What are the dominant narratives of generative AI, and what is at stake in their circulation? In this keynote, Jill Walker Rettberg discusses her ongoing research on AI-generated narratives in the AI STORIES project, which starts from the hypothesis that LLMs replicate and perhaps increase certain narrative patterns, which could mean that we lose diversity in storytelling. Research so far suggests this is true – the thousands of AI-generated stories we have analysed in the AI STORIES project emphasise stability and nostalgia, telling remarkably similar stories of threatened communities saved by reconnecting with heritage. The theme of DHA2025, Digital Archipelagos, reminds us both of our diversity and our interconnectedness – but can we retain these when using large language models (LLMs)? Is it possible to use large language models (LLMs) without succumbing to the digital colonialism of the large tech companies that sell them to us? How should we, as researchers and educators, respond to political and institutional pushes to use genAI? What does it mean for our digital archipelagos that Trump has issued an executive order banning “woke AI” and an AI Action Plan to ensure US allies use the “full AI technology stack” that aligns with American values? Generative AI is normalising, erasing the outliers and exceptions and replacing them with statistical probability. So would it help to use local models, or is the technology itself a problem? By understanding how LLMs really work we can gain the tools to decide when not to use it, and when it might add value. Rettberg will close by highlighting examples of how researchers might use LLMs with care, in ways that resist homogenisation and keep the archipelago alive.