Onrec Explores / Talent Acquisition

When every application looks perfect, how do you find the real talent?

AI can help candidates write stronger applications and prepare convincing answers. For recruiters, the challenge is deciding what those answers actually prove—and how to assess ability without making every applicant feel under suspicion.

Onrec Explores · October 2026 · Analysis

Industry perspectives

Matthew BurneySenior Strategic AdvisorIndeedJames MaliaPresident UK & EuropePrezzeeSanjoe Tom JoseCEOTalviewMatthew JefferyEditor of Better HirerBetter Hiring InstituteVedran ObradovićFounder & Brand DesignerVizura Studio

The application ticks every box. The covering letter reflects the job description. The interview answer follows a neat structure, with a clear challenge, a decisive action and an impressive result.

But what has the employer learned about the person behind it?

It is a question that feels increasingly familiar to James Malia, President UK & Europe at Prezzee. He describes CVs and interview preparation becoming more alike, and recalls seeing a candidate use AI during a video interview.

“Candidates can and will research the business and the role using AI, as you would expect, but you want the output to be them, not AI,” he says.

Candidates have always sought help presenting themselves, from careers advisers, friends and professional CV writers. Generative AI makes that help available instantly. A clearer account of someone’s experience may reveal talent that an awkwardly written application would have obscured. A convincing account of experience they do not have can conceal its absence.

For recruiters, the challenge is to build a process that can tell the difference.

What does a good application actually prove?

Morgan McKinley’s 2026 Global AI Report illustrates the pressure. Its published summary reports that 78% of job seekers use generative AI in applications, while 29% of employers struggle to verify genuine candidate skills during initial screening. The research surveyed 1,962 employees and 144 employers across seven markets; those findings should be read in that context rather than treated as a measure of every recruitment process. Read the research and methodology.

The immediate temptation may be to look harder for signs of AI. A more useful starting point is to ask what each stage of recruitment is supposed to establish. Persuasive writing may matter for one role. Diagnosing an equipment fault, managing a shift or handling a difficult customer may matter much more for another.

Matthew Burney, Senior Strategic Advisor at Indeed, argues that this makes skills-first hiring increasingly important. Qualifications, previous job titles and polished applications offer only part of the picture.

“Businesses still need people who can ask the right questions, apply judgement, understand context and make decisions,” he says.

That shifts the emphasis from how impressive an answer sounds to the evidence behind it. A CV can start the conversation, but the later stages need to give candidates a fair opportunity to demonstrate what they know.

Draw the line before candidates apply

Employers also need to decide which kinds of help they will accept. A blanket instruction that candidates must not use AI leaves plenty unresolved: does it include spelling tools, help structuring a covering letter, interview preparation or generating answers during a live assessment?

For Matthew Jeffery, Editor of Better Hirer at the Better Hiring Institute, the essential distinction is whether the underlying claim is true.

“Using technology to improve the presentation of your genuine experience is perfectly reasonable. Inventing qualifications, achievements or employment history is not. The dividing line should be truthfulness, not whether someone used ChatGPT.”

A practical policy can make that distinction explicit: assistance with presentation may be acceptable, while every claim about experience and qualifications must remain accurate. Individual assessments may have different rules depending on the skill being tested. Candidates should know those rules before they begin, with a clear route to discuss accessibility needs.

For roles where AI is already part of the work, an assessment may be more revealing when candidates are allowed to use it. Burney’s question is how they take responsibility for the result.

“The question shouldn’t simply be whether a candidate used AI to reach an answer, but how they used it.”

Did they challenge the output, spot mistakes or apply their own knowledge? Those follow-up questions turn a discussion about tool use into one about judgement. The CIPD’s discussion of AI in applications similarly highlights transparency and the need to probe underlying skills.

Make room for the person behind the answer

One way to do that is to ask candidates to show their working. Consider a short, fictional customer-service exercise: an order is late, the available information conflicts and the customer wants a resolution. The candidate prepares a response, explains which facts they would check and describes when they would escalate the problem. Then the interviewer changes one detail: the proposed replacement is unavailable.

