Intentional News

260918 Everyone Using AI To Get Hired Now What

19 min read

Thesis (Alf’s read): AI has not broken recruitment — recruitment was already a funnel that treated people as volume, and AI has industrialised both ends of it at once. Candidates now mass-produce applications and employers mass-filter them, so the signal in an application has collapsed toward zero while the cost of applying has collapsed too. The BBC’s own evidence points away from “AI is ruining the job market” and toward a sharper claim: the process has lost the one thing it needed — a human on the other side — and the losers are the people with least to fall back on. At just under 1 million young people NEET in the UK, hiring practice has stopped being an HR question and become a welfare question.

Everyone’s using AI to get hired. Now what? (BBC News, 18 September 2026)

Summary

The programme is structured as a two-sided investigation: jobseekers on one side, employers on the other, with an AI-in-recruitment consultant in between.

The jobseekers. Three young candidates — Neve, Addie and Santina — have sent roughly 50, 100 and 200+ applications respectively in the last year. Addie describes running an automated agent that scrapes the internet for opening roles and deadlines and narrows the list for her; the others use AI to rewrite CVs so they mirror the keywords in the job advert. All three describe the experience of meeting an AI rather than a person — one recalls an AI interviewer telling her she “lacked empathy”. The reporter’s summary of their feeling: they want to reach a person before anything else, and that is the stage they cannot get to.

The employers. Matt Weston, Senior Managing Director UK & Ireland at Robert Half (the firm places 100,000+ candidates a year globally), says CVs have become “very samey” and that the work is now vetting: verifying real experience rather than reading polished paper — “a CV is essentially a two-dimensional piece of paper”. Kate Young, whose consultancy advises companies using AI recruitment tools, makes the counter-argument: human hiring was always deeply flawed (the firm handshake, the eye contact, the irrelevant interview question), so well-guarded AI could screen out noise and bias — with the emphasis on “if the guard rails are there… it’s a big if.”

The volume numbers. Competition is fierce, vacancies are down, graduates are at a record high, and AI is multiplying applications. The programme’s headline comparison: graduate vacancies averaged 38 applications per job 20 years ago; 140 last year.

The surprise in the middle. Candidates and employers do not agree even on what is happening. One survey finds 75% of candidates believe AI is being used to screen CVs — against 21% of recruiters who say they use it there. The Information Commissioner’s Office position, cited approvingly, is that these tools can let employers “process high volumes of applications consistently and quickly”. The reporter adds the concern that when many employers use similar algorithms, the same candidates get favoured repeatedly — researchers call it algorithmic monoculture — while noting the counterpoint that privileged candidates dominated the best jobs long before any algorithm.

The AI interview. Some processes now open with an AI avatar asking questions and responding to answers; others record fixed answers against a countdown timer. The World Economic Forum is cited as saying AI-led interviews can expand opportunities for overlooked candidates, reduce recruiter bias, increase efficiency and make rejection more constructive; the candidates’ experience in the programme is closer to the opposite.

The conclusion the programme leaves you with. Not “AI ruined hiring” but something more uncomfortable: “You get behind if you don’t use AI today in this market. It’s a must — unfortunately.” One candidate calls it “survival of the fittest”; the reporter’s own line is that human judgment is still an essential part of the process, and that the genie is out of the bottle — the question is not if AI is used but how.

The numbers, checked

Checked against the primary sources rather than the transcript (the broadcast rounds and compresses).

