Stereotypes at the intersection of perceivers, situations, and identities: analyzing stereotypes from storytelling using natural language processing

Stereotypes in the real world are complex, varying by perceivers, situations, and intersecting identities. These modifying factors are often overlooked in the literature. We address this gap by having participants freely write stories about how people view 9 different groups across 6 representative situations and of 30 intersectional identities. To identify stereotypes in these stories, we leverage large language model GPT to detect stereotypes. Across two pre-registered studies with U.S. representative participants (N = 1,721), we found that although stereotypes are considered products of social learning, they varied considerably across perceivers. While seen as overgeneralized associations about the groups, stereotypes significantly varied by situation. Multiple categorizations do not shape intersectional stereotypes equally, with Asian showing dominating effects likely due U.S. population’s lower exposure to this group than familiar identities (Black/White, women/men, old/young). These findings highlight how individual experiences, situation norms, and identity representativeness interact to shape stereotypes in naturalistic contexts.