This article presents a creative writing course incorporating AI, taught at Université Lumière Lyon 2 (France) in September–December 2024. The course combined exploration of LLMs with critical reflection on their use. It drew inspiration from Si Rome n’avait pas chuté by Raphaël Doan (2023), written with ChatGPT 3 and 3.5, which imagines a world where industrialisation began under Nero. Students used ChatGPT to write counterfactual stories or fan fictions: these literary genres are especially useful for creating a three-way relationship between the human writer, the LLM, and an external referent that drives the interaction between the two. The pedagogical progression was structured around three phases: idea generation, narrative structuration, and writing. I generated a prompt live in class: ‘What if Medea had killed Jason instead of her children?’ The students conceived extended chains of prompts to guide the model’s responses and refine narrative variations. Some engaged with contemporary novels, others reimagined Graeco-Roman myths, which will be the scope of this article. Mythological rewritings prove particularly rich for AI-assisted storytelling, revitalising courses on classical myths in a playful, dialectical, and interactive manner. They raise a wide range of ethical issues, especially regarding the sources from which LLMs draw and the content policies that lead to the suppression of certain aspects of ancient literature. Asking ChatGPT to rework classical myths also reveals its tendency to homogenise narratives. This highlights the ongoing need for human oversight: when approached critically, generative AI can provide an effective introduction to classical mythology.