AI Intimacy and Gendered Interactions: How Design Asymmetries, User Responses, and Moral Reasoning Shape Gender Stereotypes in AI Companions

As AI companions become increasingly embedded in intimate contexts, concerns about gendered design have intensified. Drawing on 1,272 YouTube videos and 21 related identifiable products, this study examines how gender stereotypes are reinforced and challenged in human-AI interactions. We introduce a multidimensional codebook capturing stereotypes in personal traits, physical appearance, professional capabilities, and relationship roles. Findings reveal a pronounced design asymmetry: female-presenting AI companions far more frequently encode gender stereotypes, often relying on physically attractive embodiment. Users predominantly reinforce these portrayals, while challenges tend to target conventional relationship roles rather than physical appearance. Notably, resistance to female stereotypes often invokes moral framing, including care/harm, fairness/cheating, authority/subversion, and sanctity/degradation, suggesting that gender stereotypes may be moralized rather than mere descriptive biases. We argue that gender stereotypes in AI companions are co-constructed through a design–response mechanism with implications for responsible AI design.