ARTICLE

AI Cheating in Hiring: How to Stop It on Pre-Employment Tests

AI cheating in hiring turns untimed online tests into open book exams. Learn the prevention stack that works and how to review integrity evidence fairly.

AI Cheating in Hiring: How to Stop It on Pre-Employment Tests

TL;DR

Untimed online tests with fixed questions are now effectively open book, because a chat assistant can answer them in seconds. This owner-focused playbook covers the layered prevention stack and how to review integrity evidence fairly.

To prevent AI cheating in hiring, stop relying on untimed online tests with fixed questions. Instead, give each candidate a different set of timed questions, one at a time. Use picture-based items that are slow to paste into a chatbot, and block copying. Then record integrity signals for a person to review before anyone decides.

No single control is enough. Layered together, these controls make outside help slow, awkward and visible.

AI cheating in hiring became a real concern once free chat assistants could answer most text questions in seconds. For an owner hiring without a recruiting team, that turns a cheap screening step into an open-book exam. This playbook explains why that happened and walks through the five-layer prevention stack that works. It also shows how to match the stack to the role and how to review the evidence without accusing anyone.

Why an untimed online test is now open book

An untimed online test with a fixed question set gives a candidate everything an AI assistant needs: the full question as text, unlimited time and a second browser tab. Before chat assistants, outside help meant asking a friend or searching the web, which was slow and patchy. Now a candidate can paste a question and get a plausible answer faster than they could read it carefully.

What changed for small employers

Large employers often have structured interviews and supervised testing behind every assessment. Small businesses tend to send a single link, often the same test for every applicant, sometimes built in a free form tool. That setup has three weak points:

  • Fixed questions: once one candidate saves or shares them, the answers can circulate.
  • No time pressure: a candidate can look up, paste and rephrase every item.
  • Text that copies cleanly: the question goes straight into a chatbot, and the answer comes straight back.

The result is a score that may reflect how well someone uses an AI assistant rather than whether they can do the job. That defeats the purpose of testing, which is to judge job-relevant ability more fairly and consistently than a resume can.

Is there such a thing as an AI-proof hiring test?

No. Owners often search for an AI proof hiring test, but no online test can guarantee that a candidate worked alone. A determined person can use a second device, a helper in the room or a phone camera. The realistic goal is to make outside help slow, risky and visible. Then most scores reflect real ability, and the unusual ones get a closer look.

Thinking in layers also protects you from overconfidence. Treat any promise of a guaranteed AI proof test with caution, because remote testing cannot deliver that certainty. What it can deliver is a set of controls that each close a different gap.

The prevention stack: five layers that work together

Each layer below targets a different route to outside help. Alone, each one can be worked around. Together, they compound, because getting around one layer usually costs the time that another layer has taken away.

Layer 1: Timed questions served one at a time

A per-question time limit is the most effective change most owners can make. Each question has a short limit and appears only after the previous one is answered. That leaves little time to copy a question, wait for a chatbot, check the answer and reword it. Serving questions one at a time from the server also means the full test never sits in the browser, where it could be saved or captured in one go.

Set limits from how long a capable person needs, not from a guess. Pilot the test on one or two strong current employees and add a reasonable margin. Timing also interacts with fairness. A candidate with a disability may need extra time, which you should grant individually, as covered in our guide to pre-employment test accommodations.

Layer 2: Randomized question pools

A randomized question pool draws each candidate's questions from a larger bank, often in a different order, with answer options shuffled. Two candidates applying for the same job see different tests. A list of answers shared online or between friends therefore loses most of its value, and a single leaked question does limited damage.

The trade-off is comparability. Questions in a pool need similar difficulty so that scores stay fair across candidates. A well-built test library handles this with consistent items. A homemade pool of uneven questions can treat candidates unequally without anyone noticing.

Layer 3: Picture-based and abstract items

Picture-based items include abstract reasoning matrices, shape sequences, charts and diagrams. They are harder to hand to a chatbot than a block of text. A candidate has to capture an image, upload it and wait for a reply, all inside a short time limit. Abstract reasoning also shows how someone spots patterns in unfamiliar problems. That makes it useful for a general hire whose exact tasks will vary from week to week.

Picture-based items have a limit. Current AI models can interpret images, and some can solve visual puzzles. These items raise the effort and time cost of getting help, but they do not remove the possibility. They work best combined with per-question time limits and randomized pools.

Layer 4: Copying blocked

Blocking text selection, copying and right-click on the test page closes the easiest route from question to chatbot. Copy blocking will not stop someone from retyping a question or photographing the screen, which is why it belongs in a stack rather than standing alone.

Recording paste events in answer boxes adds a second view. A long, polished written answer pasted in one go is worth a look. It may still have an innocent explanation, such as drafting in a notes app first.

