AI is interviewing you — and employers don't trust what AI sends them back

Nearly half of UK jobseekers — 47% — have been interviewed by an artificial intelligence system, and 82% of those were never told beforehand that a machine would be running the conversation [1]. Cross to the other side of the desk and the picture is almost symmetrical: 73% of students and graduates now use AI somewhere in their job applications, up from 55% a year earlier [2], while 67% of early-careers employers believe graduates and apprentices are using AI to misrepresent their skills [4]. British hiring in 2026 has automated itself into a standoff. Both sides are using the machines; neither trusts what the other side's machine produces.

The consequences have moved beyond awkwardness. Since 5 February 2026 a new legal regime for automated hiring decisions has been in force [5]. The Information Commissioner's Office has found employers making solely automated decisions about candidates without realising it [7], and final, tougher guidance is expected later this year [6]. And at the highest-stakes end of selection — surgical training, accountancy qualifications — institutions have begun pulling assessment back into physical rooms because they no longer believe what remote processes tell them [9][10]. That last development is the one to watch, because it signals where verification is heading for everyone else.

How many candidates now use AI to apply?

The cleanest UK measure comes from Prospects at Jisc's Early Careers Survey 2026, which questioned 4,990 pupils, students, apprentices, graduates and young people not in education or work between 8 January and 9 February, publishing on 8 June [3]. It found 73% of students and graduates used AI at some stage of the application process, up from 55% the year before [2]. The breakdown matters: 57% used it to edit CVs and cover letters, 47% for interview preparation, 46% to write CVs or cover letters outright, and 33% to answer application questions [2].

AI is also inflating volume. One in five recent graduates in the survey had applied for more than 100 jobs, and one employer reported receiving 2,000 applications for a single graduate role [2]. When generating an application costs minutes, candidates rationally produce more of them — and every written application becomes weaker evidence of anything.

The same respondents, though, do not extend to employers the licence they claim for themselves. Asked about AI on the assessing side, 89% said an employer using AI to conduct live online interviews would be unfair; 83% said the same of recorded interviews, and 62% of CV screening [2]. This is the fairness paradox at the centre of the 2026 hiring market: near-universal adoption by applicants, near-universal objection when the tables turn. It is not a confident cohort either — 35% believe AI has reduced their career opportunities, and 41% doubt they can build the AI skills they will need [3].

Do employers believe what lands in their inbox?

Increasingly, no. The Institute of Student Employers' Student Development Survey 2026 — 144 employers, fieldwork in January and February, published 7 May — found 67% believe graduates and apprentices are using AI to misrepresent their skills, up from roughly half a year earlier [4]. That is a belief measure, not a detection rate; no one has audited how much misrepresentation actually occurs. But in hiring, belief is what drives behaviour, and a jump of roughly 17 points in a year means employers are discounting the written and recorded signals in front of them.

What the same dataset does not support is the popular claim that AI is deleting entry-level work. Among ISE employers, 87% expect AI to reshape graduate and apprentice roles — but 58% foresee only minor adjustment, and just 29% expect significant change, concentrated in FMCG and tourism (56%) and legal (41%) [4]. Two in five employers expect to replace no entry-level roles at all with AI, and a further 42% expect it to affect no more than a tenth of them; 43% say roles have already quietly evolved without formal redesign [4]. Reshaped, not replaced, is the measured picture on these employers' own forecasts. What is eroding fastest is not their appetite for early-careers hires but their trust in assessing those hires remotely.

The 2026 trust gap, in numbers

How often are candidates interviewed by AI — and are they told?

Research commissioned by Greenhouse, the hiring-platform vendor, and reported by People Management on 30 April puts the AI interview squarely into the mainstream: 47% of UK jobseekers say they have been interviewed by an AI system [1]. The finding that should worry employers is disclosure. Some 82% were not told beforehand that AI would be involved, and 24% only realised once the interview had begun [1].

Candidates are voting with their feet: 30% have abandoned a hiring process on discovering an AI-led interview [1]. On those figures, an employer running undisclosed AI interviews across a 200-candidate shortlist should expect roughly 60 withdrawals — our analysis, and a directional one, since the survey's UK sample size was not stated in coverage and the research comes from a vendor with a stake in the market. Treated cautiously, the numbers still point one way: 59% of respondents want disclosure to be a legal requirement, 45% want the option of a human interviewer, and 28% want evidence of bias audits [1]. Perceived bias is not marginal — 27% reported age bias and 17% race or ethnicity bias in AI interviews, and only 10% said the employer had a clear AI policy [1]. Greenhouse's chief executive, Daniel Chait, warned that AI is being layered onto a hiring process that was already broken, with the risk of making a bad system worse [1].

What does UK law now say about automated hiring decisions?

The legal ground shifted on 5 February 2026, when section 80 of the Data (Use and Access) Act 2025 came into force through the Commencement No. 6 Regulations [5]. The old UK GDPR position — a prohibition on solely automated significant decisions, subject to exceptions — has become a right of challenge with safeguards [6]. Employers may now rely on any ordinary lawful basis, including legitimate interests, to run significant automated decisions on ordinary personal data, provided candidates are told, can challenge the outcome and can obtain human review; the Act's new "recognised legitimate interest" basis cannot be used for automated decision-making [6].

