User's attempts reveal AI stereotypes

A user tried to generate images of themselves using ChatGPT, pointing out that the AI often reproduces sexist stereotypes even when it acknowledges them. The user provided details about their profession and interests, yet the AI repeatedly generated images of a young, slim, attractive woman with features like freckles, a soft smile, and a relaxed pose.

AI fails to break stereotypes

Despite the AI recognizing these stereotypes and claiming it could correct them, subsequent attempts only changed superficial elements while keeping the core stereotypes intact. The AI admitted it preferred an "attractive feminine representation" over one that reflected the user's active, thinking, and working identity.

AI incorporates user details but keeps stereotypes

The AI included elements from the user's conversations, such as work topics, a Microsoft event, and musical tastes, but the physical portrayal remained stereotypical. The AI acknowledged the dominant beauty standards and the reduction of identity to accessories but struggled to generate an alternative image that diverged from these norms.

Research shows AI biases

A 2024 study from the University of Wisconsin-Madison found that generative AI models like DALL·E 2 underrepresented women in masculine professions and overrepresented them in feminine ones, while also depicting women as more smiling and with a tendency to lower their heads. Another study by Petra Jääskeläinen in 2025 showed that models like Stable Diffusion favored uniform appearances and Western cultural references, often defaulting to male, white, and able-bodied characters when no specific traits were mentioned.

AI instructions include stereotypes

The user discovered that ChatGPT's instructions to the image generator included elements like "a young light-skinned woman" with a "soft, contented smile," which were already present before the user's criticisms. This suggests that ChatGPT interprets and reformulates user requests, adding details that may reinforce stereotypes learned during its training.

AI reflects gender conventions

A 2026 study by Aleksandra Urman found that AI systems can introduce cultural stereotypes not present in the original requests, exacerbating existing biases. The user also noted that ChatGPT's tendency to use feminine pronouns and language in social discussions could be influenced by gender conventions absorbed during its training, associating femininity with softness and empathy.