Generative Artificial Intelligence (GenAI) has been offering unprecedented opportunities for language education. However, its capacity to embrace linguistic diversity, particularly for learners of dialect-rich languages like Vietnamese and Mandarin, remains underexamined. Without careful consideration, GenAI risks reinforcing language hegemonies, thereby contributing to the recolonization of language learning landscapes by marginalizing minority dialects in favour of preferred standards. Adopting sociolinguistics interview, this study explores GenAI's (namely ChatGPT's) ability to recognize and generate dialect-specific content in discussing several pre-determined questions in both Vietnamese dialects (i.e., Northern, Southern, and Central) and Mandarin varieties (i.e., Mainland Standard Mandarin, Taiwanese Mandarin, and Singaporean Mandarin). A multi-stage role prompt, focusing on the topic of food, was used for both Vietnamese dialects and Mandarin varieties to generate responses. Our study reveals major inconsistencies in the representation of Vietnamese dialects and Chinese varieties within AI-generated output, raising critical questions about generative AI's role in perpetuating linguistic hierarchies. We conclude by emphasizing the need for tailored language learning approaches that leverage generative AI's capabilities to not only accommodate but also celebrate the rich tapestry of global dialects and languages, ensuring equitable access to language education for all learners.