A new study published in Nature has found that women are systematically depicted as younger than men across a wide range of online platforms and by artificial intelligence models. The research analyzed 1.4 million images and videos, along with nine large language models, to assess how gender and age are represented online.
The team examined content from sources such as Google, Wikipedia, IMDb, Flickr, and YouTube. They also included major language models like GPT2 in their analysis. Their findings indicate that women consistently appear younger than men across 3,495 occupational and social categories.
“This kind of age-related gender bias has been seen in other studies of specific industries, and anecdotally, such as in reports of women who are referred to as girls,” said Berkeley Haas Assistant Professor Solène Delecourt, one of the study’s co-authors alongside Douglas Guilbeault of Stanford’s Graduate School of Business and Bhargav Srinivasa Desikan from the University of Oxford/Autonomy Institute. “But no one has previously been able to examine this at such scale.”
The researchers noted that the distortion is most pronounced in high-status, high-earning occupations. Additionally, they observed that mainstream algorithms amplify this bias: When generating nearly 40,000 resumes using ChatGPT, the system assumed women were younger and less experienced while rating older male applicants as more qualified.
“Online images show the opposite of reality. And even though the internet is wrong, when it tells us this ‘fact’ about the world, and we start believing it to be true,” said Guilbeault. “It brings us deeper into bias and error.”
To conduct their analysis, the researchers used multiple methods for assessing gender and age in visual media. In some cases they hired thousands of online workers to classify gender and estimate ages; in others they cross-referenced image timestamps with subjects’ birthdates for accuracy.
Across all approaches—human judgment, machine learning assessments, or objective information—the results showed a strong association between youthfulness and women compared to men. This trend intensified with job prestige or larger pay gaps between genders.
Text-based analysis supported these findings. The researchers looked at billions of words from Reddit, Google News, Wikipedia, and Twitter. Words related to youth were much more often linked with women than men.
“One concern people might have is that images and videos are kind of unique in that people can wear makeup or apply filters… That’s why we also looked at text, and we found exactly the same pattern,” Delecourt said.
The study also explored real-world effects through two experiments involving both human participants and AI systems like ChatGPT (gpt-4o-mini). In one experiment with around 500 participants searching for occupation-related images on Google Images versus unrelated objects (such as apples), those exposed to images featuring women estimated lower average ages for jobs compared to those seeing male-dominated imagery or no relevant imagery at all.
In another experiment generating resumes for 54 occupations using distinctively male or female names matched on popularity and ethnicity factors, ChatGPT assumed female candidates were younger by an average of 1.6 years—with more recent graduation dates—and had less work experience than their male counterparts. When evaluating resumes for identical positions, ChatGPT rated older men higher regardless of whether names were provided by researchers or generated by the model itself.
These patterns point toward a problematic feedback loop: biased portrayals reinforce public assumptions about age-gender roles in various professions—potentially influencing hiring practices—while AI trained on such data further amplifies these stereotypes.
Guilbeault highlighted concerns about how increasing reliance on algorithmically curated information could reinforce stereotypical expectations: “This is of particular concern given the internet is increasingly how we learn about the social world… Our study shows that they are reinforcing stereotypical expectations about how the world should be.”
Delecourt noted possible long-term impacts on young people absorbing biased representations online: “What was most striking to me… was how this online presentation has a much broader effect than I imagined when going into this… These misrepresentations feed directly into the real world in ways that could be widening gaps in the labor market.”
She added: “Overall, our study shows that age-related gender bias is a culture-wide, statistical distortion of reality, pervading online media through images, search engines, videos, text…and generative AI.”
“To fight pervasive cultural inequalities,” Delecourt said,“the first step is to recognize how stereotypes are coded into our culture, our algorithms,and our own minds.”
The full paper can be accessed here:
Age and gender distortion in online media and large language models (Nature).
Authors include Douglas Guilbeault (Stanford Graduate School of Business), Solène Delecourt (UC Berkeley Haas School of Business), Bhargav Srinivasa Desikan (Oxford University & Autonomy Institute). The project received funding from several organizations including The Fisher Center for Business Analytics; The Center for Equity,G ender,and Leadership; The Barbara & Gerson Bakar Fellowship;and The Universityof California,Berkeley.
