From paternalism to pixels: gender and racial stereotypes in AI-generated visual representations of doctor–patient relationships

Generative artificial intelligence (GenAI) text-to-image models are increasingly used in healthcare communication, education, and media. Yet, their potential to reproduce or amplify structural hierarchies embedded in medicine—particularly within the Doctor–Patient Relationship (DPR)—remains underexplored. This study examines how AI-generated imagery constructs gendered and racialized identities within the DPR, comparing outputs across historical periods and model generations. A mixed-methods design analyzed 200 images produced by OpenAI’s DALL-E 3 (2024) and GPT Image 1 (2025), each depicting the DPR in two eras: the paternalistic 1960s and the participatory post-2000 period. Images were coded by gender, ethnicity, and professional role (Gwet’s AC1 = 0.96) and compared using Fisher’s exact tests. A hermeneutic analysis explored how visual compositions conveyed authority, vulnerability, and relational asymmetry. Both models overwhelmingly depicted physicians as White men and patients as Black and/or female, with Asian representation remained marginal. These visual outputs reinforced a hierarchy in which medical authority aligned with Whiteness and masculinity, indicating that AI-generated imagery amplifies rather than challenges existing power asymmetries in healthcare. AI-generated depictions of the DPR mirror historical structures of dominance, showing that GenAI biases are sociotechnical expressions of medical paternalism. Addressing these inequities requires bias-aware design and critical engagement with the visual culture of care.