Researchers have documented a gendered age gap within image databases used to train machine learning algorithms.
The findings suggest that artificial intelligence-based tools used by employers to review resumes may advantage older men while disadvantaging older women and younger applicants.
Douglas Guilbeault, assistant professor of organizational behavior at Stanford Graduate School of Business and lead author of the study, explained that when ChatGPT generates resumes, it relies on prevailing notions of desirable candidates and relevant skills, which introduces numerous opportunities for stereotypes to influence outcomes.
After analyzing nine large language models trained on billions of words, the researchers identified a biased and distorted representation of older women, indicating that bias persists even in the absence of visual depictions.
Guilbeault and colleagues prompted ChatGPT to generate over 34,500 unique resumes for 54 occupations using both male and female names.
Although AI companies can implement filters to block overtly biased or stereotypical content, Guilbeault emphasized that gendered ageism must be addressed at a foundational level.
Mainstream algorithms further amplify age bias.
In an analysis of nearly 1.4 million images and videos from platforms such as Google, Wikipedia, IMDb, Flickr, and YouTube, as well as nine large language models, the study found that searching for occupational images intensifies age-related gender bias in users' beliefs and hiring preferences.
ChatGPT tends to assume that women are younger and less experienced, often rating older male applicants as higher quality.
The research demonstrates that gender and age are jointly distorted across the internet and its mediating algorithms.
The study, "Age and Gender Distortion in Online Media and Large Language Models," was published in Nature in October 2025.
Inaccurate stereotypes about older women are not only prevalent in online images and videos but are also perpetuated and amplified by large language models.
Although occupational terms such as hairdresser, nurse, soldier, or mail carrier are gender-neutral, the researchers determined that algorithms consistently privilege male-associated content.
While there are more images of men available online, the study notes that sufficient images of both women and men exist across all occupations examined.
The research identified widespread bias against older women on prominent image and video platforms, as well as within the algorithms that drive widely used artificial intelligence tools such as ChatGPT.
In portrayals of higher-status and better-paid occupations, women are systematically depicted as younger than their male counterparts.
For example, images of software developers disproportionately feature men, whereas written descriptions are more likely to include women.