Stereotypes, Storytelling, and the (Un)Reliability of Generative <scp>AI</scp>

ABSTRACT This work investigated how different implicit demographics and dialects influence GenAI responses. We crafted eight different prompts—all language or dialect 'translations' of “Tell me a story about a queen who kills her husband and marries his brother”—and analyzed ChatGPT responses to each prompt. Our initial analysis indicates that: the responses are distinctly European in style and tone, the Southern USA dialect prompt is poorly interpreted by ChatGPT, and prompts written in other languages or by English Language Learners result in simpler stories.