AI-generated CVs are slowing UK hiring

Two thirds of UK hiring managers say that screening for AI-generated applications has made hiring slower. Sixty-seven per cent, in a Robert Half survey of 500 UK hiring managers and 1,000 professionals, fielded in June 2026 and published on 14 September. Only 10 per cent of the professionals surveyed use no AI at any point in applying for a job.

Every factor on the list is a checking cost

Robert Half names four things making candidate validation harder:

• managing an increased volume of applications from unqualified candidates, 28 per cent
• distinguishing between AI-generated and authentic CVs, 27 per cent
• needing additional in-person interviews or screening to validate skills and experience, 23 per cent
• spending more time verifying candidate credentials, 23 per cent

Three of those are obviously verification: is this document real, are the credentials real, can this person actually do the work. The fourth looks like a volume problem and is not, because you cannot know a candidate is unqualified until you have assessed them. The cost sits in the assessing, not the arriving.

That is the finding worth carrying out of this research, and it is not that employers are getting more applications. It is that every item on the list is the cost of establishing whether what you are holding is true, and better sorting touches none of them. Each one needs a person to go and find something out, and people are the expensive part.

A CV was never evidence. It was a claim, and it worked as a filter because producing a polished, closely tailored claim used to take real effort, and that effort was a rough signal of how much somebody wanted the job. AI removed the effort and kept the polish.

The work moved, it did not disappear

A candidate who uses AI to tailor an application saves time. The hiring manager who cannot tell that application from the next one spends it. Both are real, they land in different places, and that is why nobody adds them up. The person doing the saving does not work for the organisation doing the spending.

Inside an organisation the same thing happens between departments and between grades. There, it should be somebody's job to notice.

Your usage dashboard cannot see this

The research covers external applicants rather than internal AI use, so what follows is my reading rather than their finding.

A lot of organisations are holding two numbers that do not speak to each other: usage reporting that looks healthy, and processes that have got slower. Both are true, and neither explains the other, because usage reporting records whether somebody opened the tool and cannot see where the work went next.

So ask a different question. Who is now checking output they did not produce? In the rollouts we have worked on, the licences go to the widest group available and the checking settles on a function nobody consulted, usually one that was already busy. It is rarely in anybody's job description, and that person rarely has the authority to send something back.

Then disclosure, which is the cheapest fix available and the one most organisations skip. Twenty-seven per cent of those hiring managers are spending time telling AI-generated material from authentic material, yet very few application processes ever ask candidates to say.

If the advert and the application form say nothing about AI, applicants are left to guess what is acceptable. A short statement changes that: we know many of you will use AI to help with your application, we welcome people who use it well, and at interview you may be asked how you used it and what you checked. It tells candidates what good looks like and gives whoever is screening something to work with. If an organisation has not said how it wants people to apply, or asked for the information it needs to decide who goes forward, it can hardly complain about what arrives. The same applies inside your own organisation: you can simply ask people to say so.

That only works where saying so is safe. Somebody who thinks admitting to AI use will be read as cutting corners, or as evidence their role could go, will keep quiet, and every colleague downstream pays the checking cost instead. No prompt course touches that. It is a question about culture and climate.

The tempting move, when hiring slows, is to speed it back up with another tool. Automating a bad process makes inefficiency faster. An AI screen in front of a stack of AI-written applications does not restore the signal, it moves the decision further from anyone who could interrogate it.

Three things worth doing this quarter

Find where the checking landed. Pick two or three processes that now take in AI-assisted material. Ask who verifies it, how long it takes them, and whether anybody decided they should be the one doing it. The answer to the last question is usually no.

Make disclosure normal and specific. Not a policy saying AI use must be declared, which produces nothing, but a working habit of saying which part was AI-assisted and what the person did to check it. Leaders go first, on their own work, including the drafts that went wrong. Do the same at the front door: say on the job advert that you expect applicants may use AI and welcome those who use it well, then ask at interview how they used it and what they checked.

Measure the conditions, not the usage. Before spending anything more on tools or training, get an honest read on whether the structures, the climate, the perceived usefulness, the sense of threat and the actual capability in your organisation support what you are asking people to do.

uptakeAI is an organisational psychology practice helping organisations adopt AI in a way that is human centred, practical and sustainable. The work starts with PRISM, our psychometrically enabled diagnostic of AI adoption readiness, which measures the conditions that decide whether people use the AI tools they already have.

Installing the tools is the easy part. The cost in this research is not being paid by the technology. It is being paid by the people who now have to check everything it produces, and until somebody names them, nobody is going to help them.

Source: Robert Half, survey developed by Robert Half and conducted by an independent research firm in June 2026 among 500 UK hiring managers and 1,000 UK professionals across finance and accounting, technology and IT, marketing and creative, legal, administrative and customer support, and HR. Published 14 September 2026.