Verification tests have had something of a second wind with the emergence of ai-cheating. As well as trying to prevent candidates cheating in real time with the different anti-cheat measures, you can also tell them upfront that a supervised re-test exists to validate their scores.
The announcement alone stops most cheating before it starts, and for those that do manage to cheat, you will likely catch them when they have to perform without the assistance of AI in-person.
What's a verification test, exactly?
A verification test is a shorter version of the original assessment, delivered at a later stage of your process, in a supervised setting. Usually that means in person at an interview day or assessment centre. You compare the two scores. If they're broadly consistent, the original result stands. If there's a significant discrepancy, you've learned they likely cheated in the original one.
AI cheating inflates scores which creates a detectable gap
When a candidate uses AI to help them through a remote cognitive ability test, what they're really doing is borrowing a score that isn't theirs. Historically this would have been a friend taking the test for them, but now it's using AI to assist them. The remote result looks strong. But the candidate's underlying ability hasn't changed.
Put them in a room, take away the AI, and ask them to do it again. If the first score was genuine, the second will be similar. If it wasn't, there'll be a gap. An extremely capable AI tool might push a candidate's remote score to the 90th percentile. If their actual ability sits at the 50th, that's a 40-point discrepancy.
Telling candidates a verification test exists deters most cheating before it starts
You don't necessarily need to catch anyone for a verification test to be effective. The announcement does a lot of the work.
There's a widely cited stat in assessment circles that around 83% of people say they'd cheat if they thought they could get away with it. The key bit being "get away with it". When candidates know a verification test exists, cheating is a much riskier endeavour.
Most candidates will just decide it isn't worth it, and just try their honest best.
Conclusion and next steps
Verification tests tell candidates a re-test exists, and most will decide not to cheat. For those who do, the supervised re-test will expose the gap.
That said, there is a huge practical downside which you may have pondered reading this... If someone does cheat and you only find out at a late in-person stage, that's highly inconvenient (for everyone!). You've already spent time on them, and so have they. And it means cheaters stay in your process longer than you'd like before you catch them (not good, and not fair for the honest people who ended up getting rejected)
The cleaner solution is to use an assessment that doesn't give AI much to work with in the first place. That's what MindmetriQ is built to do. Rather than relying on catching cheaters after the fact, it's designed so that anyone using AI would likely perform much worse due to the AI's inability to produce a quick or useful output. If you want to understand what actually makes an assessment AI-resistant, we've put together a full breakdown of that.
Ben Schwencke
Chief psychologist at Test Partnership. MSc in Organisational Psychology with over ten years experience in psychometric testing.