Can employers use AI to reject candidates in the UK?

In more circumstances than before, yes — which is precisely why scrutiny is rising rather than falling. On 31 March the ICO published its Recruitment Rewired report on automated decision-making in hiring, built on engagement with more than 30 employers between March 2025 and January 2026 [7], and opened a consultation on tougher draft guidance that closed on 29 May, with final updated guidance — including recruitment-specific material — expected later in 2026 [6]. Its headline finding: many employers believed their screening tools were merely decision support, when in practice several were making solely automated decisions about candidates without the required safeguards [7]. Once the final guidance lands, an employer arguing it had not realised its process was automated will be arguing against a regulator that has already published the opposite finding.

The regulator's expectations, as set out in analyses of the report, are concrete. Human involvement must be meaningful and active — in the ICO's words, not "a token gesture or a rubber stamp" — and the fact that a human designed or built the system does not count [6]. Candidates must be told when automated decision-making is used, how it works, and how to challenge a decision and obtain human review; employers need a valid lawful basis, data protection impact assessments and regular bias testing, with the report suggesting routine bias reviews and consultation with vendors on their testing protocols [8].

What rights do you have if an algorithm rejects you?

Three, in practical terms: to be told automated decision-making is being used, to challenge the decision, and to obtain a review by a human whose involvement is real rather than ceremonial [6][8]. For candidates, that means a rejection from a process you suspect was automated is no longer necessarily the end of the road — a short written request asking whether the decision was solely automated, and if so for human review, now stands on statutory ground. For employers, the exposure runs the other way: the Greenhouse disclosure figures suggest a large share of current AI interviewing would fail the transparency expectations the ICO has set out [1][6]. None of this is legal advice — employers making decisions about automated screening should take their own counsel — but the direction of regulation is not ambiguous.

Why is high-stakes selection going back into a room?

While most employers wrestle with disclosure policies, the institutions with the most to lose have reached a starker conclusion: stop assessing remotely at all.

In January, the Confederation of British Surgery publicly backed a return to face-to-face interviews for surgical training recruitment, after warnings that so-called ambient AI — background software that displays suggested answers on a candidate's screen in real time, requiring no typing and minimal eye movement — is close to undetectable in the virtual interviews adopted during the pandemic [9]. The Four Nation Postgraduate National Recruitment Programme Board has banned AI technologies in online interviews outright, from generated responses to live transcription tools and automated scripts [9]. The same report cited Google's disclosure that more than half of candidates used AI tools during virtual coding-challenge interviews [9].

ACCA, the global accountancy body with more than 500,000 students and 257,900 members, reached the same verdict by another route: in a decision announced in late 2025 and taking effect from March 2026, it has withdrawn most remote examinations and returned session exams to physical test centres, citing rising misconduct and AI tools that remote supervision cannot reliably defeat [10].

These are not nostalgic gestures. They are risk decisions by gatekeepers who decide who operates on patients and who signs company accounts — and their common conclusion is that the only assessment they still fully trust is the one conducted in a room with the candidate. When the professional bodies with the strongest incentives to get selection right retreat to in-person verification, it tells ordinary employers where the integrity layer of hiring now sits.

What should employers do before the ICO's final guidance lands?

The compliance list writes itself from the ICO's expectations: a data protection impact assessment for any automated screening; a documented lawful basis; a candidate-facing notice that says plainly where AI is used; a working route to human review that is more than a rubber stamp; and a bias-testing log, informed by your vendor's own protocols [6][7][8]. Employers who cannot produce those five documents are carrying exactly the exposure the ICO found when it went looking [7].

But the sharper opportunity is competitive rather than defensive. In a market where 82% of AI-interviewed candidates were never told [1] and 89% of early-careers candidates consider live AI interviews unfair [2], a truthful claim that a human will meet, interview and answer every applicant is a differentiator — one that costs process design, not technology spend. Employers building autumn volume-hiring pipelines can put that claim to work directly: at the London Careers Fair on 6 November, every conversation is a disclosed, human, two-way assessment by definition.

What should candidates do differently this autumn?

Use AI the way the data says survivors use it: to edit and prepare, not to fabricate. The 46% using AI to write CVs or cover letters outright [2] are feeding a channel two-thirds of employers already suspect of overstating skills [4] — and the professional bodies' crackdown shows where verification is heading: live, in person, unassisted [9][10]. The asymmetry to exploit is that a face-to-face conversation is now the highest-signal channel left, precisely because it cannot be ghost-written. Ask, when invited to interview, whether AI will be used and how; if an automated process rejects you, request human review — the safeguards in force since February exist to be used [5][6].

And rebalance your channels. When a single graduate role can attract 2,000 applications [2] — in the thinnest advertised graduate market on record — the marginal hour spent on application number 101 is worth less than an hour spent where employers are physically present and answering questions. Roles on our jobs board include listings from employers who exhibit at events around the country, and the application black hole that drives so much jobseeker exhaustion has no equivalent in a hall where every employer has paid to stand in front of you.

Hiring's trust problem will not be solved by better prompts on either side of the desk. It is being solved, where it is being solved at all, by people meeting each other. The autumn careers fair season runs from late August to early November across the UK — entry is free for candidates, and every employer in the room is there to answer questions in person.

Sources