Algorithmic minds, human bias: How AI-generated images shape mental health stigma

Mental health stigma is shaped by complex social and representational processes increasingly influenced by digital and algorithmic environments. Despite the rapid diffusion of generative artificial intelligence (AI), little is known about how AI-generated visual representations may affect stigma-related responses. Because generative AI systems may reproduce and amplify stereotypes embedded in their training data and in the prompts used to guide image generation, they represent an increasingly relevant channel through which visual representations of mental illness are produced and disseminated. This study examined whether AI-generated images of mental disorders influence social distance depending on the use of stereotype-based prompts. Using a within-subject experimental design, 311 participants evaluated images depicting depression, bipolar disorder, and schizophrenia generated with or without explicitly stereotypical prompts. Overall, stereotype-based prompts were associated with slightly lower social distance compared to the no-prompt condition (d = 0.44). However, this effect varied significantly across diagnostic categories. No differences emerged for depression, whereas bipolar disorder showed a large decrease in social distance in the prompt condition (d = 1.18). In contrast, schizophrenia showed the expected pattern, with higher social distance following stereotype-based prompts (d = -0.64). Mental health knowledge was associated with lower stigma overall but predicted greater sensitivity to stereotype-based representations in schizophrenia. Prior exposure to mental health-related experiences was linked to lower stigma but did not moderate experimental effects. These findings suggest that AI-generated representations may differentially shape stigma across disorders, highlighting the need to consider algorithmic environments in stigma research and in the development of digital communication and anti-stigma interventions.