  • 140 applications per graduate vacancy (vs 38 in 2002–03) — Institute of Student Employers, the highest since its records began in 1991. It is not an AI effect: the ratio was already above 100 before ChatGPT. Context in the same data: graduate hires down 8% in 2024/25 with a further 7% fall forecast for 2025/26, and youth unemployment (16–24) around 16.2% — the worst since 2015.
  • 73% of students and graduates used AI at some stage of the application process — Prospects at Jisc Early Careers Survey 2026 (~5,000 respondents), up from 55% the year before. Most common uses: editing CVs and cover letters (57%), interview preparation (47%), writing CVs/cover letters (46%), answering application questions (33%). Applications to 100+ roles rose from 8% to 13%; 20% of recent graduates had applied to more than 100 jobs.
  • The asymmetry — candidates use AI freely, and object to employers doing the same. Same survey: 89% think AI use in live online interviews is unfair, 83% for recorded interviews, 62% for CV screening — while only 38% accept AI screening. This is the most telling finding in the whole story.
  • 75% of candidates believe AI screens CVs; 21% of recruiters say they use it there — CV-Library’s 2026 AI in Recruitment Survey (~500 recruiters, 1,000+ candidates). Same survey: 83% of recruiters use AI to speed up hiring, but only 36% think it actually delivers speed in practice; 53% of candidates believe they have been rejected by AI without human review; 28% are less likely to apply where AI is involved; 40% have abandoned or considered abandoning an application because of it.
  • Employers give almost no guidance on AI use — Ada Lovelace Institute: 45% of UK employers tell candidates nothing at all about AI in applications, and a further 29% discourage it without enforcing anything.
  • AI interviews are now ordinary — Totaljobs / HR Grapevine (May 2026): 47% of UK jobseekers have been interviewed by an AI system; 30% have walked away from a process that included one; 75% want AI use in hiring fully disclosed; 80% oppose AI making the final hiring decision.
  • Where the law sits. EU: AI used to screen, filter, rank or evaluate candidates is high-risk under Annex III 4(a) of the EU AI Act — obligations include risk management, data and bias governance, technical documentation, human oversight (a “rubber-stamp” review explicitly does not satisfy it) and informing candidates. The Digital Omnibus (Regulation (EU) 2026/1744), in force 27 July 2026, deferred the high-risk deadline to 2 December 2027; the Article 50 transparency duty (telling people they are interacting with an AI) has applied since 2 August 2026. UK: no equivalent statute — the ICO’s guidance is the closest thing, which is why the programme can describe the technology as essentially unregulated here.

The Parachute framework, against AI hiring

The owner’s frame — from Richard Nelson Bolles’ What Color Is Your Parachute? — is that every interview is really three questions, and both sides get to ask them:

  • Can you do the job? (capability)
  • Do you want to do the job? (motivation)
  • Do we want you — do you fit with us? (compatibility; Bolles’ own guide puts the candidate’s mirror-question alongside it: do I want to work here?)

Bolles’ point was that the interview is a conversation with two phases on each side — the employer gathers, then sells; the candidate gathers, then sells — and that both sides only “sell” after deciding they are interested. Held against that frame, what this programme documents is not that AI gives bad answers. It is that AI-mediated hiring removes the conversation in which those three questions can be answered at all:

  • “Can you do the job?” is the one question AI genuinely helps with — keyword matching, structured screening, and (done well) consistent criteria applied at volume. This is where Kate Young’s defence lands, and it is a fair defence. The catch: the same tools that assess capability have made the evidence of capability worthless, because every application is now polished by the same class of model. When 140 people submit 140 similarly-optimised applications, the screen stops discriminating between them and starts selecting on whatever noise remains.
  • “Do you want to do the job?” is the question AI hiring is worst placed to answer, and the one it quietly destroys. Motivation shows up in the paragraph someone wrote themselves, the question they ask back, the thing they did unprompted. When a model writes the application and a model reads it, “do you want this” is answered by comparing two generated texts — and the candidate’s actual motive never enters the process at all. That is the “impersonal and transactional” quality the owner names, and it is structural, not stylistic: the medium has no channel for it.
  • “Do we want you?” is where the damage compounds. Fit is a judgement about a person in a team, which is why Bolles frames it as a conversation. An AI screen optimises for similarity to a pattern; research on algorithmic monoculture says the result is the same candidates surfaced everywhere — so “fit” is being decided by which pattern you resemble, not by whether the team would work better with you in it.
  • The candidate’s own question — “do I want to work here?” — has simply been deleted. Almost nobody gets enough contact to form a view before the offer, and the programme shows why: the candidates’ goal has collapsed from “get the job” to “reach a person”.

So the “doom loop” is real, but it is not an AI story. It is a story about a process that had already stopped being a conversation being scaled up. AI did not remove the human from hiring; it removed the last place the human was visible, and made the absence cheap to ignore.

