The Perfect Storm: How AI Broke the Jobs Market

2026-02-10 · 8 min read

Something is happening in the UK labour market that most people haven't fully grasped yet. Two forces are converging — quietly, simultaneously, and with compounding effect — to create a crisis that neither employers nor candidates are equipped to navigate. One is about the disappearance of work itself. The other is about the collapse of the tools we use to find it.

Separately, either would be serious. Together, they form a perfect storm.

The displacement wave

Let's start with the numbers, because the numbers are hard to argue with.

UK job vacancies have fallen from over 1.3 million at their post-COVID peak to 726,000 as of January 2026. That's according to the ONS, not a think tank with an agenda. The vacancy-to-unemployment ratio has shifted from 1.9 unemployed people per vacancy to 2.6. More people are competing for fewer positions, and the gap is widening.

This isn't evenly distributed. Of the 18 industry sectors the ONS tracks, 14 saw year-on-year vacancy declines. Construction vacancies dropped 32.4%. Administrative and support services lost 36,000 positions. Wholesale and retail shed 94,000 workforce jobs in a single year.

The instinct is to explain this away as cyclical. Post-pandemic correction. Employer caution ahead of National Insurance increases. A cooling economy doing what cooling economies do. And some of that is true. But it doesn't explain the pattern.

The sectors losing the most vacancies are precisely the ones where AI and automation have the most to offer. Administrative roles — scheduling, data entry, document processing, customer service triage — are being absorbed by systems that don't need a vacancy posted to replace them. The work doesn't disappear overnight. It erodes. A team of eight becomes a team of five with better tooling. The three roles that vanished never get posted. They just stop existing.

Meanwhile, AI is creating new roles — prompt engineers, AI safety researchers, machine learning operations specialists — but not nearly at the pace it's eliminating or transforming existing ones. And the people displaced from administrative roles in Birmingham are not, by and large, the same people qualified for AI engineering roles in London.

This is a structural shift, not a blip. The labour market is not going to snap back to 1.3 million vacancies when confidence returns. The jobs that AI can do cheaper and faster are not coming back, and the new jobs it creates require entirely different skills, in entirely different places, for entirely different people.

The signalling collapse

Here's where it gets properly strange.

At the exact moment the labour market is tightening — more candidates chasing fewer roles — the tools designed to connect people with work have started making the problem worse.

Both sides of recruitment adopted generative AI at roughly the same time. Employers began using it to write job descriptions. Candidates began using it to write CVs and cover letters. Applicant tracking systems began using it to filter the results. And what emerged was an absurd closed loop: AI-generated job posts attracting AI-generated applications, evaluated by AI-powered screening tools.

The human signal — the thing that actually matters — has been stripped out at every stage.

This isn't speculation. SHRM estimates that 40 to 80 per cent of applicants now use AI to write their resumes and cover letters. Applications per job seeker have increased 47 per cent, according to Huntr's 2025 report — not because more people are searching, but because AI makes mass-applying trivially easy. Why spend an hour tailoring one application when you can generate twenty in the same time?

The result is an avalanche. An average of 22 applications per vacancy. Recruiters drowning in polished, virtually identical submissions. Every cover letter hitting the same keywords. Every CV structured to pass the same ATS filters. The quirky career path, the genuine enthusiasm, the unconventional background — all of it algorithmically erased before a human being ever looks at it.

And it's not just candidates gaming the system. Employers are doing it too. MIT Sloan research found that AI-drafted job posts were 15 per cent less likely to result in a hire. The ease of generating a listing meant that companies with no genuine hiring intent — or only a vague sense of what they actually needed — flooded the market with posts that attracted hundreds of carefully tailored applications for roles that were never going to be filled, or that bore little resemblance to the reality of the job.

Think about what this means in practice. A job description written by AI attracts a cover letter written by AI, which is filtered by an ATS powered by AI, and the candidate who gets through is the one whose AI was best at gaming the other AIs. At no point in this process has anyone demonstrated that they can actually do the work. The entire pipeline has become a contest between language models, with human beings as bystanders to their own careers.

The core problem is what you might call a signalling collapse. Written signals — CVs, cover letters, job descriptions — used to carry meaningful information. A well-crafted cover letter signalled effort, attention to detail, genuine interest. A detailed job description signalled that the employer had thought carefully about the role. These signals were imperfect, but they worked well enough for decades.

Generative AI destroyed their reliability almost overnight. When anyone can produce a flawless cover letter in thirty seconds, the cover letter stops telling you anything. When every job description reads like it was written by the same consultant, the description stops telling you anything either. The signal has been drowned by noise that looks identical to signal.

The consequences are measurable

You'd think that all this AI adoption would be making recruitment faster and cheaper. After all, that's the pitch. Automate the busywork. Screen candidates at scale. Match people to roles with algorithmic precision.

The data says the opposite.

Sixty per cent of companies reported increased time-to-hire in 2024, up from 44 per cent in 2023. Cost-per-hire has risen alongside it. Both metrics have been climbing for the past three years — the very period in which AI adoption in recruitment accelerated most dramatically.

This is counterintuitive until you understand the dynamics. AI didn't reduce the workload. It shifted it. Instead of spending time finding candidates, recruiters now spend time verifying them — trying to work out which of the hundreds of identical-looking applications represent people who can actually do the job. The screening problem hasn't been solved. It's been multiplied.

For candidates, the experience is even worse. Job seekers report sending hundreds of applications with response rates as low as 2 to 3 per cent. The process is demoralising, repetitive, and dehumanising. Every application requires adapting a CV, writing a cover letter, filling out forms that duplicate information the CV already contains, and then waiting in silence. Most applications disappear into a void. No feedback. No acknowledgement. No sense that a human being ever looked at what was sent.

And here's the cruel irony: the people who suffer most are the ones who were already worst served by the written application model. Tradespeople, care workers, early-career professionals, anyone who is brilliant at their work but struggles to write about it — AI hasn't helped them. It has raised the floor of what a “good” application looks like, while making every application sound the same. The bar went up, but so did the noise.

The perfect storm

These two forces are not separate problems. They are converging.

On one side, more people are entering the job market. Vacancies are down. Roles are being eliminated or transformed. The ratio of candidates to openings is climbing. The competition for every available position is intensifying in ways the UK hasn't seen since before the post-COVID hiring boom.

On the other side, the tools designed to help — the job boards, the application platforms, the AI writing assistants, the ATS screening systems — are actively making the matching problem harder. Every candidate looks the same on paper. Every job description reads like every other job description. The needle-in-haystack problem is exponentially more difficult when every piece of hay has been carefully manufactured to look like a needle.

More supply. Less demand. And a matching system that is producing more noise on both sides while systematically destroying the signal that actually matters.

This is the perfect storm.

It's not that AI is bad for recruitment. AI is extraordinarily capable. The problem is that we deployed it at every stage of an existing process without asking whether the process itself still made sense. We automated the writing of documents that nobody reads properly. We automated the screening of applications that no longer carry reliable information. We built faster, more efficient machinery for a fundamentally broken pipeline.

The question isn't how to use AI better within the current system. The current system — CVs, cover letters, job descriptions, keyword-matching ATS filters — was designed for an era when written documents were hard to produce and therefore carried real signal. That era is over. Generative AI ended it. And no amount of prompt engineering or ATS optimisation is going to bring it back.

What's needed is not better tools for the old process. It's a different process entirely.

But before we get to solutions, it's worth understanding what this storm actually does to the people caught in it. The statistics tell one story. The human cost tells another.

Next in this series: what the broken jobs market is doing to real people — and why the candidates who need the most help are getting the least.