AI systems, such as ChatGPT, often face challenges when using language in ways that fit social and cultural contexts, especially when making requests. While these models are strong in grammar and meaning, they frequently face challenges in capturing social and cultural aspects, leading to misunderstandings. To explore this issue, two frameworks were used: Taguchi’s Pragmatic Appropriateness Model and Brown and Levinson's Politeness Theory. A mixed-methods approach compared AI and human responses to specific scenarios testing power, familiarity, and obligation. The findings reveal common problems, such as AI being overly formal in casual situations, misusing honorifics, mixing dialects, and misunderstanding context. These issues highlight the need for AI to better adapt to social and cultural differences, particularly in diverse environments. Integrating linguistic theories into AI training can enhance its ability to comprehend context and establish trust with users. This research stated that AI struggles to adapt to social norms, especially in situations where making requests requires accuracy. Most concerns involve being too formal, using honorifics incorrectly, mixing dialects in unnatural ways, and misunderstanding the context. These results highlight the need to include sociolinguistic principles in AI training to improve its understanding of culture and context. Furthermore, the results of the current study can help AI developers and policymakers in the MENA region. Highlights The paper identifies specific pragmatic failures encountered by AI systems when processing Arabic, highlighting issues such as misinterpretation of context, cultural nuances, and idiomatic expressions that affect communication effectiveness. Through a detailed analysis of various communication scenarios, the study demonstrates how AI's inability to grasp social cues and contextual subtleties leads to misunderstandings, thereby impacting user experience and trust in AI applications. The paper offers actionable recommendations for enhancing AI's performance in Arabic, including the integrating culturally relevant training datasets, improving natural language processing algorithms, and emphasizing the importance of human-in-the-loop systems to mitigate pragmatic errors.