Abstract
<jats:p>This study examines how large language models construct intercultural communication scenarios in English language teaching by analyzing AI-generated passages for their treatment of social roles, cultural diversity, and gendered pragmatic tendencies. Using a mixed-methods approach, short narratives were analyzed through computational techniques and qualitative interpretation. The findings show that LLM outputs reproduce hierarchical role patterns, with authority figures using directive modal verbs and service or peer roles relying on softer expressions. Cultural representation was uneven, with Japan and the United States appearing most often and several regions receiving limited attention. Gender differences also emerged, as males used direct and assertive language, while females displayed relational and emotionally focused discourse. These results highlight the need for critical AI literacy to ensure AI-generated ELT materials remain pedagogically sound and culturally inclusive.</jats:p>