How to stop cheating in online tests?

To stop cheating in online tests, change what the test asks before you add cameras. A question that can be answered by a search, a chat window or a friend on a second laptop is the actual vulnerability, and proctoring only tells you it happened. The strongest setup pairs the two: assessments built so answers cannot be looked up, plus enough session evidence that a human can judge a flag fairly.

That order matters more in 2026 than it did two years ago, because the tooling candidates reach for is no longer exotic. It is a browser tab, a phone propped on a stack of books, and, judging by what people now search for, sometimes a camera the size of a shirt button.

TL;DR

• Test design beats surveillance. Questions with a single lookup-able answer leak first, no matter how closely you watch the room.
• Deterrence works better than appeals to honesty. In one controlled study, an honesty pledge changed nothing and a warning that detection was in place cut observed cheating from 70% to 15%.
• Covert hardware is the new gap. A pinhole camera feeding an AI assistant defeats screen-only monitoring, which is why a second camera angle matters.
• AI help is now normal behavior, not an edge case: 84% of developers use or plan to use AI tools, and about half use them daily.
• A flag is evidence, not a verdict. Under US selection law an assessment used to decide who advances is a selection procedure, so how you act on a flag needs to be consistent and documented.

What counts as online assessment cheating?

Online assessment cheating is any use of help the assessment did not authorize: looking up answers, getting them from another person, having someone else sit the test, or passing the questions to a tool that answers for the candidate. The practical definition is narrower than the moral one. What you care about is whether the score still describes the candidate's own ability.

Here the research is less tidy than vendor copy suggests, and it is worth knowing before you spend a budget. In a study of an online biology course published in Online Learning, Pleasants and colleagues observed 70% of students cheating on unproctored tests. Appealing to students' honesty and asking them to sign a pledge did not work. Warning them that the instructors had technology that could detect cheating, with real penalties attached, cut it to 15%. The finding that should change your thinking is the third one: the researchers did not find that exam performance changed when cheating behavior changed.

So cheating is common, deterrence works, and the score impact is smaller than the panic implies. That argues for proportionate controls on the assessments that actually gate a hiring decision, not maximum surveillance on everything.

Mini camera and AI cheating: can a rig solve your exam?

Yes, and this is the vector screen monitoring was never built to catch. The setup is simple: a small camera or phone lens captures the question off the screen or the page, an AI assistant reads it and returns an answer, and the candidate reads that answer off a second device or hears it through an earpiece. Nothing unusual happens inside the test browser, so tab-switch detection and copy-paste logging see a clean session.

Pause on why that works. Almost every control in a standard proctoring stack watches the test window or the primary webcam's narrow cone. A rig that lives outside both is invisible to it. The countermeasure is not a better screen check, it is a different angle.

This is the one place where a second camera earns its cost. Testlify's dual-device proctoring turns the candidate's phone into that second camera, positioned so it captures both the candidate and the laptop screen, and it is a hard gate rather than a nice-to-have: the assessment will not start until mobile monitoring is active, and a disconnection blocks the session and is reported to the test administrator. A desk shot showing hands, screen and surroundings is what makes a hidden lens or a second monitor hard to run.

The honest caveat: a second angle raises the effort required, it does not make cheating impossible. Treat it as the control you switch on for the assessment that decides a shortlist, not for a 10-minute screening quiz.

How do candidates cheat in an online aptitude test?

Mostly by finding the questions, not by outsmarting the proctor. Aptitude and reasoning tests are drawn from item banks, and item banks circulate. Researchers at Queensland University of Technology audited an engineering school's assessments and found published answers matching content from more than half of the 41 subjects they examined, nearly 1,200 question and answer pairs in total, with almost 400 from a single subject. Speed is the part that should worry a hiring team: as Inside Higher Ed reported, half of mechanical engineering questions posted to one homework-help site got a response within about 84 minutes by 2020, down from 216 minutes in 2015. Chegg disputed the findings and pointed to a tool that lets educators block questions during a set assessment window.

