Automatic question generation (AQG): A systematic review of methods, applications, challenges and implications in educational technology

Abstract

The rapid transition to digital education has intensified the demand for personalised evaluation, yet the manual creation of high-quality assessment items remains critically resource-intensive. To address this challenge, this systematic review examined automatic question generation (AQG) methods, applications, challenges and implications in educational technology. A total of 127 peer-reviewed studies published between 2018 and 2025 were analysed. AQG approaches are classified into template-based, rule-based, machine learning and hybrid methods. While template and rule-based systems offer stability and interpretability, they face scalability and rigidity constraints. Conversely, machine learning-based techniques, particularly transformer architectures, demonstrate superior adaptability, though performance remains hindered by data scarcity. Emerging hybrid models show promise in balancing precision with flexibility. Key challenges identified include multilingual corpus availability, the generation of complex cognitive questions and ethical concerns regarding algorithmic transparency. We recommend embedding AQG within adaptive platforms to enhance instructional efficiency. Future research should prioritise expansive multilingual data sets, automated rule formulation and robust ethical frameworks to guide responsible deployment.

Implications for practice or policy:

• Educational technology leaders should embed AQG into adaptive systems to enable real-time feedback
• Teacher training must include AQG competencies, focusing on question validation.
• Higher education institutions should pilot AQG within learning management systems to automate assessments.
• Policymakers must establish regulatory frameworks for data privacy and algorithmic transparency.
• Researchers should conduct longitudinal studies to provide empirical evidence on learning outcomes.