For the first time in its ten-year history, Eploy's 2026 UK Candidate Attraction Report asked in-house recruiters about candidates using AI to apply. It went straight into the top three sourcing challenges, cited by almost a third of the 700+ recruiters surveyed.

That was the topic of the second session in our webinar series with Eploy, with Charlotte Clark, Account Executive at Eploy, and Greg Dunbar, our Chief Commercial Officer.

‍

Charlotte opened with a question most of us skip. Is a CV that wasn't written entirely by the candidate actually new?

"I remember my first CV. I asked a friend of a friend who worked in HR to give me tips. There are CV consultants who take a fee. What's changed is that AI has made that polishing free, scalable, and instant. So the real question isn't why the candidate is using AI. It's why that polished document ever became a gatekeeper at all."

Greg agreed on the history but emphasized the difference is now the scale. Getting help was always available; getting it infinitely, at zero cost, for every job, while you sleep, is a different market. "We're approaching a world, if we're not already there, where AI is the norm, and if you're not doing it you're at a disadvantage." That structural shift, he stressed, has happened. For talent teams doing nothing is not an option.

‍

As proven in Daniel Pink's Drive: for tasks requiring judgment, external rewards tend to narrow focus and make performance worse. Hiring, Charlotte argued, has become exactly that kind of carrot-and-stick task.

"CV applications are a narrowing structure that rewards machines that can tick boxes, and excludes candidates demonstrating the nuanced skills around motivation and initiative. I don't think candidates are cheating. This is proof the system is rewarding the wrong behavior."

We should not be shocked that a machine got very good at chasing the carrot we hung out.

‍

As the moderator I put the blunt question to the panel: detect it and block it? Both said no, for two reasons.

The first is practical. Charlotte: "You can't say no to AI. The toothpaste is out of the tube. You can't untoast the bread." The second took the strategic, long-term vision - the candidates trying new tools before their employer forces them to are curious and show initiative. A no-AI policy will screen out precisely the trait most teams say they are hiring for.

Greg explained how we handle this at Hubert. HubertDetect looks at signals in a candidate's responses and flags likely AI use to the recruiter, with the context of exactly where it appeared. Candidates aren't stopped from using AI but it will be available for human review by the recruiter. The recruiter will decide what that information means for that role. "This is the billion-dollar question in the industry right now. You can't block it. You've got to redesign your process."

‍

The instinctive response to a flood of perfect applications is stricter sifting: tighter filters, application caps and more hurdles for the candidate. But this only treats the symptom. Charlotte's goal is "not stricter sifting out; it's learning how to sift in, aligned with what both good candidates and good recruiters care about." Candidates want a chance to demonstrate they can do the job. Recruiters want the same evidence - we are on the same team.

That is what a structured, skills-based interview at the point of application does. Instead of scoring a document, Hubert gives every candidate the same competency-based conversation, by chat or voice, in their own time, with feedback within minutes and all directly within their ATS, including Eploy. It looks forward at what someone can do rather than backward at what a CV claims.

‍

Greg was equally direct about the industry's own failures. Much of the backlash against AI in hiring is earned: too many tools rushed to market with a large language model doing the scoring. Probabilistic, inconsistent, unexplainable black-box large language models potentially deciding a candidates future.

Hubert is built differently. Our assessment layer runs on deterministic AI models: same input, same output, full explainability. If a candidate gave the same answers 100 times, they would get the same score and ranking 100 times. That is what makes a shortlist legally defensible when a candidate or regulator challenges it, and it means no personal data goes anywhere near a language model that could train on it or leave the EU.

We ran out of time for the other hard question: what does a *meaningful* human-in-the-loop look like, when a recruiter who rubber-stamps an AI recommendation is arguably a proxy for an automated decision? Greg promised a proper answer. We'll holding him to it - join the rest of the series to hear more here.

‍