Can candidates use ChatGPT on a 12-minute aptitude test?

The clock is the biggest safeguard

A take-home project gives someone hours. That's plenty of time to paste the prompt into an AI tool, tidy up the answer, and send it in as their own. A timed aptitude test works the other way. Ours is 50 questions in 12 minutes, which works out to about 14 seconds per question.

Now picture what getting AI help actually involves. The candidate has to type the question or snap a photo of it with their phone, wait for a response, read it, find the matching choice, and click. Those steps vary from person to person, but together they usually take longer than 14 seconds. Do that on every question and the candidate simply reaches fewer of them before time runs out.

That's the part most people miss. On a speed test, the score depends on how many questions you get through as well as how many you get right. Someone leaning on an AI tool spends their clock waiting on it. It rarely pays off.

What the test records

You don't have to take the timing argument on faith. The test page also watches for the most common ways people reach for outside help on the same computer, and shows you what it saw:

• Copy and paste are blocked. Nothing in the test needs typing, so any attempt to copy or paste is stopped and counted.
• Leaving the test window. Every time the window loses focus, it's logged, along with the total seconds spent away.
• Time on each question and answer changes, so you can see the pace of the attempt, not just the final score.
• Questions drawn from a rotating pool in a random order, so there's no fixed answer key to pass around online.

Before the test starts, every candidate also agrees in writing to work alone, without a phone, other devices, AI tools, or notes. That matters. It means you're holding them to a rule they knowingly accepted.

What no online test can promise

Some testing companies say they block AI assistants. It's worth being clear about what that can and can't mean. A web page can notice what happens inside the browser. It can't see a phone in someone's lap, a second laptop on the desk, or a friend sitting just out of view.

The only way to get close is heavy monitoring: webcam recording, screen capture, browser add-ons. We chose not to go there. That kind of setup tends to put off good applicants, creates barriers for people using assistive technology, and still doesn't catch a phone held below the camera. We'd rather be straight with you about the limits and let the clock and the signals do the work.

How to read a scorecard when you're worried

• An honest attempt usually looks steady. The candidate works through questions, slows down on hard ones, and stays on the page.
• An attempt worth a second look shows a combination: several trips away from the window, long pauses, and correct answers on hard questions right after those pauses, over and over.
• One signal on its own means very little. A single focus loss is often a notification. A high score on its own isn't suspicious either. Some people are just quick.

What to do if something looks off

Don't accuse anyone, and don't quietly toss the result. Flag the attempt for review in your dashboard and offer a fresh try with a new code at a time you agree on. You can also ask them in the interview to talk through a similar problem out loud. Someone who did the work honestly loses a few minutes. Someone who didn't usually becomes clear pretty quickly.

AI isn't going away, and you don't need to treat every applicant as a suspect because of it. A short timed test, a clear rule, and a few signals you can actually read give you a fair way to tell who can do the thinking themselves.