AI in recruitment: what actually works for UK agencies in 2026

Morgan McKinley's 2026 Global AI Report found 55% of UK employers don't use AI in recruitment at all, more than in any other market it surveyed. Plenty of agencies have tried a tool or two, usually for adverts, without seeing much change in how the desk runs. Meanwhile the rules have moved: UK data protection law changed in February 2026 to allow more automated decision-making, with conditions attached that many agencies using AI screening tools don't yet meet. This post covers which AI recruitment tools are worth using in a UK agency and what to check before you let software decide who gets a call back.

Where AI in recruitment pays off

The tasks where AI pays off are repetitive and low-risk, and a consultant checks the output before anyone else sees it.

• Advert drafting. This is where most agencies start. Give a tool the job spec, the client's notes and your house style, and you get a usable first draft in seconds. It's also good at stripping out jargon, gendered wording and the 15-bullet requirement lists that put candidates off. A consultant still checks the salary, location and anything the client said off the record.
• Search and matching inside your own database. Most agencies are sitting on years of candidates they never look at because keyword search in the CRM is poor. AI search that understands 'forklift counterbalance, nights, within 10 miles of Warrington' and surfaces people you placed three years ago is often worth more than any new sourcing channel.
• Notes and admin. Call summaries, interview write-ups, turning a messy client brief into a structured job record, drafting a shortlist email. This is time consultants hate spending and the risk is low, because they read it before it's saved or sent.
• Candidate messages out of hours. Answering 'is this role still open?' or 'what are the shift times?' at 9pm, booking a call, and chasing missing right-to-work documents. Candidates in temp markets often apply in the evening, and the agency that replies first tends to get the worker.
• Reporting. Pulling fill rates, time-to-fill and margin by client into a readable summary for a Monday meeting.

None of these leave a decision about a candidate to the software, which is why the risk is low.

Screening is where the law applies

CV screening and candidate ranking are the uses most vendors lead with, and they're the ones that need the most care, because the output is a decision about who progresses.

The Data (Use and Access) Act 2025 changed the UK's automated decision-making rules from 5 February 2026. Before, solely automated decisions with a significant effect on someone (rejecting a job applicant counts) were broadly prohibited unless narrow conditions applied. Now they're allowed on wider grounds, including legitimate interests, provided you put safeguards in place: you tell candidates automation is being used, give them a way to challenge the decision and ask for a person to review it, and test the tool for bias. The old, stricter rules still apply where the decision uses special category data, such as health or ethnicity.

In March 2026 the ICO published Recruitment rewired, a report on automated decisions in recruitment based on its work with more than 30 employers. Its main finding was that many didn't realise they were making automated decisions at all, so none of the safeguards were in place. The ICO was clear that a person clicking 'approve' on a list the software produced doesn't count as human involvement. To count, the reviewer needs the authority and the information to overrule the tool, and has to actually use them. Its updated guidance on automated decision-making is due this winter.

The ICO's earlier audit of AI recruitment tool providers (November 2024) is worth reading before you buy anything. It made almost 300 recommendations, and found some tools inferring candidates' gender and ethnicity from their names, and others collecting more data than they needed and keeping it indefinitely.

There's also the Equality Act 2010. If a screening tool filters out older candidates, women returning from career breaks or people whose CVs follow a non-UK format, that's indirect discrimination, and the agency using the tool is liable for it. Your contract with the vendor doesn't move that liability.

In practice, for an agency:

1. Know which of your tools make or shape decisions about candidates. Ask vendors directly.
2. Keep a consultant in each rejection decision, with time to look at the CV, or treat the process as automated and apply the safeguards above.
3. Tell candidates, in your privacy notice and at the point they apply, where AI is used.
4. Ask vendors for bias testing results and what data the model was trained on. If they can't answer, don't use it for screening.
5. Carry out a data protection impact assessment before you switch it on.

If you place candidates with clients in the EU, the EU AI Act classes recruitment and screening tools as high-risk. Those obligations were due in August 2026 but have been pushed back to 2 December 2027, so there's time, but they will apply.

Where the consultant still wins

Some parts of agency work don't improve with AI, and a few get worse.

Client relationships and negotiation. Winning a PSL place, holding a fee, talking a client round on a rate that won't attract anyone, knowing a hiring manager will say yes to the second-best candidate if you call before 10. AI tools don't help with any of it.

Judging motivation and fit. A five-minute call tells you more than a CV about whether someone will turn up on Monday, and on a temp desk that matters more than any ranking score.

Safeguarding and compliance. In care, education and other work with vulnerable people, the Conduct Regulations put the duty to check qualifications, references and suitability on the agency. AI can chase the documents, but a trained person has to decide whether the checks are good enough.

Saying no well. Candidates remember how they were rejected. A short, human message after an interview does more for your reputation than a fast automated one, and in a small specialist market you'll often want to place the same person later.

Candidates are using AI too

Applications are now written with the same tools. The same Morgan McKinley report found nearly a third of employers (29%) struggle to verify candidates' real skills at the first screening stage, as AI-written applications flood their inboxes. Every CV now reads well, so CV quality tells you less than it used to.

The fix is to rely less on the CV. Add two or three role-specific questions to the application form that are hard to answer generically, do a short structured phone screen sooner, and use practical tests where the role allows. AI-detection tools aren't reliable enough to reject people on.

Candidates are also wary. The same report found 28% of UK candidates are less likely to apply to an employer that uses AI screening. Being open about where AI is used and where a person decides helps.

Why most agencies don't see a return

When we look at agencies whose AI tools haven't changed much, the same two causes come up.

The first is data. AI matching can't find candidates whose records are incomplete, duplicated or split between the CRM, a website database and a spreadsheet. If your website, job board and CRM don't share one candidate record, that's the first thing to fix, and it pays off with or without AI.

The second is training. Buying a licence and leaving consultants to work it out produces a handful of drafted adverts and little else.

A lot of what gets sold as AI is ordinary automation that doesn't need a language model. Shift confirmation texts, timesheet reminders, alerts when a worker's DBS is about to expire, notifying a client when a booking is filled: these are simple rules, and they often save more hours a week than any AI feature, with nothing to go wrong in front of a candidate.

A sensible way to start

• Pick one high-volume, low-risk task (advert drafting or call notes) and do it well before adding more.
• Clean up and connect your candidate data first.
• Train consultants on the tool and on what it mustn't be used for.
• Keep people in charge of decisions about candidates, and document how.
• Review every tool that touches screening against the ICO's expectations and the Equality Act.
• Measure the result: hours saved, time-to-fill, response times. Drop what doesn't move them.

Build it into the systems you already use

AI tools work best when they sit on top of clean, connected systems: a website that feeds applications straight into your CRM, a job board that shares the same records, and temp workflows where reminders and confirmations run themselves. That's the work we do at Nodex, and it's what Stafr does for temp and contract desks, with shift booking, timesheets, geofenced clock-in and payroll export in one place. If you want to automate the repetitive parts of your agency without handing decisions about candidates to software, [talk to us about automation for your agency].