, EDLAB, Maastricht University.
 

References

[1] Cressey, D.R. (1953) Other people’s money: A study in the social psychology of embezzlement, Glencoe, IL: Free Press. 

[2] Hamid, N.A., Dangi, M.R.M, Sabli, N., Adnan, M.F., Wahab, R.A., (2017). Academic Cheating and Fraud Triangle Theory: Undergraduate Students’ Perspectives. Advanced Science Letters, 23(11), 10577-10581(5); https://doi.org/10.1166/asl.2017.10106 

[3] Digital Education Council (DEC), The Next Era of Assessment, 2025.