Best practice — a recruitment process that values candidates

The owner’s ask: what should humans actually do? Two lists, because both sides have obligations. The guiding principle is Bolles’ frame: an application process exists to answer three questions honestly — and the cheapest way to answer them is a conversation.

For employers (the ones who hold the power):

  1. Say what you will do with AI — and mean it. The Ada Lovelace finding that 45% of employers give no guidance at all and 29% “discourage without enforcing” is the single cheapest thing to fix. Publish a one-paragraph AI policy: where AI is used (screening, scheduling, interview), what it does not decide, and what you expect candidates to disclose. Silence is what turns a tool into a grievance.
  2. Never let AI make, or effectively make, the rejection. Screening may rank; a human must decide — and must be able to explain the decision. Under the EU AI Act this is a legal requirement from Dec 2027 (human oversight a rubber-stamp does not satisfy); in the UK it is currently just decency. 53% of candidates already believe they have been rejected by a machine, and they are frequently right about the feeling even when the count is wrong.
  3. Tell candidates where they are, every time. The complaint in the programme is not rejection — it is silence. A status page, a “we got it”, an actual rejection with one sentence of reason. This is where AI should be spent: on feeding people back, not on filtering them out.
  4. Shorten the funnel and lengthen the contact. Trial shifts, work samples, a 15-minute human call before any automated test. The six-skills video is a catalogue of things that cannot be assessed on paper: collaboration, creative thinking, communication, organising. The bar manager in that film solves this by putting candidates behind the bar. Every hour of a trial shift buys back more signal than another filter stage.
  5. Do not open with the AI interview. If you must use one, use it late, declare it up front, and never for the stage that tests motivation. A recorded interview against a countdown is the exact place where “do you want the job?” becomes unanswerable.
  6. Measure what your screen is actually doing. Disparate-impact checks on rejection rates by school, postcode, gender, ethnicity and age; keep the risk-management record current. Bias is not removed by having a model — an opinionated pattern learned from historical hires reproduces them.
  7. Value the unusual route in. The programme’s best moment is a politics graduate in the construction industry and a cheerleader-turned-quantity-surveyor. If your screen selects for a pattern, you are filtering out exactly the transfers that make teams better.
  8. Publish outcomes, not vibes. How many applications per hire, how many got feedback, how many were interviewed by a human. Employers measure everything else; measuring their own candidate experience is what stops “we care about people” from being marketing.

For candidates (given the market you actually have, not the one you deserve):

  1. Read the three questions and answer them yourself first. Capability, motivation, fit — and add Bolles’ fourth: do I want to work here? If your application cannot answer the motivation question in your own words, it is a form-filling exercise.
  2. Target density, not volume. 140 applications per vacancy means 140 applications is a lottery ticket. The ISE’s own data shows a large share of roles are hard to fill — those are the ones where being a real person counts. 10 careful, tailored applications beat 100 generated ones.
  3. Use AI as an editor, not an author. The candidates in the programme who sound strongest use it to tighten what they wrote. Keyword-matching without a story is precisely what employers say has become undetectable noise.
  4. Go around the funnel. Referrals, direct approaches, walking in, informational conversations with people doing the job. The FT’s point quoted in the programme — that recruiting is “going retro” because when every application is polished, polish tells you nothing — is a map of where the openings actually are.
  5. Ask for the human, in writing, once. “I’d value 15 minutes with the hiring manager — happy to do any assessment you need.” Sometimes it works; it always distinguishes you from the 139.
  6. Treat rejection as data, not verdict. With these ratios most rejections are statistical. That is not comfort — it is accuracy, and accuracy is what keeps you applying.
  7. Protect your own time. 30% of UK jobseekers have walked away from an AI-interview process. Walking away is a legitimate choice when the process refuses to tell you what it is.

The NEETs and welfare consequence — why this is not just an HR matter. The UK has just under 1 million young people not in education, employment or training, and youth unemployment at 16.2% is the worst since 2015. NEET status is sticky: the longer the gap, the more each application has to explain. A process that returns silence and machine rejections does not merely fail to help — it teaches a cohort that applying is pointless, and the cost of that lesson is paid by the welfare system, by the NHS, and by a labour market that will need these people in ten years. NFER’s own forecast in the companion video is that up to 3 million low- to mid-level jobs could go by 2035 — the same band where entry-level hiring happens. If the entry rung is automated away and the process of reaching it becomes opaque, the welfare consequence is not a risk, it is the arithmetic. Employers’ candidate experience is now a public-policy lever, whether they treat it as one or not.

