The hand of someone reviewing internet job listings with a magnifying glass. Credit: Getty Images
Research

In Study of DEI Language in Job Ads, UMass Amherst Professor Finds Applicants Aren’t Fooled by ‘Superficial Language’

AI analysis of nearly 29 million UK advertisements shows inclusive language can either reinforce or disrupt the status quo.

An interdisciplinary study co-authored by a University of Massachusetts Amherst researcher has found that despite regulatory and cultural shifts encouraging the use of language promoting diversity, equity and inclusion (DEI), job advertisements containing such language often failed to produce significant changes in workforce diversity. 

The study of 28.6 million job listings in the United Kingdom identifies distinct patterns in how ad language is linked to the demographic makeup of the workforce. This language often signals characteristics expected of an ideal candidate, acting as a gatekeeper to the labor force. At the same time, the composition of the labor force itself can shape the wording of ads.

“This is significant because we found bidirectional effects. The ad language affects industry-level workforce composition and, in turn, the composition affects the language,” explains Monideepa Tarafdar, Charles J. Dockendorff Endowed Professor at UMass Amherst’s Isenberg School of Management and principal investigator of the grant that funded the project. The study, published in PNAS Nexus, employed an artificial intelligence model trained on a literature-based and contextualized word inventory to measure bias in the ads, which were posted between 2018 and 2023. This analysis was combined with labor force data at the industry level.

The findings reveal that the words used in job postings can either reinforce or disrupt gender and racial segregation in the workplace, offering insights into how workplace inclusion initiatives intersect with workforce composition.

Monideepa Tarafdar

If it’s just superficial language, people can figure that out. DEI does not exist in words and phrases. It exists in what you do.

Monideepa Tarafdar, Charles J. Dockendorff Endowed Professor in the UMass Amherst Isenberg School of Management


For instance, the study shows that postings with “family-friendly” terms such as “flexible work” or “parental leave” tend to attract more women, increasing their representation. Conversely, ads with explicit gendered pronouns, such as “he/she,” and traditionally “feminine” descriptors like “caring” or “attentive” can actually deter women from applying, perpetuating gender imbalances.

“If it’s just superficial language, people can figure that out,” Tarafdar notes. “DEI does not exist in words and phrases. It exists in what you do.”

When it comes to how ads are written, the study demonstrates that industries with a higher proportion of women or racial minorities were more likely to feature language signaling inclusivity and diversity. It also identifies patterns of “compensation,” where employers in male-dominated or racially homogeneous sectors used inclusive language to attract underrepresented groups.

However, these strategies are not uniformly effective. For example, industries with higher minority representation often included more workplace inclusion language, but this had little measurable effect on increasing diversity further. The researchers hypothesize that this could be due to skepticism about the sincerity of these commitments or structural barriers in hiring processes.

Tarafdar says this underscores the need for more targeted and meaningful approaches to DEI in hiring that don’t unintentionally exclude potential candidates. 

“Employers should not use boilerplate language. Instead, they should actually describe what they do to promote an inclusive workplace,” she says. “Inclusion happens through everyday practices of organizational colleagues.”

The team of researchers who collaborated on this study also includes Yang Hu, professor of global sociology at Lancaster University; Nicole Denier, associate professor of sociology at the University of Alberta; and Lei Ding, doctoral student at the University of Alberta, among others. The project was supported by the Economic and Social Research Council and the Social Sciences and Humanities Research Council under the Canada-UK Artificial Intelligence Initiative