<p>Although artificial intelligence (AI) is transforming language education, its application in early childhood oral language curricula remains&#xa0;underexplored. This design-based research developed and evaluated a teacher-mediated, AI-enhanced oral language curriculum and blended professional development programme for kindergarten teachers in Yogyakarta, Indonesia. Six play-based modules integrate Google Gemini for real-time story generation and NotebookLM for preparing structured materials; all AI interaction is teacher-mediated, with children engaging through guided play rather than direct screen use. A mixed-methods design involved 94 teachers and 186 children aged 4 to 6 across six schools over 16&#xa0;weeks. Following intervention, teachers demonstrated significant self-efficacy gains (<i>d</i> = 0.87), implementation fidelity averaged 81.3%, and children showed significant oral language improvement (<i>d</i> = 0.46),&#xa0;with notable gains in narrative production (<i>d</i> = 0.43) and vocabulary (<i>d</i> = 0.38). These findings provide&#xa0;empirical evidence&#xa0;for a teacher-mediated AI curriculum in early childhood oral language education, offering a replicable, developmentally appropriate model for integrating generative AI into play-based curricula without direct child–screen interaction.</p>