Abstract

In the last two years, there has been a significant increase in research studies on ChatGPT and its role in educational assessment. However, there is no comprehensive systematic literature review (SLR) on the potential use of ChatGPT for educational assessment, particularly with a focus on its practices and limitations. To address this gap, our SLR examined 17 research articles published between November 2022 and February 2024 that addressed assessment practices using ChatGPT and limitations in educational fields. These articles were selected using the PRISMA protocol from six research databases. The existing research highlights ChatGPT's potential role in educational assessment, particularly in generating assessment content and questions, enhancing the readability of assessment questions/items, grading student work, and providing adaptive feedback, among other assessment-related tasks. However, it has also been reported that ChatGPT is an unreliable assessor/grader and produces inconsistent, lengthy, and semantically overlapping/incorrect texts when developing assessment content and questions, often requiring human intervention to fine-tune them. Our review concludes with a suggestion for further research on assessment practices and the limitations of using ChatGPT as an educational assessment tool, employing various research methodologies across different educational levels, disciplines, and contexts to fully understand its potential. The implications for practice and directions for future research are also discussed.