The article makes significant contributions by providing the first large-scale evidence that age-related gender bias is a widespread distortion, present in digital visual content (images, videos) and in nine language models, and that it is also systematically amplified by algorithms. According to this bias, there is a tendency to assume that women are younger—and therefore less experienced—than men in relation to their professions or social roles. 

The relevance of this study lies in the rigorous quantification of this bias against verifiable objective anchors—in particular, US Census data showing that there are no systematic age differences between women and men in the working population—which allows us to move beyond the controversial debate about the accuracy of stereotypes. The study causally demonstrates that Google Image searches amplify the perceived age gap by 5.46 years, and that ChatGPT propagates this bias by generating CVs that assume women are younger and less experienced than men, especially in high-status, high-income occupations. 

This highlights the urgent need for intervention, particularly given that bias is strongest where women face persistent pressure to appear youthful (the “beauty tax”) and where older women suffer disadvantages in hiring and promotion (gender ageism). This work is related to the work carried out by ELLIS Alicante on attractiveness bias and beauty filters (What is beautiful is still good: the attractiveness halo effect in the era of beauty filters), as beauty filters tend to make people look younger (5.87 years on average).

The technical soundness of the article is high, characterised by a large-scale methodology combining the analysis of nearly 1.4 million images and videos from Google, Wikipedia, IMDb, Flickr, and YouTube, a pre-registered human experiment with a nationally representative sample in the US (n=459), and a quantitative audit involving nearly 40,000 CVs generated by ChatGPT. The methods are carefully controlled to generalise the results, including comparison with census data and the use of objective age information, as in the case of famous people. 

The main limitation is that, while the study confirms algorithmic amplification, identifying the precise causal mechanisms by which industry-specific aesthetic norms or biases are transferred to generative AI remains a critical area for future research. It would also be important to develop strategies to mitigate this bias, as well as to extend the work to other regions of the world, since both the census data and user studies have been conducted with representative populations from the United States. 

EN