Generative AI has made many employers ask whether online test results can still be trusted. Some reports suggest widespread cheating; others suggest the fear is ahead of the evidence. This article looks at what the research shows, why the risk varies so much between types of test, and how Viewpoint's Validity Checks help employers keep online assessment fair and trustworthy.

How big is the problem?

Estimates vary widely. Some assessment vendors report that a large share of candidates try to cheat on online tests, especially for entry-level and technical roles. Independent research paints a more measured picture. Clevry's State of AI in Talent Assessments 2026 report found that while 62% of HR professionals believe candidates are using AI to cheat, only 26% said they had seen evidence of candidates manipulating assessment results [1]. The report concluded that the risk can be managed through better assessment design and simple deterrents.

The balanced view is that AI-assisted cheating is real and worth taking seriously, but it is not a reason to abandon online assessment. The question is how to design and run tests so that results stay meaningful.

Why some tests are harder to game than others

The risk is not the same for every kind of test:

• Knowledge-style questions with a single factual answer are the easiest for an AI tool to answer.

• Timed reasoning tests are harder, because copying a question into another tool, waiting for a reply and checking it takes time that a well-timed test does not allow.

• Visual and speeded tests, such as checking, concentration and game-based tasks, are hard to pass to an AI tool quickly and accurately.

• Situational judgement tests and personality questionnaires ask about judgement and preferences. An AI tool can suggest a "safe" answer, but the scoring keys and the follow-up at interview make it much less useful than people assume.

Research on unproctored internet testing has long recommended combining sensible test design with verification of results for shortlisted candidates, for example a short supervised re-test [2]. That advice is just as relevant in the age of AI.

How Viewpoint's Validity Checks work

Viewpoint's Validity Checks are a set of safeguards that employers can switch on for a test and for their own candidates. They are designed to deter misuse, flag sittings that need a closer look and make the evidence easy to read:

• A clear warning before the test. Candidates are asked to close other browsers and not to use AI, and they confirm this before they start.

• Identity marks on the test itself. The candidate's name and ID appear discreetly across each question, with a copyright notice and a visible note to AI assistants that this is a live recruitment test. This discourages photographing or copying questions into other tools.

• Leaving the test page. The platform records when and for how long a candidate leaves the test page, for each page and overall.

• Timing checks. Answer times that are implausibly fast or follow unusual patterns are highlighted, with time per question available for review.

Results are summarised in one overall flag for each sitting, with each check shown underneath it, so a busy recruiter can see at a glance whether anything needs attention and click through for detail.

A flag is a prompt, not a verdict

A Validity Check flag shows that something about a sitting is worth a closer look. It does not prove that a candidate cheated. A candidate might leave the page because of a power cut, a family interruption or assistive technology. That is why flags are best used to guide the next step: a short supervised re-test, a structured interview question about the candidate's approach, or a conversation about what happened. Treating flags as automatic rejections would be unfair and could disadvantage candidates who use reasonable adjustments.

The other side of the coin: AI skills

As the Clevry report points out, some of the skills candidates might use to game a test, such as writing good prompts and working quickly with AI, are skills employers increasingly want [1]. The answer is to measure those skills openly, in the right setting. Viewpoint's AICheck does exactly that: it tests how well people use AI at work, including checking what the AI produces. Our guide to AI skills tests explains how.

Practical steps for employers

• Choose test formats that suit online delivery for each role, and use a mix of test types.

• Explain the rules clearly to candidates and ask them to confirm they will follow them.

• Switch on Validity Checks where the stakes justify it, and review flagged sittings rather than rejecting them automatically.

• Verify key results for shortlisted candidates, through a supervised re-test, a work sample or structured interview questions.

• Measure AI skills openly where the role needs them.

See our ability tests, our situational judgement tests, or contact us about switching on Validity Checks for your campaigns.

Frequently asked questions

Are online tests still worth using?

Yes. Well-designed online tests remain a fair and efficient way to compare candidates, especially when combined with sensible safeguards and verification of shortlisted candidates.

Does a Validity Check flag mean the candidate cheated?

No. It means the sitting is worth a closer look. Flags should guide follow-up, not trigger automatic rejection.

Can Validity Checks be switched on for some tests and not others?

Yes. They can be switched on for particular tests and for particular clients.

Is using AI always cheating?

Not always. In some roles, using AI well is part of the job, and it can be measured openly with a test such as AICheck. In a test of the candidate's own reasoning, it is not allowed.

References

1. Clevry (2026). State of AI in Talent Assessments 2026. Clevry.
2. Tippins, N. T., Beaty, J., Drasgow, F., Gibson, W. M., Pearlman, K., Segall, D. O. & Shepherd, W. (2006). Unproctored internet testing in employment settings. Personnel Psychology, 59(1), 189–225.