Translate that to hiring. If your assessment uses the same 30 items every candidate sees, the first few candidates effectively publish it for everyone after them. The fixes are unglamorous:

1. Randomize from a pool several times larger than the test length, so two candidates rarely see the same set.
2. Rotate items on a schedule, and retire anything you find circulating.
3. Time-box tightly enough that a 45-minute round trip to an answer site does not fit.
4. Ask for work product, not recall. A short task with a defensible method cannot be answered from a leaked key.

Why candidates search for TestDome cheating answers

Because the platform name is how candidates describe the test they have been sent. Searches pairing a named assessment tool with the word cheating are answer-seeking, and they tell you something useful: candidates assume a named, widely used test has a published key somewhere. Often they are right, for exactly the item-bank reason above.

For a hiring team the lesson is not which vendor gets searched. It is that brand-standard tests carry brand-standard leakage, so the question set you rely on should either be large and rotating or specific to your role. If you are comparing platforms on how they handle this, our Testlify and TestDome comparison covers the integrity controls side by side, and the TestDome alternatives guide sets out what else is available.

How do you detect cheating during online exams?

You detect cheating during online exams by gathering several weak signals and reading them together, never by trusting one. Identity checks, environment controls, behavior logs and a second camera angle each cover a different gap, and the pattern across a whole session is what carries meaning. No single signal proves anything. A tab switch might be a notification. A face leaving frame might be a delivery driver. The pattern across a session is what carries meaning, which is why the useful output is reviewable evidence rather than a pass or fail stamp.

These are the layers worth running, and what each one genuinely does:

How do you stop candidates cheating using AI in hiring assessments?

Start by accepting that AI use is the default working condition, then decide per assessment whether it is cheating at all. In the 2025 Stack Overflow developer survey, 84% of respondents said they use or plan to use AI tools in their development process, up from 76% the year before, and 47.1% use them daily, rising to 50.6% among professional developers. Trust is a different story: 46% actively distrust the accuracy of AI output and only 3.1% highly trust it.

Read that against your own rules. A ban on what half your engineering candidates do daily, for a job where you will hand them the same tools in week one, measures compliance rather than skill. So split your assessments in two.

• AI-allowed assessments. Say so explicitly and raise the bar. Ask the candidate to critique, correct or extend AI output, or to defend a decision. The skill you are measuring is judgment, which survives the assistant.
• AI-restricted assessments. Reserve these for the specific abilities you need to observe unaided, state the rule before the first question, and back it with controls that detect assistance behaviorally: AI tool and browser extension detection, full-screen enforcement, per-question timing, and a second camera angle.

How to stop a candidate cheating using AI on a live test

For one candidate in front of you, the sequence is short. State the rule on the screen before the first question, so there is no ambiguity about what was allowed. Instrument the session rather than watching it live. Then, if a flag comes back, do not adjudicate from the log: put 15 minutes on the calendar and ask the candidate to talk through two of their own answers. Someone who did the work explains their reasoning easily. Someone who pasted it cannot, and that conversation is both fairer and more reliable than any score you could infer from a timing anomaly.

One thing to avoid: judging whether an answer was AI-written from how the prose reads. Liang and colleagues at Stanford found that GPT detectors are biased against non-native English writers, consistently misclassifying their writing as AI-generated while identifying native samples correctly. Used as a pass or fail test in hiring, that is a fairness problem with your name on it. Detect tool usage, not writing style.