The original answer becomes one piece of evidence alongside the candidate’s questions, reasoning and ability to adapt. The same approach can be applied to reviewing code, prioritising a workload or finding weaknesses in a sample project plan.

Sanjoe Tom Jose, CEO at Talview, suggests a related exercise: give candidates an AI-generated answer containing flaws, then ask them to critique it and explain their reasoning live.

“You learn whether they can tell when the AI is wrong, which is the skill that matters now,” he says. He also advocates giving everyone the same hosted AI assistant so access to different tools does not determine the result.

The exercise still needs to be brief, relevant and assessed consistently. The CIPD’s selection guidance identifies job-relevant skills assessments and structured interviews as methods employers can use to evaluate candidates. An elaborate task is of little help if the hiring team cannot explain what it measures.

Malia’s preference for direct conversation, and face-to-face meetings where possible, reflects the same desire to get beyond a prepared script. He values confidence and potential, but that raises a further question for employers: how will they recognise ability in someone who is less comfortable in an interview?

Instinct can prompt a useful follow-up. It should not replace job-related evidence, and the format of the conversation should take candidates’ accessibility needs into account.

Check the claim, then consider the context

Demonstrating capability does not resolve every concern. An employer may also need to verify qualifications, employment history or whether the person taking an assessment is the person applying. These are different questions, and each check should have a clear purpose.

Jeffery argues that employers must examine their own role in creating processes that reward keywords and polished performances. Practical assessments, identity verification and proportionate fraud checks can all help, he says, but adding friction indiscriminately risks driving genuine applicants away.

Suppliers are responding with tools aimed at candidate fraud and verification. Hirevue describes measures to mitigate cheating and fraud, while Talview markets checks across the hiring journey. Those are suppliers’ descriptions of their products, rather than independent evidence of how well they will work in a particular employer’s process.

For buyers, the important questions concern what happens when a system flags something: what has it actually established, what else might explain the signal, and who reviews the evidence?

Jose points to Talview’s work with Persistent Systems. Of 78 interviews selected for a live second-camera check, fraud was confirmed in 18. That describes the selected interviews; it is not a measure of fraud across all candidates.

“Where technology gets it wrong is treating a flag as a verdict. A second voice might be a family member. Watch the recording, think about the context and let the candidate explain before you act,” he says.

Source: Talview’s Persistent Systems case study. The figures are supplier-reported and do not establish an overall detection rate.

Give genuine candidates a reason to stay

Every additional check asks something of an applicant: time, information, preparation or trust. Hiring teams should be able to explain what it establishes and why it is needed. Candidates should know how long an exercise will take, which tools they may use and who to contact if something goes wrong.

That clarity should extend to the employer’s own use of technology. Vedran Obradović, Founder & Brand Designer at Vizura Studio, argues that candidates need to understand which steps AI handles, which decisions a named person makes and where they can demonstrate their ability through a task, portfolio review or live conversation.

“"AI-powered" sits next to "people-first", and neither tells a candidate what will actually happen to their application,” he says.

In his identity and website work for Recruitsome, a Dutch AI recruitment platform, Obradović paired imagery of people with the product’s existing “R” mark. The intention was to make the software and the people visible as separate parts of the process. He did not measure how candidates interpreted that distinction, so this is a design example rather than evidence of an improved candidate experience.

His practical recommendation brings the emphasis back to clear communication: “write one plain sentence for each stage of your process, saying what the tool does and who decides, and put it where candidates look before they apply.”

The relationship matters as well as the assessment. As Malia puts it: “Interviews are about both sides working out whether the role and the company are right for each of them.”

A process that treats everyone as a potential fraud risks undermining that conversation. Equally, accepting every polished answer at face value gives employers little confidence in their decisions. Clear expectations, relevant tasks and careful follow-up offer a way to examine capability while giving candidates space to explain themselves.

Jeffery’s conclusion brings those responsibilities together:

“The answer isn't to catch candidates using AI. It's to build hiring processes that can distinguish genuine ability from a convincing performance.”

A well-presented application can earn someone a closer look. The hiring process then needs to establish what sits behind it—and give the person a fair chance to show what they can do.