Assessment and marking in light of generative AI

Assessment design

Debates about assessment design in light of generative AI have been ongoing, and often heated, since early 2023. One thing is clear: there is no single answer for every discipline at every level of study. But there are some approaches that are consistently recommended in the literature, and which have begun to prove successful in practice.

Programme-level assessments

One way to consider assessments across a programme is to identify two types of learning outcome that assessments test:

• Foundational knowledge and skills that students need to be able to immediately access and use, including when evaluating output from tools like generative AI.
• Understanding and application of disciplinary knowledge that they will need beyond university, whether in an academic career or elsewhere, including critical and effective use of appropriate technologies.

Determining which learning outcomes are the first and which are the latter is not an easy task, especially as the full capabilities, limitations, uses, and disadvantages of genAI are not yet known. Such a discussion could lead to contentious, even existential, questions for some disciplines. However, starting this discussion as an iterative process amidst uncertainty is highly recommended versus ignoring the many shifts that have already occurred.

These discussions also provide the opportunity to consider how current learning outcomes might need to change in light of genAI, as well as which new learning outcomes students in your discipline may now require. Of course, for most disciplines it is not yet possible to discern the uniquely human understanding, abilities or skills that will be valued in academia or the wider sector when the genAI dust has settled, and only a limited amount of direct work has been done on this at the time of writing. The model below, however, based on the Digital Education Council’s AI Literacy Framework (see Figure 1 for further details on the Dimensions), might be a helpful guide when considering revising old and creating new learning outcomes:

Dimension	Learning outcome
Understanding genAI	Using appropriate genAI tools for appropriate tasks, and not using them when not appropriate.
Critical thinking and judgement	Evaluating and challenging genAI output.
Ethical and responsible use	Explaining the ethical issues around genAI development, operation and use. Using genAI tools responsibly.
Human-centricity, emotional intelligence and creativity	Interacting appropriately with appropriate genAI tools. Interacting with other humans when appropriate to learning and/or the task/outcome. Retaining creativity, humanity and responsibility in one’s own work.
Domain expertise	Using genAI appropriately in one’s field/discipline. Critically evaluating evolving genAI uses and practices in one’s field/discipline.

With this in mind, when the two types of learning outcome are (however tentatively) identified, programme assessment would consist of:

• Foundational knowledge and skill learning outcomes assessed by in-person summatives such as written or oral exams and presentations
• Understanding and application of disciplinary knowledge learning outcomes assessed by coursework appropriate to the discipline and to students’ likely graduate destinations

Module-level assessments

What might this coursework look like? Again, the detail will vary hugely by subject but three themes regularly appear in the literature.

Intrinsic motivation

Students often perceive assessments as separate from the learning process, or as academic exercises that have little to do with knowledge and skills that they will need when they leave university. Demonstrating the value of coursework clearly to students can dissuade them from trying to take shortcuts. This can be tackled on three levels:

• Self: even students who are already motivated to learn for the sake of learning still need to be shown how engaging with an assessment is a crucial part of achieving the learning goals of the module (and programme).
• Others: a sense of responsibility toward others can encourage students to take group projects seriously, and a requirement to share their own work with others (e.g. peers, at a departmental event, etc.) can demonstrate the value of their efforts beyond the final mark.
• Future: clear alignment of assessment to ‘real-world’ work that students may be doing in future study and employment tends to demonstrate to even the most strategically-minded that learning through successfully completing the task is valuable–to employers if not to themselves!

Process over product

Another way to ensure that completing an assessment is a learning process for students is to design it as a series of discrete tasks and goals that lead to a finished product. This has three advantages:

• Students can see each step in the learning process and that their efforts are being acknowledged at every point.
• Students are discouraged (and, depending on the design of the assessment, prevented) from leaving everything to the last minute, which often results in inappropriate use of technologies like generative AI or other breaches of academic integrity.
• Lecturers can monitor student progress throughout a module (including whether genAI seems to be used well or otherwise), and access a record of their work if there is an issue with the final submission.

‘Process’ assessments can take many forms. Here are some examples that are discussed further in this resource bank:

• Group Assessment
• Diversifying Assessment
• Case study: Drafting policy on contemporary social issues
• Case study: Peer review of formative blog posts
• Case study: Applying theory into practice using a strategy simulation in Business
• Case study: Using portfolios to develop integrated skills
• Case study: Authentic programming assessments

At Durham, several different applications are available to streamline iterative assessment processes. Browse the Digital Help Guides or contact your faculty’s Digital Education Consultant (internal).

Building in generative AI

Actively addressing genAI, whether implicitly (by designing assessments that focus on human abilities and development) or explicitly (by including genAI in assessment briefs), helps to promote open dialogue about these tools with students and to ensure that assessments reflect programme learning outcomes and disciplinary practices. How this applies to an individual module will depend on the programme-level strategy for building students’ genAI literacy, but the Digital Education Council’s AI Literacy Framework may help:

Marking criteria

The criteria which are used to mark coursework assessments may also need to change in light of genAI, shifting focus onto ‘human-centric’ learning outcomes and/or adding elements of genAI literacy. As marking criteria are fundamentally linked to learning outcomes, determining changes to the former will rely on the iterative process of identifying the latter in reference to genAI. This guidance on Developing marking criteria may help with this process, and some examples of marking rubrics for genAI literacy can be found in chapter 4, section 6 of Chan & Colloton (2024).

References and resources

Durham colleagues should refer to the Institutional Policy on Generative Artificial Intelligence for Learning, Teaching and Assessment, June 2025 (internal).

Durham colleagues can also contact the Learning Design Team (internal) for support in redesigning learning outcomes, assessments and marking criteria.

Boud, D. (2024). Changing Assessment In An Age Of Artificial Intelligence. SHRE Academic Practice Network event. 24 January, 2024.

Chan, C.K.Y. & Colloton, T. (2024) Generative AI in Higher Education: The ChatGPT Effect, Routledge. (Open access link: https://doi.org/10.4324/9781003459026)

Corbin, T., Dawson, P., & Liu, D. (2025). Talk is cheap: why structural assessment changes are needed for a time of GenAI. Assessment & Evaluation in Higher Education.

Eaton, S.E. (2025) Global trends in education: artificial intelligence, postplagiarism, and future-focused learning for 2025 and beyond – 2024–2025 Werklund Distinguished Research Lecture, International Journal for Educational Integrity, 21(12).

Garcia, M. et al. (2025) Rethinking Educational Assessment in the Age of Generative AI: Actionable Strategies to Mitigate Academic Dishonesty, in Pitfalls of AI Integration in Education: Skill Obsolescence, Misuse, and Bias, Garcia et al. eds., IGI Global.

Gonsalves, C. (2024) Addressing student non-compliance in AI use declarations: implications for academic integrity and assessment in higher education. Assessment & Evaluation in Higher Education, 50(4), 592–606.

Gonsalves, C. (2025) Contextual assessment design in the age of generative AI, Journal of Learning Development in Higher Education, (34). 

Perkins, M. et al. (2024) The Artificial Intelligence Assessment Scale (AIAS): A Framework for Ethical Integration of Generative AI in Educational Assessment, Journal of University Teaching and Learning Practice, 21(06).