AI has broken recruitment. Here’s how employers can fix it

How IT and HR leaders can create a fairer and more effective recruitment process

Image:
AI has broken recruitment. Here’s how employers can fix it

AI recruitment tools can process applications quickly, but systems trained on historical hiring data may repeat existing biases and reject candidates with unconventional experience. Employers should combine automation with human intervention, examine rejected applications, track outcomes across demographic groups and scrutinise training data before procuring hiring technology.

Key takeaways

AI HR tools may have made initial screening of candidates more efficient, but at the terrible cost of eliminating jobseekers who do not fit narrow criteria, exacerbating inequalities and causing frustration among applicants and hirers alike.

The increasing use of artificial intelligence in recruitment has made candidates feel their own true voice is unacceptable. Applications are tweaked to make them more machine-parsable; chatbots are deployed to craft “more professional” CVs with the result that all applications start to look alike, each written in slick, bland corporatese.

Most likely, this effort will be to no avail. The majority will be rejected automatically, without a human being as much as glancing at them, the applicant left in the dark as to what they might do better next time. Meanwhile, organisations complain about being unable to fill vacant positions.

How AI recruitment tools screen out suitable candidates

Machines are judging machines and it has broken recruitment.

image
Profile image
description
Abhijit Bhaduri

“It's a very, very poor system,” says Abhijit Bhaduri, a former Microsoft general manager for global learning and development and author of several books on skills and recruitment, including the forthcoming AIdentity. Automated systems cannot judge what really makes people tick, he contends; nor can they predict how they will perform as part of a team.

“I think it's absolutely stupid to outsource the best part about building a really unique combination of people and to do it through machines. It's a tragedy.”

Bhaduri acknowledges that HR AI systems can be efficient for sorting, but says that like any tool they can be used the wrong way. They optimise what can be measured rather than what actually makes an individual valuable, and the two are not necessarily the same.

Karen Blake is head of the Equality, Diversity, and Inclusion in Science and Health (EDIS) coalition and a member of the Women in Tech Taskforce at the Department for Digital, Culture, Media and Sport. She believes that automated tools have introduced a distance between candidate and hirer.

“On one side, so many talented people are being screened out, often without feedback and sometimes within hours of applying. On the other, organisations are becoming increasingly indecisive, with roles left open for months, processes that stall and candidates left waiting without any word.”

Automated hiring reproduces bias at scale

She mentions another problem with AI systems: they recruit for what they already know.

AI tools trained on historical hiring data tend to select profiles that match people previously hired and promoted. The apparent objectivity of the technology can therefore conceal bias in a way that can be hard to detect. Blake offers a couple of pertinent examples of how people can be excluded to the detriment of both sides.

“Someone who spent three years caring for a parent with dementia has learned patience and judgement that would serve them well in many jobs, and someone who helped run the family shop as a teenager probably understands margins and customers better than most graduates, yet a keyword filter sees a gap in one case and a mismatch in the other.

“‘Kept the family business going’ doesn’t register as ‘P&L management’, even if it’s the same skill learned under much more pressure.”

Biases built unwittingly into automated systems have serious implications for diversity in tech, particularly disadvantaging returners, women, neurodivergent candidates, older workers and those with unconventional career paths. And they disadvantage them at scale.

“These systems process applications in huge numbers, so a single flaw gets repeated thousands of times and ends up deciding who is treated as a candidate at all.”

AI-written CVs and automated screening are creating a vicious circle

There’s an obvious paradox here: candidates are encouraged to demonstrate their individuality but have to do so in a way that is increasingly standardised and optimised for machine-readability. It’s a self-reinforcing vicious circle in which neither side achieves what it wants, says Blake.

“Jobseekers use AI to cope with the number of applications they have to send. Recruiters use it to cope with the number they receive, and each side’s shortcut makes things harder for the other.”

Of course, the recruitment playing field has never been level. Those who went to the right school or have been coached in professional-sounding corporate language and presentation skills have always had an advantage. But the mass standardisation of speech patterns by LLMs has created a new norm that “feeds recruitment’s oldest habit of hiring the safe bet,” Blake asserts.

“AI makes it quicker and easier to pick the candidate who looks like the last person who did well, and the people who don’t quite fit the usual profile, who might once have caught a curious recruiter’s eye, are dropping out before anyone looks twice. Edge cases are becoming cliff edge cases.”

Can personal branding help young people overcome the experience gap?

The promise that AI will reduce headcount is already threatening the first rung of the career ladder in several professions. Denied this foothold, and unable to demonstrate the years of work experience required by many job ads, some younger people are attempting to create social media brands for themselves. There are positive and negative aspects to this.

