Abstract
<title>Abstract</title> <p>The integration of generative artificial intelligence into educational assessment presents both opportunities and challenges for teacher practice. While AI offers potential for automating time-consuming evaluation tasks, there remains a notable gap in operational frameworks that guide teachers through the complete assessment cycle—from question design to personalised feedback—while maintaining pedagogical integrity and human agency. This study addresses this gap by examining a practical instantiation of the Integrated AI Triad (IAT) model within a professional development programme involving 138 teachers across four cohorts. Analysing a dataset of 70 constructed-response questions, over 330 AI-simulated student responses, AI-driven error diagnoses, and AI-generated feedback messages, the study employed qualitative content analysis guided by formative assessment theory, feedback models, cognitive load theory, and self-regulated learning frameworks. Findings reveal three key patterns: (a) teachers predominantly designed scenario-based questions targeting higher-order thinking (analysis, application, evaluation = 80%), with science as the most frequent subject (45%); (b) six recurrent conceptual error types were identified—causal confusion, over-simplification, foundational misconceptions, failure to recognise causal relationships, narrowing of abstract concepts, and confusion of physical and chemical changes; and (c) AI-generated feedback consistently followed a three-part structure—positive opening, guiding question, encouraging closure—aligning with theoretical principles of effective formative feedback. Synthesising these findings, the study proposes a five-step operational framework for assessment automation and introduces the concept of pedagogical stimuli—AI-simulated responses that serve as catalysts for teacher reflection and error diagnosis. This framework preserves teacher agency while leveraging AI for efficiency, contributing both theoretical insights and practical guidance for AI-assisted assessment in K-12 education.</p>