Manuscript received April 1, 2026; accepted April 16, 2026; published May 19, 2026.

Abstract—Due to advancements in generative Artificial Intelligence (AI), there is growing interest in using Large Language Models as conversational models in language learning. Although previous studies have shown the effectiveness of AI chatbots in accelerating second-language acquisition, most prior research focuses on English and other standard language varieties, with little attention given to Arabic dialects. This work-in-progress paper proposes the design of a custom Generative Pre-trained Transformer (GPT)-based learning system aimed at vocabulary acquisition and learner satisfaction in the Kuwaiti Arabic dialect, along with a proposed framework for evaluating the learning system. The custom GPT learning system will focus on conversational tasks centered on vocabulary learning, based on dialect-specific conversational tasks, and will provide interactive and context-specific vocabulary practice using the native language. The study adopts a design-based research approach that integrates instructional design principles with iterative system development, with the use of a mixed-methods evaluation framework. The study will be conducted on adult non-native women based in Kuwait who are learning Kuwaiti Arabic. Data collection will include vocabulary assessments, learner satisfaction surveys, semi-structured interviews, and platform usage analytics. The research aims to analyze and evaluate learning impacts, along with students’ perceptions of learning environments that utilize artificial intelligence. By addressing the underrepresentation of Arabic dialects in AI-assisted language learning studies, this study aims to provide design insights and guidelines for GAI language learning that is culturally responsive.