The increasing adoption of generative AI raises concerns regarding its potential to reinforce gender stereotypes, especially in content aimed at children. AI-generated images that reflect traditional gender roles can inadvertently influence children’s perceptions, embedding unconscious biases. Through co-design sessions with domain experts, we developed a child-friendly generative AI web application designed with dual purpose: (1) analysing children’s choices to determine potential underlying bias and whether children’s own gender identity influences these choices, and (2) investigating if specific non-aesthetic traits prompt the AI to produce characters with stereotypical presentations.