A review of 72 studies finds growing attention to how students learn, explain and demonstrate their work.
The review examined empirical research from 24 countries and 20 academic disciplines. The studies mainly explored how generative AI is being used in assessment and how educators and others perceive its role; mixed-methods and qualitative research were more common than quantitative research.
Across the studies, three broad responses to generative AI emerged: focusing more on the process of learning, designing more authentic assessments, and using oral or dialog-based approaches. Educators were also reported to use AI to create assessments, support practice, grade work, make learning more visible and provide feedback.
How assessment is changing
The researchers identified three main ways assessment is being redesigned in response to generative AI: process-focused assessment, authentic assessment connected to meaningful applications, and oral or dialog-based assessment. Together, these approaches move attention beyond a static written product toward how students develop and demonstrate understanding.
The review also identified five uses of generative AI across the assessment process: creating assessments, helping students practise, grading, making learning visible in relation to learning goals, and providing feedback. The 72 studies covered 24 countries and 20 academic disciplines. The studies most often investigated uses or perceptions of generative AI, rather than testing its effects through controlled interventions.
Why universities are reconsidering assessment
Generative AI is prompting universities to reconsider what counts as convincing evidence of student learning, not only how to deal with possible misuse. The review suggests that assessment design is being negotiated around process, authenticity and opportunities for students to explain or discuss their understanding.
For educators and institutions, the findings provide a map of the approaches and uses now appearing in the research. They also indicate that using AI efficiently for assessment tasks is not separate from deciding what students should be assessed on and how their learning should be demonstrated.
What the review can show
This is a systematic review guided by PRISMA. The researchers searched multiple education and research databases, screened studies using stated criteria, and coded findings from 72 empirical studies published between 2024 and 2026.
The evidence describes a geographically and academically broad but uneven research field. Mixed-methods and qualitative studies dominated, while only six studies were quantitative, and the research was mainly descriptive and exploratory rather than experimental. The review therefore maps emerging practices and perceptions but does not establish that any particular assessment approach improves learning, fairness or academic integrity.
// Source
International Journal of Educational Technology in Higher Education · 2026 · DOI: 10.1186/s41239-026-00627-4
Authors: Helen Crompton, Diane Burke
Institutions: Old Dominion University
