A few years ago, the biggest worry with online tests was a friend sitting next to the candidate. Today the friend is an AI chatbot, it is free, and it answers in seconds.
So it is a fair question. If a candidate can paste your question into ChatGPT, is an online test still worth anything?
The short answer is yes, but only if the test is built for the world we now live in. Some test formats are easy for AI. Others are very hard for it. This guide explains the difference, and what you can do to make sure your results still show the real person.
How candidates are using AI in hiring
Most candidates are not trying to trick you. Many use AI the way they use a spell checker: to tidy a CV or plan an answer. But some go further. The common patterns are:
- Pasting questions into a chatbot during an untimed or loosely timed test.
- Using a second device, such as a phone, to photograph a question and get an answer.
- Getting AI to write take-home tasks, from cover letters to coding exercises and case studies.
- Using real-time "interview assistants" that listen to a video interview and suggest answers on screen.
Each of these works best against one type of assessment: tasks with no time pressure, a single fixed answer, and no way to see how the candidate got there.
Which tests are easy for AI to beat
Untimed knowledge questions
If a question asks for a fact, and the candidate has as long as they like, AI will usually get it right. "What does VLOOKUP do?" is easy to look up. It tells you very little about whether someone can actually build a working spreadsheet.
Take-home written tasks
A cover letter, a short essay or a written case study can now be produced in under a minute. These tasks still have some value at interview, where you can ask the candidate to explain their thinking. As a filter on their own, they no longer separate strong candidates from weak ones.
The same questions for everyone
If every candidate sees the same fixed set of questions, answers can be shared, saved and fed to AI ahead of time. Once a question bank leaks, it stops measuring anything.
Which tests are hard for AI to beat
Timed reasoning tests
Numerical, verbal, logical and abstract reasoning tests give a short time for each question. Copying a chart, a passage or a pattern into a chatbot, waiting for an answer and checking it takes time the candidate does not have. A candidate who tries usually finishes fewer questions, not more.
Picture-based questions add another barrier. Pattern and diagram puzzles are slow to describe in words, and AI tools are still unreliable on them.
Questions drawn at random from a large bank
When each candidate gets a different mix of questions, there is no single answer sheet to share. Randomised tests also stop a candidate from simply retaking the same test with the answers ready.
Situational judgement and personality
Situational judgement tests ask how a person would handle a real workplace moment. Personality questionnaires ask how they prefer to work. AI can give a "nice" answer, but the best answer depends on your role, your team and your values, which the chatbot does not know. And a personality profile with no clear right answer gives AI very little to aim at.
Practical skills tests
A job skills test built around the real work, such as reading a spreadsheet, spotting errors in a document or reasoning through a customer problem, tests applied skill rather than recall. It is much harder to fake doing a task than to fake knowing a fact.
Seven ways to keep your tests honest
1. Use timed tests. A sensible time limit is the single biggest defence. It leaves enough time to think, but not enough to outsource.
2. Use randomised questions. Choose a platform that draws different questions for each candidate.
3. Mix several types of test. A candidate might get help on one test. It is much harder to fake a consistent profile across reasoning, personality and job skills together. When results don't fit together, that is a signal worth exploring.
4. Watch the test behaviour, not just the score. Good platforms record how a test was taken, for example if a candidate left full-screen mode or moved the mouse away from the test window. These are not proof of cheating on their own. They are prompts to ask a question at interview.
5. Test early, check later. Use an online test to build your shortlist. Then confirm key skills face to face, or on a short live task. A candidate who scored highly on numerical reasoning should be able to talk through a simple calculation with you.
6. Tell candidates the rules. Many people use AI because they are not sure if it is allowed. A short, clear statement such as "please complete this test on your own, without AI tools" removes the grey area. Most candidates will respect it.
7. Stop relying on tasks AI can do for them. If the cover letter is now written by a machine, give it less weight. Put your effort into the parts of the process that show real ability.
Should you ban AI completely?
Not always. In many jobs, using AI well is now part of the work. A marketer or analyst who can use AI tools properly may be more productive than one who can't.
The answer is to decide what you want to measure. If you want to know how someone thinks under their own steam, use a timed, randomised reasoning test with clear rules. If you want to know how well they work with AI, design a separate task that allows it, and judge the quality of the result.
What you should not do is assume a polished application means a skilled person. That link is weaker than it has ever been.
The bottom line
AI has not made testing useless. It has made bad testing useless.
Fixed, untimed, knowledge-based tasks are now easy to game. Timed, randomised tests that measure reasoning, judgement, personality and practical skill together are still very hard to fake, and they give you a much clearer view of the person behind the application.
At Test Candidates, our assessments are built to reveal the real skills and traits of each candidate. Questions are timed and drawn at random, and you can combine aptitude, personality, soft skills and job skills tests in one assessment. See how it works or start free.
Frequently asked questions
Can candidates use ChatGPT during an online aptitude test?
They can try, but in a well-timed test it rarely helps. Copying a question, waiting for an answer and checking it uses up most of the time allowed. Picture-based questions such as pattern and diagram puzzles are also hard to describe to a chatbot quickly.
How can I tell if a candidate cheated on a test?
Look at the test behaviour data, such as leaving full-screen mode, and compare results across different tests. Then check key skills at interview. A large gap between a test score and what the candidate can show you in person is the clearest sign.
Are personality tests affected by AI?
Less than knowledge tests. There is no single right answer to aim for, so AI cannot simply "pass" one. Candidates can still try to present themselves in a good light, which is why personality results work best alongside reasoning and skills tests.
Should I let candidates use AI in assessments?
It depends on the role. For roles where AI is part of the daily work, a separate task that allows AI can be useful. For measuring a candidate's own reasoning, set clear rules that ask them to work alone.
