P030 Optimizing Prompt Engineering for Arabic Text Generation: A Comparative Multi-Platform Study Utilizing PICCO, CLEAR, and PARTS Frameworks

Abstract

In the digital transformation era, many Arabic Language Education students directly copy texts from Artificial Intelligence (AI) for writing assignments (kitabah), unaware that AI frequently produces disorganized sentence structures when given sub-optimal prompts. Rather than replacing organic writing skills, AI should be optimized as an intelligent writing assistant through systematic prompt engineering. However, if this assistant provides flawed linguistic guidance, such a habit will ultimately impair students' authentic writing abilities and degrade the quality of Arabic language education on campus. Therefore, this study aims to determine the most optimal prompting framework among PICCO, CLEAR, and PARTS, while comparing the performance accuracy of ChatGPT and Google Gemini in generating high-quality Arabic texts. Adopting a qualitative approach, this study employs content analysis. The research objects comprise formal Arabic (fusha) text drafts generated by AI. The data collection tool utilized is the documented textual transcripts of prompt interactions structured under the guidelines of the PICCO, CLEAR, and PARTS frameworks. The collected data were then manually analyzed based on Nahwu-Sharf syntactic-morphological rules and semantic appropriateness, followed by validation through expert judgment. The findings indicate that the PICCO framework is the most optimal model for generating strict grammatical accuracy, followed by CLEAR for lexical variation, and PARTS for creative writing styles. Furthermore, ChatGPT proved to be the most compatible platform for executing PICCO-based prompts, whereas Google Gemini excelled in executing the CLEAR model. This study is limited to formal Arabic texts. Future researchers are encouraged to expand the scope to colloquial Arabic ('ammiyah) dialects.