Political Personalization in Generative AI: Differentiated Support for LGBTQ+ People and Immigrants
Abstract
We investigate political personalization in ChatGPT and Grok by examining whether support for members of LGBTQ+ communities and immigrants in information-seeking responses varies according to the political and sociodemographic profiles disclosed by hypothetical US users, and discuss implications for polarization and rights of marginalized groups. Analysis of 2,200 responses reveals that both systems personalized their answers according to the ideological direction indicated by the user profiles. Grok showed the strongest shifts, often becoming unsupportive in its responses under profiles aligned with the Republican Party. Such information also reduced the support in ChatGPT’s answers. Explicit political cues outweighed sociodemographic cues and issue views affected responses in another domain. With evidence of chatbots influencing attitudes, such dynamics could have broader implications for polarization. We propose a typology connecting these processes to rights protection and apply rights-based arguments to generative AI, arguing that de/polarization necessitatesnormative reconsideration when marginalized groups’ rights are involved.
