In this study, we explored the effective design of prompts using generative AI in the educational field for evaluating reports. The experiment confirmed that prompts that included “presenting specific examples” and “setting specific goals” improved the clarity and persuasiveness of reports, and were more likely to receive high scores in automated grading. On the other hand, vague instructions and broad theme settings tended to result in scattered content and low scores. It was clear that detailed instructions were essential for eliciting appropriate information. In addition, it was suggested that the consistency and accuracy of AI scoring could be improved by incorporating evaluation guidelines that reflect the scoring criteria into the prompts for automatic scoring. This research provides guidelines for the effective use of generative AI in educational settings, and contributes to reducing the burden on teachers and improving the learning effectiveness of students. In the future, it will be necessary to improve the accuracy of prompt design and automatic scoring models through comparisons with human scoring and verification using different themes.
