Something strange has happened to hiring. The tools have never been more powerful. The data has never been richer. The process has never been more automated. And yet, finding the right person for the right job has never felt harder.
Applications are flooding in. Inboxes are full. Shortlists are long and mostly wrong. Recruiters are exhausted. Candidates feel unseen. And somewhere in the middle, the actual goal – connecting the right human to the right opportunity – is getting lost.
We need to be honest about why. Not because AI is the villain, because it isn’t. But because AI, as it’s currently being used in recruitment, is accelerating a problem we created long before the algorithms arrived.
The volume trap
For years, the default assumption in hiring was that more applications meant better outcomes. Post a job, cast the widest net, sift the pile. The logic seemed sound, but it wasn’t.
A role posted today can attract hundreds of applications within hours, yet industry data shows that 77% of those applications don’t match the role they’re trying to fill – that’s not a talent pipeline, that’s noise. Even skilled recruiters cannot meaningfully engage with that volume, spending three hours, on average, sorting through shortlists that are, by the numbers, mostly wrong.
AI did not create this problem. But cheap, poorly designed AI has made it dramatically worse. Tools that make it frictionless to apply – auto-fill, AI-generated cover letters, one-click submissions – have flooded hiring pipelines with applications that were never really meant to be there. The result is a market where everyone is technically applying, and no one is really being seen.
This is the volume trap, and most of the industry is still walking straight into it.
What ‘more efficient’ actually means
There is a version of AI in recruitment that promises to solve this by making the filtering faster. Screen more CVs, score more candidates or reject more automatically. The maths looks compelling until you ask: what are we actually optimising for?
Speed, applied to the wrong goal, doesn’t fix the problem, it scales it. An AI that learns to replicate historical hiring patterns will screen out career changers, flag atypical profiles as anomalies, and quietly discriminate in ways no one programmed it to, because it learned from data that already contained those biases. This is not a theoretical risk. The Amazon case in 2018 is the textbook example: their AI recruiting tool was trained on a decade of hiring decisions, most of which favoured men, and the system had simply learned what ‘good’ looked like from the past.
A poorly designed AI applied to recruitment doesn’t just speed up the hiring process. It can reinforce exclusion, standardise profiles, and erase the very notion of human potential. If AI is taught to find more of the same, that is exactly what it will do – efficiently, at scale, invisibly.
That’s not a hiring solution. That’s a machine for missing people.
The jobs AI should actually be doing
The right question isn’t how we make AI filter faster. It’s how we use AI to free up human judgement for the moments that actually matter.
This is a distinction that changes everything about how you build a product. When AI handles the mechanics – drafting job descriptions, scanning CVs, scheduling interviews, sorting applications against clearly defined criteria – recruiters get time back. Not time to review more CVs. Time to have real conversations, to understand context and to ask the question a CV can’t answer: does this person belong here?
Used well, AI should introduce more serendipity, not less – surfacing unexpected matches, not just confirming existing assumptions about what a ‘good’ candidate looks like. The career changer with an unconventional path. The junior candidate who doesn’t fit the template but has every quality that matters. The person who, by any algorithmic measure, looks like a risk, but by any human measure, looks like exactly what the team needs.
These are the candidates who get missed when the process is optimised purely for speed. They are also, often, the candidates who turn out to be the most valuable hires.
Human judgement is not optional
There is a hard line here: AI should never make the final call. Not in responsible products, and not in responsible hiring practice.
This isn’t just an ethical position, though it is that. It’s also becoming a regulatory requirement. The EU AI Act classifies recruitment tools as high-risk AI systems, with obligations due to apply from August 2026: transparency toward candidates when AI influences a decision, human oversight of all significant outcomes, and non-discrimination requirements with audit trails to prove it.
These aren’t burdens. They’re the floor. The principle that humans remain responsible for every significant hiring decision – that AI supports but never supplants human judgement – is what allows the whole system to work. The moment you remove that principle, you stop hiring people and start selecting outputs.
Responsible platforms build around a simple rule: AI does the mechanics, humans make the decisions. That means mandatory human review before any candidate is progressed or rejected as well as bias-checking job ads before they’re published. It also means disclosing to both hirers and candidates wherever AI is used and it means not training matching models on historical decisions – because the past is not always a reliable guide to what ‘good’.
What transparency actually requires
One of the most important things we can do right now is be honest with candidates. Not just legally, but genuinely.
Candidates deserve to know when AI has influenced something that affected them. Not buried in terms and conditions, and not as a footnote. As a clear signal – this is where AI was used, this is what it did, and here’s what happened next.
Simple visual indicators – a colour code or clear label – can mark where AI has been applied in a platform. It’s a small thing , but it matters because it tells candidates: we are not hiding the machine.
Nine out of ten candidates now ask about AI in their hiring process – that shift tells us that trust has become a feature, not an optional extra. Candidates who understand how AI is being used, and who trust that it’s being used responsibly, engage more fully whereas those who feel processed or scored in ways they can’t see or understand, disengage. The experience of being hired matters as much as the outcome.
Where the industry needs to go
The talent market is at an inflection point. The technology is not going away. Nor should it. But the question of how it’s used – whether it expands the field of human possibility or narrows it – is still very much open.
The path forward requires a few things the industry has been reluctant to commit to. Genuine transparency about AI use, not performative disclosure. Human oversight that is real, not rubber-stamp. Governance structures that include legal, HR, and DEI in AI decisions – not just product and engineering. And a willingness to measure something harder than speed: match quality, candidate experience, retention, belonging.
It requires asking, for every AI feature built or purchased: what is this optimising for? And who does it miss?
The companies that get this right will not just hire better. They will build something increasingly rare: workplaces where people feel found, not selected. Where the process of joining tells you something real about what it means to belong.
That is the version of AI in hiring worth building. Not faster noise. A better signal.
