“Learn AI skills” has become the default advice for anyone worried about their job — from career coaches, from governments, from the tech companies selling the tools. What almost nobody has done is test that advice in a real hiring market, with real vacancies and real employers, to see how much those skills are actually worth to the people who need the biggest boost.
A team at Anglia Ruskin University did exactly that, and the results published in Industrial Relations: A Journal of Economy and Society are more honest than most of the conversation. AI training helps. It does not fix the system.
🔍 THE BOTTOM LINE: In a field experiment involving more than 3,000 job applications submitted for real London vacancies during 2025, women from underrepresented backgrounds who highlighted a six-month AI-related qualification on their applications received interview invitations at a rate of 14%, up from 11% without the AI signal — a relative improvement of just under a third — but still well below the 25% rate for comparable applicants who faced none of the same barriers. Skills opened doors; they did not remove the door frame.
What the experiment actually did
The study, led by professor Nick Drydakis of Anglia Ruskin University, created fictional but highly comparable female candidates and submitted applications for private-sector roles that did not require a college degree, across 10 different workplace fields in London during 2025. The applications varied on only a few factors: race, age, sexual orientation, autism spectrum disclosure, and — the variable that matters for this story — whether the application highlighted a six-month AI-related professional development qualification completed through a vocational training provider.
The comparison group’s numbers set the scale of the problem first. Among women not from underrepresented groups, 25% of applications that did not mention AI skills still drew an interview invitation. Among underrepresented women without AI skills, the same rate was just 11% — less than half. Those are applications for jobs where the qualifications matched; the only difference was who the applicant appeared to be.
Add the AI qualification, and the underrepresented applicants’ rate rose to 14%. In relative terms — the way a student deciding whether to enrol should read it — that is a lift of just under a third. In absolute terms it is three percentage points recovered out of a fourteen-point gap, if the non-underrepresented baseline is taken as the reference.
The researchers also tracked what kind of interviews the AI signal bought. Underrepresented applicants were more likely to receive invitations for lower-paid vacancies among the jobs they applied for — a pattern the study describes as “wage sorting.” Having AI-related skills improved access to better-paid vacancies too, but notable gaps persisted even there.
The study’s own caveat, and it cuts both ways
The authors are careful about scope: the experiment tested whether AI qualifications improved outcomes for underrepresented women. It did not measure whether the same qualification would add the same boost for majority-group applicants. A reader cannot conclude from this study that the AI credential is worth more to one group than another — only that for the group that history suggests needs help most, the credential measurably helped and measurably under-delivered relative to the gap.
The framing from Drydakis is unusually direct for a skills study: “AI training can help open doors, but it should not be viewed as a substitute for fair recruitment practices. If we want technological change to create opportunities for everyone, investment in digital skills must be matched by strong efforts to tackle discrimination and ensure qualifications are assessed consistently across all applicant groups.”
That matters because of who is funding the “learn AI” push. Governments across the English-speaking world have poured money into reskilling programmes on the quiet promise that skills equal opportunity. In New Zealand, government AI-spend doctrine has favoured training tied to sectors the country actually earns money from — agriculture, food processing, logistics — rather than certificates for their own sake. This study is the first serious audit of that promise’s fine print, and it says: fund the training, expect roughly a third more doors to open, and budget for the fact that the rest of the gap needs different tools entirely.
Why it is still — genuinely — good news
It is worth resisting the reflex to file this story under gloom. Three things in the data point up, not down.
One: the largest-ever controlled test of AI credentials in live hiring found the credentials work. They are not snake oil. Applications carrying an AI qualification were invited to interview about a third more often. For a six-month vocational course, that is one of the higher returns measured on any training investment in recent labour research.
Two: the study found the AI signal improved access to better-paid vacancies — it pushed applications up the wage-sorting ladder, even if not all the way.
Three: the experiment exists at all. Labour economists have argued about AI’s effect on hiring for years mostly with survey data; here is a randomised design run against real vacancies with a peer-reviewed paper attached. The field is moving from vibes to evidence, and the evidence so far supports a moderate, testable optimism: invest in skills, audit the recruiters.
The findings also line up with what older workers have already demonstrated here. APAC employment data showed older workers pivoting into AI-adjacent roles with surprising success, and LinkedIn’s count of 750,000 AI-linked new jobs since 2023 documented that the new roles exist and are not all for computer-science graduates. The ARU study adds the missing sentence: the door is real, it opens a bit wider with the right credential, and it still does not open for everyone equally.
What it means for New Zealand
The closest local benchmark is SEEK’s New Zealand job-ad data from July, which showed postings mentioning AI skills up 94% in a year — local demand for the exact credential type this study tested. The supply side is the harder half. The national unemployment rate sat at 5.6% in the June 2026 quarter with underutilisation at 13.8% and 440,000 people underutilised, per Stats NZ — a pool of talent that reskilling programmes are explicitly built to reach.
Two practical reads follow for NZ readers. First, for job seekers in the underutilised pool: the evidence says a short, vocational, work-focused AI qualification is worth putting on the application — measurably, though modestly. Second, for whoever is designing the programmes: the study’s warning lands here too. New Zealand employers hire from smaller applicant pools than London’s, where a single biased shortcut in a screening process can quietly decide whole sectors’ hiring. Skills funding without recruitment-fairness auditing buys the three points and leaves the fourteen on the table.
❓ FAQ
How much did AI skills improve interview chances in the study? Applications from underrepresented women that highlighted a six-month AI qualification received interview invitations 14% of the time, up from 11% without it — a relative improvement of just under a third.
Did AI skills close the hiring gap entirely? No. Comparable applicants from non-underrepresented groups received invitations 25% of the time even without AI skills named. Skills narrowed the gap; they did not eliminate it.
How was the study conducted? Researchers submitted more than 3,000 applications for real private-sector vacancies in London during 2025, varying only the candidates’ apparent characteristics and whether an AI-related qualification was named. The paper appeared in Industrial Relations: A Journal of Economy and Society.
Is this relevant to New Zealand job seekers? Yes. NZ job ads mentioning AI skills rose 94% in a year per SEEK data, and the study suggests short vocational AI credentials carry a measurable interview-odds benefit — while also showing why programmes must be paired with fair-recruitment checks.
🔍 THE BOTTOM LINE
The first serious real-world audit of the “learn AI skills” advice found it works — and that it is not enough. A six-month AI credential lifted interview odds for underrepresented women from 11% to 14% in live London hiring, a genuine relative gain from a short course, before a stubborn 11-point gap to comparable majority applicants took over. For career-planners the arithmetic is clear: get the credential, because the door opens a little further. For programme designers, be honest that skills policy alone is buying a third of the problem’s solution. Both halves are supported by the data. Neither is served by pretending the other half does not exist.
📰 Sources
- Nick Drydakis — “Artificial Intelligence-Related Digital Skills and Employment Outcomes for Underrepresented Women”, Industrial Relations: A Journal of Economy and Society (online 29 September 2026, DOI 10.1111/irel.70047)
- Anglia Ruskin University via Phys.org — “AI skills not enough to level the playing field — study” (8 October 2026)
- Stats NZ — “Unemployment rate at 5.6 percent in the June 2026 quarter”
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
