AI Models Have Replicated Human Attractiveness Ratings
Large language models show consistent biases by prioritizing age and gender in facial aesthetics evaluations.
Updated on Oct. 3, 2026 in Artificial Intelligence
Live Poll
Do you trust AI models to provide objective evaluations of human traits?
A recent study published on Research Gate found that AI models, including Claude, Gemini, ChatGPT, and Grok, rate human faces with a consistent bias. These models evaluated 2,513 faces, demonstrating a tendency to mirror human-like preferences for youth and specific gender markers.
Why it matters
The findings suggest that AI models encode common human aesthetic biases, potentially impacting how these systems handle automated image filtering or sorting tasks. This alignment indicates that LLMs have internalized societal norms regarding physical appearance through their training data.
Researchers tested 2,513 human faces across four major models to determine rating consistency. The models consistently converged on narrow attractiveness ranges, favoring younger faces and women's features over men's, closely mirroring established human rank-ordering patterns.
The players
Claude
An AI model developed by Anthropic that uses constitutional AI to align performance with human preferences.
Gemini
A multimodal AI model suite developed by Google for integrated text, image, and data processing.
ChatGPT
An AI chatbot platform powered by OpenAI, widely used for reasoning and generative tasks.
Grok
A large language model developed by xAI designed for real-time data integration and conversational tasks.
The details
Researchers displayed 2,513 facial images to models, including Claude and ChatGPT, to measure how these systems interpret aesthetic appeal. The study reveals that AI models reproduce rank-orderings similar to human groups because they share common facial characteristics that inform attractiveness evaluations. By analyzing the output of these models, the researchers confirmed that the systems consistently applied biases toward youth and gender without needing explicit instructions to prioritize those traits.
Timeline
October 3, 2026: The research findings were published in Research Gate.
The Tech Race
This study situates AI aesthetic evaluation within the broader field of algorithmic bias research using the Research Gate repository. It follows a growing pattern of audits examining how generative models replicate human subjectivity in image classification tasks.
Users interacting with AI-driven photo management or content moderation tools should anticipate that these systems will mirror human-like biases toward youth and gendered aesthetics. These results demonstrate that subjective grading criteria are now embedded within the standard architectures of major commercial LLMs.
The takeaway
This research confirms that AI models are not neutral observers but rather mirrors that reflect the biases present in their training data. Users should monitor future model releases for updates in how these systems handle subjective visual evaluations versus objective data classification.
Further reading
For broader context on how models handle subjective inputs, see our latest research in Artificial Intelligence.
Live Poll
Do you trust AI models to provide objective evaluations of human traits?
