AI-designed versus pre-service teacher-designed activities: Analysis and evaluation

Abstract

This study employed a comparative quantitative design to examine the quality of grammar teaching activities generated by three AI models - ChatGPT, DeepSeek, and Gemini - and by pre-service teachers. Data were collected using the Grammar Activity Assessment Form (GAAF) and evaluated by 10 field experts. Findings indicated that AI models consistently produced high or very high-quality activities, with DeepSeek achieving the highest proportion of top-tier outputs. While inferential analyses revealed no statistically significant differences in overall quality among the groups, pre-service teacher outputs displayed greater variability. These findings highlight the pedagogical viability of AI-generated activities as structured scaffolds; however, they also underscore the need for teacher educators to balance AI efficiency with the preservation of the foundational teaching craft. The study concludes that AI should be positioned as a co-designer that requires active human intervention to ensure context-specific relevance and pedagogical creativity.