Experimenting across several areas of work can certainly help develop transferable skills, says Bhaduri. In a fast-changing environment, diversity of experience can be more important than dusty qualifications, and proof of proactivity can still catch an employer’s eye.

image
Profile image
description
Karen Blake

A major caveat, however, is the importance of remaining true to oneself. Building a personal brand can simply replace one conformist metric, qualifications, with another, online visibility.

“A lot of what we’re seeing is people cosplaying professional, copying the tone and polished confidence of people they think they’re supposed to sound like,” says Blake, who offers the following advice.

“Focus on being curious rather than looking impressive. Your network matters far more than your followers. You can build a network through mentoring schemes, local groups, volunteering, online communities or simply getting in touch with someone whose work you admire.”

How jobseekers can build an AI-proof professional identity

Work is far more than simply a way to earn a crust. It is also how people satisfy their psychological needs for meaning, belonging, mastery, predictability and status. But the rapidly changing relationship between the worker, the work and the market for work is removing these supports, one by one.

Stuck in the Catch 22 of lacking experience (or having the wrong type of experience or gaps in experience), unable to find jobs that would help them gain experience; and with the first rung of the ladder disappearing, what are jobseekers to do?

In the long term they need to create an AI-proof identity, says Bhaduri, who wraps this up in the Urdu concept of “hunar”. This, he explains, is “the irreducible human signature formed by lived experience, judgment, taste, memory, victory, failure, culture, instinct and practice; the unique personal capacity that makes the same skill different in different hands.”

Hunar is distinct from talent and skill, he goes on: “Skill is what you do. Talent is how easily you do it. Hunar is the way you do it that no one else can replicate.”

Importantly, as the workplace changes, hunar cannot be replicated by AI.

Bhaduri believes that as AI transforms employment, future jobs will arise at the intersections between existing disciplines. The goal, therefore, is to avoid getting stuck in a narrow work identity in the early years. Hunar can be developed by seeking out work, voluntary projects, interests and experiences in line with, but not directly related to, one’s main field of expertise.

Candidates should think first and use AI second

Developing hunar means avoiding becoming over-reliant on chatbots, he insists. “Start with the human first. So you write it out, or you read the book, or you read the report, and you say, ‘for me, the three most important ideas that I can use are these’.”

Only then should AI be brought in, perhaps to challenge the reasoning: “What do you think I'm missing in this entire argument?”

Overuse of chatbots risks reducing an individual’s distinctiveness towards a common denominator. It also has a deleterious effect on memory and sense-building, says Bhaduri.

“You don't remember what you have not worked on. Cognitively, you need that iteration where you're struggling with something in order to make it make sense. You're fitting it into your worldview.”

How employers can stop AI overlooking suitable candidates

What should organisations do to make sure they aren't overlooking suitable candidates?

As discussed at the start of this article, organisations are struggling to fill roles with the right people, and the experience is bruising for both sides.

“People get hired for their competency and fired for their personality,” as Bhaduri puts it, adding: “Organisations need to stop confusing efficiency with effective hiring.”

In addition to skills and experience, organisations need to assess, at an early stage, candidate’s hunar – their judgement, flexibility, unusual perspectives, empathy, creativity and interpersonal skills. But the ability to do this has diminished as automated systems have become more prevalent, with fewer opportunities to exercise fine human judgement.

Bhaduri suggests experimenting with different approaches rather than simply relying on AI.

Combine automated screening with alternative routes into recruitment

“If I had to choose between, say, 10 candidates, I could choose three from an automated system, but I would choose seven from a different system where somebody who doesn't fit the mould can come in.”

Blake agrees that hiring systems should be redesigned to include what they currently miss. “Remember that AI tools are often trained on your past hiring and can only find people who look like your past, and your past is rarely what your future needs,” she states.

“As AI takes on more routine work, organisations will rely more on judgement, adaptability and the ability to see problems from different angles, and those qualities often come from the varied lives and unusual career paths that keyword filters struggle to spot.”

What IT leaders should ask when procuring AI recruitment tools

If possible, these steps should be taken at the procurement stage.

Recruitment tools are often bought for reasons of speed and cost, but procurers should also scrutinise the training data and measures used to reduce bias, not only to improve hiring outcomes but also to reduce legal risk as regulations tighten, says Blake.

“If a tool disadvantages people because of a protected characteristic, it’s your organisation that has to answer for it under equality law; and the EU’s rules on AI in hiring will apply from 2nd December 2027.”

CIOs and CTOs should look for documentation about the limits of the system, provisions for human oversight, bias and adverse-impact testing, explainability, accessibility and known errors.

Audit rejected candidates and track outcomes by demographic group

Like Bhaduri, Blake urges organisations to test whether their system is really bringing forward the best candidates.

“Have someone regularly review a sample of the candidates your system rejected, and track how different groups move through each stage. If one group keeps falling away at a particular point, the problem is the process, not the people.”

She concludes: “The organisations that thrive will be the ones that use AI to find talent they would never have noticed before, not just to hire the same people faster.”