Layer 5: Integrity evidence for human review

The final layer does not block anything. It records what happened during the session, such as tab switches, pastes and unusual timing. With the candidate's consent, it can also include webcam snapshots. A summary and a timeline let a person see whether a session looks ordinary or deserves a follow-up conversation.

Treat integrity evidence as signals, never a verdict. A tab switch can mean a chatbot, but it can also mean a notification, a dropped connection or a child at the door. Our guide on how to detect AI cheating in online hiring tests explains what each signal can and cannot tell you.

How the five layers compare

LayerWhat it makes harderWhat it does not stopTrade-off to manage
Timed questions, one at a timePasting each question into a chatbot and rewording the answerA fast helper sitting nearbyNeeds fair limits and individual extra time when requested
Randomized question poolsSharing answer lists between candidatesReal-time help during the testQuestions must be similar in difficulty
Picture-based itemsPasting text straight into a chatbotPhotographing the screen for a model that reads imagesNot every skill suits a picture format
Copying blockedOne-click copy and pasteRetyping or using a phone cameraMinor, since honest candidates rarely notice
Integrity evidenceUnusual behavior going unnoticedAnything, on its ownNeeds a person to review fairly and consistently

How to match the prevention stack to the role

Decide first what you are measuring. If the job lets people use AI tools daily, unaided recall may be the wrong thing to test. You could test judgment about AI output instead, such as spotting errors in a drafted email. If the job needs someone to think on their feet, answer customers live or calculate without help, unaided ability matters, and the full stack fits.

Some practical pairings for small business roles:

  • General hire or operations assistant: a short cognitive battery with abstract and numerical reasoning, timed per question, plus a situational judgment test. Picture-based items carry much of the integrity load.
  • Administrative and office roles: performance tasks such as typing a passage or fixing a spreadsheet. These measure doing rather than knowing, which leaves less room for pasted answers.
  • Customer-facing roles: timed situational judgment questions, a short written reply scored against your own rubric, and a spoken interview question on the same scenario.
  • Writing-heavy roles: accept that a written sample may be AI-assisted. Ask for detail drawn from the candidate's own experience, then discuss it in the interview.

For more on choosing question formats, see how to reduce AI assistance in online assessments.

Add a verification step for every finalist

The cheapest check is a short follow-up. Ask each finalist to talk through one or two answers, or to solve a similar problem on a call. Someone who did the work can usually explain the reasoning. Someone who relied on outside help often finds that harder. Use the same follow-up for every finalist, not only those with flagged sessions, so the step stays consistent and nobody is singled out.

How to review integrity evidence without accusing anyone

Reviewing integrity evidence well means weighing signals in context, looking for innocent explanations and following the same process for everyone. The steps below keep the review fair and defensible.

  1. Read the score and the evidence together. A high score from an ordinary session needs no extra step. A high score with frequent tab switching deserves a conversation, not a conclusion.
  2. Look at the timeline, not just a number. Patterns matter more than single events. Repeated switches right after each question appears tell a different story from one switch midway through.
  3. Consider innocent explanations. Connectivity problems, a shared computer, assistive technology or an approved accommodation can all produce unusual signals.
  4. Use a consistent follow-up. Offer a short verification interview or a supervised retest, and apply the same rule to everyone with similar signals.
  5. Word the invitation neutrally. Say you would like to go through a few answers together. Do not suggest wrongdoing.
  6. Record the reason for your decision. Note what you saw and what the follow-up showed. Written notes keep decisions consistent and help you explain them later.
  7. Never reject on a signal or score alone. A person makes the hiring decision after weighing the test, the interview and references.

Keep prevention fair and lawful

Tests used in hiring fall under federal equal employment law. The U.S. Equal Employment Opportunity Commission (EEOC) enforces Title VII of the Civil Rights Act of 1964, the Americans with Disabilities Act (ADA) and the Age Discrimination in Employment Act (ADEA). Title VII and the ADA generally apply to employers with 15 or more employees, and the ADEA to employers with 20 or more. The EEOC's guidance on employment tests and selection procedures explains that tests should be job-related and should not unfairly screen out protected groups.

Prevention controls raise their own questions:

  • Time limits can disadvantage some candidates with disabilities. The ADA requires reasonable accommodations, such as extra time, unless they would cause undue hardship.
  • Webcam and identity checks involve personal data, and some states regulate biometric information. Illinois does so through its Biometric Information Privacy Act. Tell candidates in advance, ask for consent and keep images only as long as needed.
  • Adverse impact monitoring helps you notice if a test screens out one group at a higher rate. The four-fifths rule, from the Uniform Guidelines on Employee Selection Procedures, is a common first check. Cognitive tests deserve particular attention here. See cognitive ability tests: what they measure and how to use them.

