When AI tells the story: Systematic stereotyping in AI-generated narratives

As Artificial Intelligence (AI) tools are increasingly deployed across digital platforms to generate content about people, questions arise about whether AI-generated narratives exhibit systematic representational biases. This study examines whether AI-generated narratives show structured framing differences across social role contexts and gender, and whether such patterns differ across three leading commercial AI tools. Using a controlled prompt experiment, we generated 469 narratives from ChatGPT (GPT-4o), Claude (Opus 4.6), and Gemini (2.5 Pro), varying social role context (professional vs. domestic) and character gender (man vs. woman). Each narrative was scored using the warmth and competence dimensions derived from the Stereotype Content Model. The results reveal three notable patterns. First, the AI models told a different story depending on the context in which each character was placed: professional narratives were consistently high in competence and low in warmth, while domestic narratives received the reverse. Second, gender shaped this pattern specifically in professional contexts, where female characters received notably warmer portrayals than male characters despite similarly high levels of competence. Third, although all three models exhibited similar framing patterns, they differed in intensity. ChatGPT generated the warmest narratives overall and assigned higher competence portrayals to domestic characters than either Claude or Gemini, whereas Gemini generated the coldest portrayals overall. These findings suggest that AI-generated narratives encode structured stereotype-consistent framing, that gender and social role context jointly shape those portrayals, and that AI tool choice carries representational consequences that are rarely considered in practice.