自治体広報誌の文字組版における生成AI導入可能性の検証

抄録

This study examines the feasibility of adopting generative AI in the text typesetting workflow for municipal newsletters. Experiments using the Claude series revealed distinct model characteristics: Haiku demonstrated stability and structural consistency, while Sonnet showed strengths in processing long texts. Prompt design was found to play a significant role in output accuracy and reproducibility, and the combination of Haiku with explicit prompts was most suitable for meeting institutional requirements. Furthermore, automated layout experiments with InDesign confirmed a high placement success rate and short post‑editing time, indicating practical effectiveness as a semi‑automated approach despite remaining challenges to full automation. This research highlights a direction for improving typesetting workflows through the integration of generative AI and DTP processes, contributing to the intersection of printing engineering and information technology.