The experimental design was deliberately meticulous. Researchers constructed fictional but highly comparable female candidates and submitted applications for genuine non-graduate, private-sector vacancies spanning ten different workplace fields. The applications differed in only a small number of carefully controlled respects: the candidates’ race, age, sexual orientation, or disclosure of being on the autism spectrum, and whether the application highlighted a six-month AI-related professional development qualification completed through a vocational training provider. Because everything else about the candidates was held constant, any difference in the rate of interview invitations could be attributed to these specific characteristics rather than to variation in experience, education, or writing quality. This matched-resume methodology is considered the gold standard for detecting discrimination in hiring, since it isolates the causal effect of a particular trait on employer behaviour in real labour markets rather than in hypothetical scenarios.

The baseline results were stark. Among women applicants who did not belong to any underrepresented group and whose applications made no reference to AI skills, 25 percent received an interview invitation. For women from underrepresented backgrounds who likewise lacked AI credentials, the figure dropped to just 11 percent. In other words, simply signalling membership in one of the studied groups more than halved the probability of being invited to interview, even though the candidates were otherwise indistinguishable on paper. The underrepresented categories examined were age, race, sexual orientation, and autism-spectrum disclosure, and the pattern of disadvantage held across all four, suggesting a broad and systemic rather than isolated phenomenon in the recruitment process.

AI training did move the needle. When underrepresented applicants signalled the AI-related digital qualification on their applications, their interview success rate rose to 14 percent, an improvement of just under a third relative to their non-AI counterparts. That increase is far from trivial for individual job seekers: a candidate who previously faced roughly a one-in-nine chance of reaching the interview stage saw her odds climb to roughly one in seven. Employers, the findings suggest, do genuinely value demonstrated competence with AI tools, such as using AI-assisted software to automate administrative tasks, analyse data, or manage digital records. The study extends Drydakis’s AI Capital theory, which frames AI-related skills as a form of human capital that can generate measurable returns in the labour market, and this experiment provides some of the first direct evidence that those returns extend to groups that traditionally face hiring barriers.

Yet the central finding is the persistence of the gap. Even with the AI qualification, underrepresented women’s 14 percent interview rate remained well below the 25 percent achieved by majority-group applicants without any AI credential at all. The skills premium, in other words, narrowed the inequality but did not come close to eliminating it. A candidate from an underrepresented group would apparently need to bring demonstrably more to the table than an otherwise identical majority-group candidate simply to reach the same stage of the hiring funnel. This asymmetry has important implications for policy debates that position digital upskilling as a primary route to labour market equality, because it shows that supply-side interventions targeting workers’ skills cannot, on their own, correct demand-side biases embedded in employer behaviour.

The study also uncovered a subtler pattern the researchers describe as wage sorting. Underrepresented applicants were more likely to receive interview invitations from lower-paid vacancies within the fields they applied to, indicating that the disadvantage they face is not only about whether they hear back from employers but also about which tier of jobs responds to them. Possessing AI-related skills improved access to better-paid vacancies, partially counteracting this downward sorting, but notable gaps persisted even at the upper end. This finding adds a distributional dimension to the evidence: discrimination may operate not just as an absolute gatekeeping mechanism but as a steering mechanism, channelling underrepresented candidates toward less remunerative positions even when they do successfully attract employer interest.

Professor Drydakis, who directs the Centre for Inclusive Societies and Economies at Anglia Ruskin University, emphasised that the results carry a dual message for policymakers and employers. As AI becomes increasingly embedded across workplaces, he noted, there is growing interest in whether developing AI-related skills, such as training in the use of artificial intelligence tools and software in workplace settings, can help improve employment opportunities for people who traditionally face barriers in the labour market. His findings suggest that employers do value these credentials and that they can improve interview prospects for underrepresented women, but he stressed that skills alone are not enough to overcome the inequalities such candidates face. In his words, AI training can help open doors, but it should not be viewed as a substitute for fair recruitment practices.

The implications extend to how governments and training providers design digital inclusion programmes. Subsidised AI bootcamps and vocational AI qualifications have become popular policy instruments, premised on the idea that equipping disadvantaged workers with cutting-edge skills will translate directly into better employment outcomes. This study partially validates that premise, demonstrating a genuine, quantifiable return on AI human capital for the groups studied. But it also issues a caution: if the returns to identical credentials are systematically smaller for underrepresented applicants, then investment in digital skills must be matched by strong efforts to tackle discrimination and to ensure that qualifications are assessed consistently across all applicant groups. Otherwise, upskilling programmes risk raising expectations without delivering proportional results, and may even place the burden of overcoming bias on the very people who experience it.

Methodologically, the research stands out for its scale and ecological validity. Because applications were submitted for real vacancies in a major labour market during 2025, the findings capture employer behaviour as it actually occurs, not as employers report it in surveys, where sensitivity to social desirability often masks discriminatory tendencies. The focus on non-graduate private-sector roles is also significant, since these positions account for a large share of employment and are typically less shielded by formalised, credential-driven selection procedures than graduate recruitment, potentially leaving more room for subjective judgement to influence outcomes. The authors note that the study focused specifically on whether AI qualifications improved outcomes for underrepresented women and did not measure whether the same qualification would provide a similar boost for majority-group applicants, a question that future research could address to determine whether AI skills confer a differential premium across groups.

As artificial intelligence reshapes occupational requirements across the economy, the question of who benefits from the AI skills revolution is becoming urgent. This experiment provides an early and sobering data point: the benefits are real but unevenly distributed, accruing to underrepresented women in the form of improved, yet still substantially disadvantaged, interview prospects. The research suggests that technological change alone will not level the playing field. Closing the remaining gap will require employers to examine their own selection processes, regulators to enforce consistent assessment of qualifications, and policymakers to pair digital skills investment with robust anti-discrimination efforts, so that the doors AI training opens are the same doors, leading to the same rooms, for every candidate who walks through them.

Subject of Research: The effect of AI-related digital skills on employment outcomes for women from underrepresented backgrounds

Article Title: AI skills not enough to level the playing field – research

Article References: AI skills not enough to level the playing field – research. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, digital skills, labour market, hiring discrimination, field experiment, underrepresented groups, women in employment, vocational training, wage sorting, recruitment, AI capital, Anglia Ruskin University