How Often Do Students Actually Get Caught Cheating Online? (2026 Numbers)

Short answer. Surveys consistently put student-reported cheating at 50–70% of college students at least once during their degree. The percentage who get formally caught is under 2%. The gap between those two numbers is almost entirely explained by exam format — unproctored exams account for the vast majority of uncaught cheating, while proctored exams catch a much larger share of attempts.

This post breaks down the real numbers, why detection is lower than the panic suggests, and which exam formats are the riskiest if your priority is not getting caught.

The headline numbers

Pulling from the most-cited surveys (ICAI's annual student integrity studies, McCabe-Treviño longitudinal work, and post-2020 online-learning follow-ups):

Stat	Number
College students who admit to cheating at least once	50–70%
Students who report being formally caught	< 2%
Cheating rate during unproctored online exams	~70%
Cheating rate during proctored exams	~15%
Students who cite "pressure to get good grades" as the reason	~71%
Students who say they would not cheat if there were a real risk of getting caught	~73%

The 50× gap between "admit to cheating" and "got caught" is the single most important number on this list. It tells you the system catches a very small fraction of what happens.

Why the detection rate is so low

There are four structural reasons most online cheating goes undetected:

1. Professors do not have time to audit every log

A professor teaching 200 students with weekly quizzes is generating thousands of quiz logs per semester. Nobody reads all of those. Logs only get opened when something else flags the student — a grade out of pattern, a tip, a duplicate answer match. If you do not generate one of those triggers, the log usually goes unread.

2. The log records events, not content

Canvas (and Brightspace, and Blackboard) record browser-level events: focus loss, answer changes, timing. They do not record what is on your screen. A professor reading the log sees that something happened, not what happened. That is a fundamentally weaker form of evidence than a proctoring video or a screen recording.

3. The bar for a formal case is high

Most schools require a "preponderance of evidence" — basically, "more likely than not." A quiz log full of focus-loss events is suggestive but rarely sufficient by itself. Professors who try to escalate on log evidence alone often get pushed back by deans, because they have seen too many false positives.

4. Most cheating is invisible by construction

Students who use a phone, a friend in the room, paper notes, or any tool that runs on a device Canvas cannot see — those leave no signature in the Canvas log at all. The log only catches the subset of cheating that happens in the browser, in a way the log can record.

Where the < 2% who get caught actually come from

If you look at how caught cases actually happen, they cluster into a small number of patterns:

1. Duplicate answer patterns. Two friends with identical wrong answers in the same order. This is the single most common path to a caught case, and it does not involve the quiz log at all.
2. Performance inconsistency. Acing the online quizzes and failing the in-person midterm. Professors notice this fast.
3. Tab-switching during a hard question. Focus-loss event right before the correct answer goes in.
4. Submission speed. Finishing a 40-question quiz in 4 minutes.
5. Tips from other students. Roommates and group chats are a more common evidence source than students realize.
6. Proctoring video review. When the exam is proctored, the catch rate jumps because there is actual video to review.

Notice that the Canvas log alone is only one of these. The biggest contributors are pattern-matching across students and proctoring video.

The unproctored vs proctored split

This is the single most predictive variable in the data. Cheating rates roughly 4x higher in unproctored exams, but catch rates in proctored exams are roughly 5x higher.

Format	Cheating attempt rate	Catch rate (approx.)
Unproctored online quiz	~70%	very low
Lockdown Browser only (no camera)	~40%	low
Proctored with webcam	~15%	moderate
Live human proctor + screen recording	~5%	high

If your exam is unproctored — no Lockdown Browser, no webcam, no live monitor — you are statistically in the format with the lowest detection rate. If your exam is proctored, the math is very different.

This is also why most AI extensions, including ExamClutch, explicitly do not function under Lockdown Browser. Proctored environments disable extensions at the OS level. You cannot use them, and the catch rate inside that environment is high enough that you should not try.