Why it matters

  • Polycrisis, not a single failure. This sits at the junction of Digital and Information Chaos (an opaque algorithmic layer deciding access to work), Economic Instability (a labour market where entry is hardest for those with least buffer) and the youth/demographic thread. AI is not adding a crisis here; it is a force multiplier on an existing one.
  • The regulatory divergence is now real. The EU treats recruiting AI as high-risk (with the deadline deferred to Dec 2027 and transparency duties already live since Aug 2026); the UK has guidance and no statute. British candidates are the test population.
  • It sharpens the “AI and labour” thread in your digests. Earlier items — UK job postings falling while AI skills hit record highs; AI boot camps; the NEETs crisis — are all the same story from different angles. This one supplies the mechanism: not “AI took the jobs” but “the bridge to the first job got narrower and less human”.
  • And it is a values question the programme mostly avoids. Bolles’ three questions are a test of respect as much as competence. AI hiring answers one of them well and two of them badly, and does so for the people with the least power to insist otherwise.

My assessment (Alf)

The programme’s own evidence refutes its framing. “AI stands accused of ruining the job market” is a hypothesis the BBC then dismantles: applications per vacancy were already 140 (from 38) and graduate hiring was already falling before generative AI arrived, and the gap between candidate belief (75%) and recruiter practice (21%) suggests much of the felt harm is opacity rather than algorithmic rejection. But the honest conclusion is worse than the framing, not better: the volume problem is structural, and AI has removed the human contact that used to make the volume bearable. Employers are not villains here — Matt Weston’s “verify true experience” is the right instinct — but verification is not recruitment, and a market where the only value-add left in hiring is detective work has already given up on the candidates.

The asymmetry finding is the most important thing in the film, and the programme under-plays it. 73% of candidates use AI; 89% think employers using it in live interviews is unfair. That looks like hypocrisy and is not: you can choose your own tools, but you cannot choose your evaluator’s. The complaint is about power, not about automation — and the CV-Library numbers back it (53% believe they were machine-rejected; 40% considered abandoning an application because of it). The fixable part is almost trivially cheap: 45% of employers give candidates no guidance on AI. A published policy costs nothing and would remove a large slice of the grievance.

Where I part company with Kate Young’s defence. She is right that human hiring was always flawed — the handshake story is real — but the conclusion “so AI could be less biased” needs an evidence base that does not yet exist. Guardrails are the whole argument, and the honest position is that we are running the experiment at scale on a cohort that cannot afford to be the test population: employers adopt at 83% while only 36% think it even delivers the speed they wanted. That is not a technology being chosen well; it is a technology being chosen without measurement.

On the Parachute frame, my answer to the owner’s question is blunt: the three questions are still the right test, and AI hiring fails two of them. Capability it can assess (arguably better, arguably worse, always at scale). Motivation and fit are relational questions, and they can only be answered where a person appears. So “the solution” is not to regulate the filter — it is to stop replacing the conversation. The most effective interventions in the programme are the least technological: a trial shift at a bar, a visit to a factory floor, a graduate who found the job by looking at a company’s website and walking in. Every one of them is a human answering the three questions.

What I would not claim. That AI hiring is causing the NEET crisis — the causality runs mainly the other way (thin entry-level demand, automation of junior tasks, a record cohort competing for fewer roles). AI’s contribution is to make the entry point harder to reach and harder to read, and that is enough to matter: it lengthens the dole queue for exactly the group with the weakest fallback. If entry-level work is automated away by 2035 and the process of getting in becomes a machine-to-machine exchange, the welfare bill is where the shortfall lands — and no employer’s cost model includes it. That externality is the argument for policy, and it is the argument the programme never quite makes. I’d also flag the honest uncertainty: none of the surveys here are longitudinal, most are vendor-adjacent or self-reported, and the one genuinely hard number (140:1) is pre-AI. Treat the direction as established and the magnitudes as soft.

Graph View

Backlinks