AI Didn’t Invent Ageism in Hiring, It Just Made It Faster

Stacey Duguid calls it Botoxing her CV. After 16 months of sending what she calls “gazillions” of job applications, the 52-year-old former fashion executive started stripping her age and experience from her CV. She told the BBC that AI “holds a mirror up to society. It’s a reflection of our bias.” The BBC could not confirm AI screening caused Duguid’s rejections specifically. Employers rarely disclose how their processes work, and that gap is itself part of the problem. Duguid’s diagnosis is right; her treatment is not. The mirror is not the fault. What it reflects has been the problem for years, and a different mirror will not change what shows up in the glass.

A Grade From A to D, and No One Who Can Explain It

Laura Holden, an AI lawyer and founder of the Bonsai AI and legal consultancy, described one CV screening tool to the BBC, which grades applications A to D and nudges recruiters toward higher grades. She said companies “often do not understand how the tools work,” which she believes can cause “significant harm.” Global regulation is lacking, she added, so bias is hard to prove since providers need not show how their systems reach a score.

Pull the automation out of the story and the pattern is familiar: a recruiter skims a stack of CVs and downgrades anyone whose graduation date implies an age they would rather not hire. No dashboard records that decision, and no audit trail exists to challenge it. The AI version is not new, it is the same sorting instinct running at volume, wearing a grade that makes it look objective. That is a governance failure, not proof that software has a unique capacity for prejudice.

The Seven-Year Gap Note Is an Old Habit, Automated

Holden flagged a specific mechanic: some tools mark a seven-year career gap as “a thing to note,” shaping a recruiter’s view before anyone reads what that person did during those years. Koeyli Jaluka, 49, headhunted for over a decade before her redundancy, has applied for 442 jobs and has occasionally been told she is “too senior,” which she reads as code for old, while Anna Cowie, after 30 years in advertising, prefers networking in person to submitting CVs.

Dr Eleanor Drage, a senior researcher at the University of Cambridge, says human recruiters are better placed to recognise the value of someone returning from a break. A model, she argues, cannot assess intrinsic value or personality the way a person can, though that does not prove unstructured judgement is fairer on its own. Drage does not reject AI outright: she wants firms to audit their own hiring practices with it, not screen vulnerable groups such as older women re-entering work.

Career coaches already tell women like Jaluka to cut the first ten years off their CV before a human recruiter opens the file. The gap gets penalised twice: once by the software’s flag, once by the advice to hide it. Automation did not invent that penalty; it scaled it and made it harder to see who applied it, and when.

Shaving Off a Decade Is the Symptom, Not the Software

In our view, a minimum-years requirement in a job description often functions as a tenure filter, a proxy for age. Botoxing a CV and shaving off a decade are rational responses to hiring criteria built on tenure and titles rather than what a candidate can do. We think a model trained on historical hiring data would reproduce that filter just as faithfully as a hiring manager working from an unspoken template of “senior but not too senior.”

The City of London Women Pivoting to Digital Taskforce puts a number on the cost of leaving this unaddressed. Chair Caroline Haines told the BBC its survey of more than 1,000 women found 68% had not been given the opportunity to retrain into digital roles. Using Department for Education projections, the taskforce estimates AI and automation could displace hundreds of thousands of women’s jobs by 2035, with firms facing more than £750m in severance costs without significant retraining investment. “When the country is desperate for economic growth, we need to use whatever resource we have,” Haines said. This describes a retraining gap, not a verdict on any screening tool, though it shows what tenure-first hiring costs at scale.

Decision-Support on the Sales Deck, Decision-Making in the Funnel

Holden’s sharpest point deserves close attention. Providers, she said, were “very quick” to claim their tools are safe and reduce bias, while designing and marketing them in ways that push employers to reject large numbers of applicants on an AI-generated score alone. Sold as decision-support, used as automated decision-making. That gap between label and function is exactly where accountability disappears. It is the same gap whether the system doing the deciding is software or an unwritten rule a hiring manager has never had to defend out loud.

Across the teams we train, the fix is never trusting the algorithm more or less. It is knowing where a human must intervene in the funnel and what evidence they check before rejecting a candidate. Regulation will eventually force that transparency; TA leaders do not need to wait for legislation to demand it of their own process.

Stacey Duguid’s mirror metaphor holds up better than the debate it has triggered. Swap the screening tool for a stricter human panel, and Koeyli Jaluka’s 442 applications would still meet the same tenure filters and the same “too senior” euphemism. Ask yourself whether you could stand behind every rejection in your funnel, human or automated, with a reason you would say to the candidate’s face. If you cannot, the algorithm was never your real problem.

Original reporting: BBC.