This section describes the laws in general terms only. Requirements vary by state and situation, so ask a qualified employment attorney about your circumstances.

A setup checklist for your next opening

Use this checklist before you send the first invitation.

  1. List the three or four skills the job truly needs.
  2. Choose tests that measure those skills, favoring timed, picture-based or performance items.
  3. Confirm the test draws questions from a randomized pool.
  4. Set per-question time limits from a pilot with current staff.
  5. Turn on optional camera checks only if the role justifies them, and explain them in the invitation.
  6. Tell candidates how the test works, what is recorded and how to request an accommodation.
  7. Plan the verification step for finalists before you see any results.
  8. Decide who reviews integrity evidence and how decisions are recorded.

Clear expectations do some preventive work on their own. A candidate who knows the test is timed and that activity is recorded has less reason to try a shortcut.

Frequently asked questions

Can candidates use ChatGPT during an online pre-employment test?

On an untimed test with copyable text, yes. A candidate can paste questions into ChatGPT or another assistant and get answers within seconds. Per-question time limits, randomized questions, picture-based items and blocked copying make that much slower and harder. Integrity signals then show a reviewer when a session looks unusual and may deserve a follow-up conversation.

Should I disqualify a candidate who switched tabs during a test?

Not on that signal alone. A tab switch can come from a chatbot, but also from a notification, a dropped connection or assistive software. Look at the pattern on the timeline and consider innocent explanations. Then use the same follow-up for everyone, such as a short interview where the candidate talks through a few answers.

Is it legal to use webcam monitoring in a hiring test?

Webcam checks are common in remote testing, but privacy and biometric rules vary by state. Illinois, for example, regulates biometric data under its Biometric Information Privacy Act. Tell candidates about camera checks in advance, ask for consent, limit how long images are kept and ask a qualified employment attorney about your state.

Do time limits make a test unfair to candidates with disabilities?

Time limits can, which is why the Americans with Disabilities Act requires reasonable accommodations such as extra time when a candidate needs them. Explain how to request accommodations in your invitation and grant them individually. Set base limits from how long capable employees actually take, not an arbitrary short clock.

Are picture-based questions harder for AI to answer?

Picture-based questions are harder to get help with, not impossible. Text can be pasted instantly, while an image must be captured and uploaded under time pressure. Some AI models can interpret images and solve visual puzzles. Picture-based items therefore work best alongside per-question time limits and randomized question pools.

Should I let candidates use AI if the job allows it?

Sometimes that is the better test. If employees use AI tools every day, consider measuring how well candidates judge and correct AI output instead of unaided recall. Keep the rules explicit by stating whether AI is allowed. Test skills such as reasoning or live communication without help when the job requires them.

Is an in-person test better than an online one for preventing AI cheating?

A supervised in-person test removes most outside help, but it costs time for you and the candidate and can narrow your applicant pool. Many small businesses screen with a well-protected online test, then use a short in-person or video verification step for finalists. That combination balances reach, fairness and confidence.

Conclusion

AI assistants did not make online testing useless. They made untimed, fixed and copyable tests unreliable. The fix is a stack: timed questions served one at a time, randomized pools, picture-based and performance items, copying blocked, and integrity evidence that a person reviews. Decide what the role really needs, verify finalists the same way every time, and keep every decision in human hands.

If you would rather not build these layers yourself, HireValid serves each candidate a different, timed test one question at a time from the server. It blocks copying and records signals such as tab switches, pastes and timing. Its Integrity Score, from 0 to 100 with an evidence timeline, summarizes what was observed for you to review, and nothing automated rejects a candidate. Optional camera checks are explained to candidates on the welcome page, and candidates are asked to consent. You can try HireValid free on 3 candidates, or see the monthly plans when you are ready to hire.

About the editorial team

Written and fact-checked by the HireValid editorial team. Sources checked . Read the editorial policy.

Does a low Integrity Score reject a candidate?+

No. Integrity signals may have innocent explanations, including connection issues or accessibility needs. A person should review the evidence and speak with the candidate before deciding.

What will candidates need?+

A browser and a reliable connection. Typing, spreadsheets and code tasks are intended for desktop. If an employer enables camera checks, candidates must receive a clear notice and a route to request an alternative.

Are the scores official qualifications?+

No. These are tools for hiring decisions. English results are CEFR-aligned level estimates, not official certificates. Cognitive scores are not clinical IQ results.

Keep exploring

LESS GUESSWORK. MORE GOOD PEOPLE.

AI can polish an answer.
Hire for the ability behind it.

Explore advanced hiring assessments and integrity tools built for the AI era.

Start freeOr try a sample test No credit card. No annual commitment.