Cheating-resistant hiring assessments for remote candidates

A cheating-resistant assessment is one where knowing the answer in advance does not help much. That is a design property, and it is cheaper to build than to police. Four moves do most of the work for a remote shortlist:

1. Ask for method, not recall. A question with one correct number leaks. A question that asks the candidate to pick an approach and justify the tradeoff does not travel well, because the answer is theirs.
2. Use role-specific scenarios. Generic items are the ones that end up on answer sites. Something drawn from your actual work has no published key.
3. Score blind, against a rubric written first. This is where consistency comes from, and it is also what makes a flagged session reviewable rather than arguable.
4. Match strictness to stakes. Light controls on a first-round screen, full identity and dual-device proctoring on the assessment that creates the shortlist.

Pro tip: run your own assessment yourself with a phone and a chat window open before you send it to anyone. Most teams find within ten minutes that two or three items can be answered straight from a search, and those items are the entire problem. Fix them and you have done more than any proctoring setting will.

For a deeper treatment of the design side, see our guide to designing cheating-resistant online tests, and how AI is used in cheating prevention for the detection side.

Is a background check the best cheating control?

No, and the mismatch is worth being precise about. A background check verifies history: identity documents, employment, credentials, criminal records where lawful. It is retrospective and it happens after the assessment. It cannot tell you whether the person who scored 82% was the person who took the test.

It does catch one thing well, and the case is documented. The FBI's Internet Crime Complaint Center issued a public service announcement on deepfakes and stolen personally identifiable information used to apply for remote work positions in June 2022, mostly for IT, programming, database and software roles, some with access to customer data and corporate systems. Complainants reported voice spoofing in interviews, where lip movements did not coordinate with the audio and actions like coughing or sneezing did not align with what was on screen. Employers caught several of these at the background-check stage, when the PII an applicant supplied turned out to belong to someone else.

So the sequence that works is identity verification at the test (photo ID with face match, snapshots through the session), interview checks for a live human, and a background check to confirm the history. Different jobs, run in that order. More on the interview half in our guide to preventing cheating in online interviews.

What a cheating flag should and should not do

A flag should start a review. It should never end a candidacy on its own, and that is a legal position as much as an ethical one. Under the Uniform Guidelines on Employee Selection Procedures, a selection procedure is any measure, combination of measures, or procedure used as a basis for any employment decision, and the definition explicitly spans the full range of assessment techniques from paper and pencil tests through informal interviews. An assessment that decides who advances is a selection procedure, so rejecting on an automated flag is a selection decision you need to be able to explain and apply consistently.

The identity half carries its own measurement risk. NIST's face recognition evaluation tested 189 algorithms from 99 developers against 18.27 million images of 8.49 million people and found demographic differentials in most of them, with false positive rates in one-to-one matching often varying by a factor of 10 to 100 times across demographic groups, higher for Asian and African American faces than Caucasian faces, and highest for African American females in one-to-many search. The most equitable algorithms were also among the most accurate, so quality matters, but a face mismatch is a prompt to look, never proof of fraud.

This is what the Testlify Assessment Integrity Framework is built around: protect the validity of a result through identity checks, proctoring controls, AI assistance detection, suspicious-behavior signals and reviewable evidence, with configurable strictness, while the final judgment stays human-led. In practice that means a three-tier flag rather than a verdict. Green means nothing unusual was observed. Yellow means some behavior was not ideal and a quick manual review is recommended. Red means cheating was confirmed. Automatic termination exists as a separate opt-in control with a threshold the recruiter sets, not something the flags do by themselves.

Two things to get right before you switch any of it on. Tell candidates what is monitored, in the rules screen, before they start. And know your retention: face verification data is deleted after 30 days and consent can be withdrawn at any time, while video and audio responses are removed after six months. If you hire in a US state with its own biometric privacy statute, or anywhere under GDPR, biometric consent is not a formality you can paper over later, and the penalties attach per candidate.

Hire with assessments you can trust

If your current test could be answered from a search bar, no amount of monitoring fixes it. Rebuild the questions, then add the controls the stakes justify. You can see how the integrity layers fit together across the Testlify test library, read the detail in the AI proctoring buyer guide, or book a demo and bring your worst-behaving assessment with you.

FAQs