Why the fear is bigger than the math

If only 2% of cheaters get caught, why does it feel like the risk is closer to 50%?

A few reasons:

• Asymmetric loss. Getting caught once is catastrophic; not getting caught is invisible. The brain weights catastrophic-but-rare more heavily than common-but-fine.
• Social media amplification. Caught-cheating stories spread on TikTok and Reddit because they are dramatic. The 98% of attempts that go undetected are not interesting enough to post.
• Survivorship bias. Every time you hear "they caught my friend," you do not hear about the hundred friends who were never caught. The stories all run in one direction.
• The post-submit dread. The strongest fear hits after the submit button. The brain replays everything and looks for what could have been noticed.

Knowing the actual base rate does not make the fear go away, but it does help calibrate decisions. Most students dramatically overestimate the per-quiz catch rate.

What actually shifts your risk

If your goal is to be in the "did not get caught" group, the math is not about whether you cheat — it is about whether you generate one of the signals that leads to a case being built. The signals to avoid:

• Tab-switching during the quiz. This is the single most common log signature in caught cases.
• Copy-paste of question text into other tools. The focus-loss is recorded and the round trip is suspiciously fast.
• Identical answers with a friend. If you and a friend are using the same tool the same way, your logs will line up.
• Finishing too fast. Even a perfect log looks suspicious if you finish in 1/5 of the average time.
• A 100 online with a 60 in person. The most powerful piece of evidence a professor ever has is your other grades.

Most of these are about workflow, not luck. A student using a paste-into-ChatGPT workflow generates two of these signals on every question. A student using an inline-DOM extension generates zero of them — but still needs to pace submissions to not look suspiciously fast.

How ExamClutch fits the math

ExamClutch is designed around the signals that lead to caught cases. The whole point of an inline-DOM extension:

• No tab switching. The extension reads the question from the Canvas page directly. No focus-loss events.
• No copy-paste. The page never loses focus to a clipboard event.
• You control the timing. Double-click pacing is up to you. Spread answers out to match a normal completion time.
• Works on Canvas, Brightspace, Blackboard, Moodle, and more. One workflow across LMSes.

The thing it does not change:

• Duplicate-answer patterns with friends. If two of you use any AI tool and get the same wrong answers, the friend-pair signal is identical regardless of how you got the answers.
• Proctored exams. Lockdown Browser disables all extensions. Do not try.
• Long-form essays. ExamClutch handles short-answer fill, not essay generation.

The honest framing: an inline tool reduces the easiest signals to detect. It does not reduce the structural ones.

FAQ

Is the < 2% catch rate actually accurate? It is the most-cited number from McCabe-era survey work. The real rate varies wildly by class, format, and school. Unproctored Canvas quizzes are at the very low end; proctored remote exams with video review are much higher.

Do schools share academic integrity findings with each other? Generally no. A finding at one school does not transfer to another. Graduate school applications often ask directly, though.

Does using a Chrome extension show up in the Canvas log? The extension itself does not announce itself in the log. What shows up is the field changes the extension makes, which look identical to a student typing. See Canvas Quiz Logs Explained for details on what the log actually records.

Why is the cheating rate so much higher in unproctored exams? Because the cost of attempting is low. No camera, no Lockdown Browser, no recording. The behavioral economics are obvious — students respond to perceived risk, and unproctored risk is perceived as near zero.

If most cheating goes undetected, why does it feel risky? Because the downside of getting caught is enormous compared to the upside of any individual quiz. Even a 2% catch rate per attempt is non-trivial when the punishment is an F or worse. The math of "this quiz only" and "every quiz this semester" are very different.

Related reading

• What Happens If You Get Caught Cheating on Canvas?
• Does Canvas Detect Cheating?
• Canvas Quiz Logs Explained
• Canvas Privacy Myths
• Canvas Quiz Answers in 2026

If you want a workflow that does not generate the signals most caught cases are built on, see how ExamClutch works or check pricing.