AI and Assessment: Managing a Real Threat Without Giving Up on Human Judgment

Artificial intelligence is now part of everyday life. It is changing how we work, learn, communicate, write, search for information, and even prepare for interviews or assessments.

In recruitment and talent management, this shift raises a critical question: how can we assess a person's real skills when AI can so easily blur the line between genuine ability and staged performance?

The answer is not straightforward. AI now goes far beyond essay responses or multiple-choice questionnaires. It can help candidates structure an answer, analyze a scenario, prepare for an interview, reword their ideas and, in some cases, respond in real time during a virtual meeting.

AI therefore poses a real threat to genuine assessment. But that threat does not make assessment obsolete. It means organizations must review their practices, strengthen their processes, and stop relying on signals that are too easy to manipulate.

AI doesn't create cheating, but it makes it harder to spot

Cheating is nothing new. Long before generative AI appeared, candidates could get help, copy answers, use prepared templates, or present an overly favourable version of themselves.

What has changed is the power, speed, and discretion of today's tools. AI can produce a clear, structured, and convincing answer in seconds. It can also adjust its tone, add context, and infer what is likely expected.

The risk is clear: a well-produced performance can be mistaken for genuine competence.

A person may look like the ideal candidate on paper, in a written response, or even in an online interview without having the judgment, knowledge, rigour, or skills their application suggests. For employers, the danger is not only that AI may be used. It is that an important decision may be made on information that has not been properly validated.

AI detectors are not enough

When faced with this risk, it is tempting to look for a simple fix: a tool that can automatically detect whether a text or response was generated by AI.

But detectors have limits. They are not infallible. They can produce false positives, miss AI-assisted content, and create a false sense of certainty. As with plagiarism, no detection mechanism can eliminate the risk entirely.

Trust in the process should therefore never depend on a single monitoring or detection tool.

The better response is not to find a perfect way to work around AI. It is to design stronger assessment processes, where cheating is harder to attempt, less useful, and easier to identify through inconsistencies.

AI exposes weak assessment processes

AI is not making assessment invalid. It is exposing a problem that was already there: some processes are too fragile.

An assessment that depends on an unsupervised written response, a loosely structured interview, or one isolated measure becomes easier to compromise. The more a process rests on a single performance, the easier it is for a candidate to gain an artificial advantage.

A stronger process draws on several sources of information. It checks whether the person's explanations, behaviour, reasoning, and results tell the same story.

This is the logic of a convergence approach. No test, interview, situational exercise, or detection tool should carry the decision on its own. But when several indicators point in the same direction, the organization can decide with greater confidence.

Improve assessment; don't do without it

Organizations therefore need to adapt their practices. This does not mean abandoning assessments. It means using them with greater rigour.

Clear rules around AI use are becoming essential. Candidates need to know what is permitted, what is not, and what the consequences are for unauthorized use. Making that distinction clear reduces ambiguity and creates a fairer framework for everyone.

Assessment methods should also reflect the level of risk. For some roles or stages in the process, an unsupervised assessment may be sufficient. For more sensitive decisions, organizations may need additional safeguards, such as supervised administration, identity verification, time limits, follow-up questions, or a structured interview or the repetition of an exercise in a controlled setting.

Finally, organizations should be careful not to overvalue polished responses. A well-written answer is not always evidence of competence. Candidates should be asked to explain their reasoning, justify their choices, apply their ideas to a specific context, or respond to an unexpected follow-up question.

Psychometric tools still matter in an AI world

Psychometric tools are not a silver bullet against AI. No tool, on its own, can guarantee that a candidate has not used technology to support or influence their responses.

Their value lies elsewhere.

Well-designed tools provide structured, comparable, and interpretable data. They move the process beyond the ability to produce a polished answer and make it possible to observe patterns, preferences, reasoning styles, and likely behaviours that can be validated later in the process.

They are especially useful when integrated into a broader process that includes structured interviews, validation questions, situational exercises, observations, and professional judgment.

The aim is not to pretend AI cannot influence an assessment. It is to reduce the risk of mistaking an artificial performance for genuine competence.

Even interviews need to be rethought

Interviews can feel daunting and stressful. But with today's tools, even an online interview can be influenced by AI. Some candidates may over-rehearse in ways that compromise spontaneity, rely on pre-recorded answers, or receive real-time assistance.

This does not mean interviews have lost their value. It means they need to be better structured.

General, predictable questions are becoming easier to rehearse. Follow-up questions, requests for concrete examples, job-specific scenarios, and cross-checks now matter more.

A good interview isn't about hearing a strong answer. It should reveal how the person thinks, how they qualify their responses, how they react to a new situation, and whether their answers stay consistent with the other information collected.

Reassure without downplaying the risk

It would be naive to dismiss AI as a minor issue, given what it can do for better or worse. It calls on organizations to be more thorough and to revisit certain habits.

But it would be just as naive to conclude that assessment no longer has value.

The right response is neither fear nor naivety. It is rigour.

Organizations that want to make sound decisions will need more structured processes, clear expectations, appropriate tools, cross-validation, and proportionate monitoring mechanisms when the context warrants them.

AI does not mark the end of assessment. It marks the end of the illusion that weak processes are enough.

Conclusion

Artificial intelligence is profoundly transforming the assessment context. It can help candidates prepare more effectively, express their responses more clearly and, in some cases, bypass certain steps in the process.

This must be acknowledged clearly.

But this reality does not mean we should stop assessing. It means we need to assess better.

For organizations, the challenge is to protect the validity of decisions without falling into generalized suspicion. This requires clear rules, secure practices, structured interviews, well-designed psychometric tools, and careful cross-checking of the information available.

The future of assessment will not depend on completely eliminating AI from the process. It will depend on designing approaches robust enough to distinguish the appearance of competence from genuine competence.

In a world where AI can help people make a good impression, human judgment, methodological rigour, and high-quality assessment tools are more important than ever.