Ninety-two per cent of major graduate employers say they have suspected or confirmed misconduct in recruitment, up from 85 per cent last year and 78 per cent in 2024. The most common problem is unauthorised use of AI during interviews and online assessments, with two-thirds of employers reporting it in interviews. At the same time, employers are becoming more relaxed about candidates using AI to draft CVs and cover letters.

This is not evidence that an entire generation of graduates is cheating. The Institute of Student Employers survey covered 155 employers and the 92 per cent figure describes employers that suspected or confirmed misconduct, not the proportion of applicants involved. But it does reveal a recruitment system struggling to decide which parts of an application still tell us anything useful about the person making it.

AI has made polished presentation cheap. The sensible response is not to ban useful technology or automate another layer of screening. Employers need clearer rules about where AI assistance is legitimate and more ways for candidates to prove what they can actually do.

More applications are not producing better signals

Competition for graduate roles is already intense. ISE reports an average of 141 applications for every graduate vacancy and 96 for every apprenticeship. Graduate recruitment itself rose by four per cent this year after falling eight per cent last year, so this is not a story of AI having eliminated graduate jobs. The immediate disruption is happening in the recruitment process.

AI makes it easier to tailor an application to a job description, rewrite a covering letter and apply to more employers. That can be genuinely useful. Someone with strong experience but weak writing skills should not be penalised because another candidate can afford professional help or happens to know the conventions of graduate recruitment.

The difficulty is that the same technology reduces the information contained in the application.

A carefully written covering letter once told an employer something about how a candidate communicated, what they had noticed about the organisation and why they believed their experience was relevant. If software can produce a convincing version in seconds, the document may still contain useful facts but its polish tells us much less about the applicant.

Volume compounds the problem. Candidates face fierce competition and have an incentive to submit more applications. Employers receive more apparently credible candidates and have an incentive to screen more aggressively. Each side then uses more technology to cope with the behaviour of the other.

Having spent much of my career in recruitment, I recognise the logic. When a recruiter has hundreds of applications for a handful of jobs, filtering is unavoidable. But a process can become very efficient at rejecting people without becoming any better at identifying who will perform well.

Clear rules matter more than blanket bans

The line between legitimate AI assistance and misrepresentation is not especially mysterious.

Using a tool to check spelling, improve structure or help somebody express genuine experience more clearly is different from using it secretly to generate answers during an interview or complete an assessment intended to measure the candidate's own capability.

Employers increasingly recognise that distinction. ISE says 64 per cent now accept AI use somewhere in recruitment, up from 46 per cent last year, particularly for CVs and covering letters. Meanwhile 82 per cent provide candidates with guidance about acceptable use, up from 55 per cent.

That is preferable to pretending AI can simply be prohibited.

It also places responsibility on employers. A candidate should know before each stage whether AI is permitted, what kind of assistance is acceptable and what must be completed independently. If the rule changes between the written application, assessment and interview, say so explicitly.

Candidates then have a straightforward obligation to follow those rules and represent their abilities honestly.

Employers should apply the same standard to themselves. Only 16 per cent of employers in the ISE survey currently use AI in recruitment, although 29 per cent plan to introduce it within the next year. Where an organisation uses automated screening while imposing an absolute ban on candidate assistance, it should be able to explain why those two uses are materially different.

Perception already matters. Separate ISE research found that 95 per cent of candidates believed employers use AI to shortlist applicants, while 73 per cent said they had used AI somewhere in their own application process. Candidates who believed employers always used AI were substantially more likely to use it themselves.

That is how an arms race becomes self-reinforcing.

Stop asking polished applications to do a job they can no longer do

Employers are already changing their processes. Seventy-seven per cent told ISE they were reviewing or redesigning recruitment, using approaches including more personal questions, time limits and randomised assessments. Some are also using automated eligibility screening or closing application windows early to manage volume.

The danger is responding to AI-assisted applications by adding more automated filters to the front of the same process.

If the problem is that written presentation has become less informative, the answer is to place less weight on presentation and more on evidence of capability.

That could mean structured interviews in which answers are probed rather than scored for fluency, short work samples, practical exercises and assessments that require candidates to explain their reasoning. For roles in which employees will genuinely use AI, employers need not pretend otherwise. Candidates can be allowed to use the tools they would have at work, then asked to explain what they accepted, what they rejected and how they checked the result.

That tests judgement rather than typing speed.

It also gives less polished candidates a fairer opportunity. Graduate recruitment has long risked rewarding those who have learned how to perform recruitment itself: the right covering-letter language, competency examples and interview formulas. Generative AI has not created that weakness. It has made it harder to ignore.

Skills policy should reward capability, not application technique

Universities and careers services have a role here too. Teaching students how to produce an immaculate CV is less useful if every applicant can produce one. Young people need help identifying evidence of what they have done, explaining decisions they have made and demonstrating skills in ways that survive scrutiny.

Government should take the same lesson into skills policy. AI literacy matters, but it should not become synonymous with knowing how to generate more applications or better prose. The useful skill is knowing how to use technology while remaining responsible for the quality of the work.

There is a reasonable objection to all this. Candidates are responding rationally to a difficult market. Someone applying for dozens of roles cannot spend hours crafting every covering letter and employers themselves have spent years automating recruitment. Asking applicants alone to return to some pre-AI ideal would be both unrealistic and unfair.

They should not have to.

The principle is simpler. Use AI where it improves legitimate work. Be clear about where assistance ends and assessment begins. Then design recruitment around evidence that remains informative.

The CV will not disappear. It is still a useful summary of someone's experience. But employers should stop asking polish to stand in for potential.

AI has made it easier than ever to look like the right candidate. Recruitment now needs to get better at finding out who can actually do the job.