By Harry Lang, Managing Director, The Oxford AI School
Key takeaways:
- AI has made hiring slower, with 67% of UK hiring managers noting that screening for AI-generated applications is taking longer.
- A human clicking “approve” isn’t true oversight. The ICO found employers using “automated support” tools were often making fully automated decisions without realizing it.
- Train the team. Bullhorn’s GRID data ties AI-ready leadership to a 40% higher chance of 2025 revenue growth.
Imagine opening a shortlist of twenty candidates. Every application is articulate, carefully tailored, and apparently enthusiastic about your organization. Your AI screening tool has ranked them. Their AI assistants have helped them anticipate what it wants. Somewhere in that exchange is the person who can actually do the job. Finding them still requires judgment, and staffing leaders need to decide where that judgment belongs.
Why faster applications are slowing down hiring
That challenge is becoming familiar. Robert Half research published in September found that 67% of UK hiring managers said screening for AI-generated material had increased hiring time. Only 10% of professionals surveyed said they used no AI in their applications. The study covered 500 hiring managers and 1,000 professionals across selected occupational fields. It describes a substantial shift, although it should not be treated as a census of every jobseeker.
For staffing leaders, this creates an awkward productivity question. If candidates can produce applications faster and employers can process them faster, why might hiring take longer? One explanation is that polished presentation has become easier to manufacture, leaving recruiters with more claims to verify. Adding another automated filter may simply move the workload further down the process.
Redesign the assessment before you add another filter
The response starts with assessment design. Consider a hypothetical finance vacancy. A fluent covering letter offers limited evidence that someone can explain a discrepancy in a forecast. A short exercise using fictional figures, followed by a conversation about the candidate’s reasoning, gives the hiring manager something more useful to examine. Keep the task proportionate, offer appropriate adjustments, and avoid asking applicants to perform unpaid work with commercial value.
Be explicit about permitted AI use. Where the role involves working with AI, an assessment might allow it and ask candidates to explain what they accepted, corrected, or rejected. Where unaided competence matters, say so and assess it consistently. A candidate who uses technology to improve their grammar should not automatically be treated as someone who invented their employment history. The relevant question is whether the evidence supports their claimed capability.
Let AI draft the job ad, not vet the candidate
AI can help recruiters gather that evidence. Take a hiring manager’s hurried brief for a customer service supervisor. An approved tool could turn it into a first draft of a job ad and suggest structured interview questions. The recruiter then checks whether the requirements reflect the work, whether the wording creates unnecessary barriers, and whether the questions test agreed criteria. This is useful preparation, provided someone owns the quality of the finished material.
CV summaries need particular care. A system might compress several years of experience into a confident paragraph while losing the distinction between leading a project and supporting it. Ask for each relevant claim to be tied to the original application, with missing evidence identified. Recruiters should verify the summary before relying on it. Otherwise, the next interviewer may assess a machine’s interpretation of the candidate rather than the candidate’s experience.
The compliance blind spot in “human-reviewed” hiring decisions
That distinction becomes more consequential when a tool ranks or rejects people. The ICO’s Recruitment rewired report drew on voluntary engagement with more than thirty employers, rather than a representative survey. It found that many employers using automated recruitment were likely making solely automated decisions, despite believing that people remained involved. The report called for stronger safeguards, clearer candidate information, and better monitoring for fairness.
A human clicking an approval button does not, by itself, resolve that problem. The ICO describes meaningful involvement as having real influence before a decision takes effect, with the competence and authority to change it. Staffing leaders should examine what happens to lower-ranked applicants as closely as successful ones. If managers only consider the top scores, a later interview cannot supply the review missing from an earlier rejection.
Vet AI suppliers like you’d vet a new hire
Procurement therefore needs practical scrutiny. Ask suppliers how their tools were tested, what limitations they disclose, and what evidence supports their claims. Establish where candidate information goes, how long it is retained, and who can access it. The UK’s Responsible AI in Recruitment guide provides a useful basis for assessing suppliers and planning deployment, including accessibility and staff capability.] Bring recruitment, data protection, and IT colleagues into those decisions before teams improvise their own arrangements.
Capability deserves a budget alongside software. Bullhorn’s global survey of nearly 2,300 recruitment professionals found that leaders who felt equipped to guide AI adoption were almost 40% more likely to report revenue growth in 2025. That is an association in a supplier-sponsored survey, not proof that training caused growth. Nevertheless, it gives staffing leaders a reason to examine management readiness alongside technology spend.
Pilot one workflow, measure it, then scale
Useful learning should reflect the actual working week. Let recruiters practice checking a job ad, challenging a misleading summary, and recognizing information that should never enter an unapproved tool. Give managers time to discuss disagreements with automated recommendations. Reward justified challenges; otherwise, a nominal right to override the system may disappear under pressure to clear applications quickly.
Start with one defined workflow and establish a baseline. Measure time saved after checking and corrections, alongside candidate feedback, progression patterns, and the quality of decisions. Agree who can pause the pilot when problems appear. Review results before extending it to another role or team. That includes deciding how applicants can question an outcome and reach someone able to respond. A faster shortlist is worthwhile only if it helps the organization hire well. The strongest use of AI will give recruiters more capacity to investigate uncertainty, speak to people, and explain their decisions. Staffing leaders should make that the test of progress.
Harry Lang is Managing Director of The Oxford AI School which provides practical AI education for businesses and professional teams. Its programs include the recently launched AI for Recruiters training module.
