Sign up free

Webcam proctoring: what it catches, misses and what students say

An evidence review of live, recorded and automated webcam proctoring: scores, false flags, bias, what students say, what courts and regulators decided.

The EdProcto editorial teamWritten and checked against the research it cites. Last checked . How the guides are made

Guides · Integrity and cheating · · 13 min read

What does webcam proctoring catch, what does it miss, and what do students think of it?

Webcam proctoring is followed by lower scores in most comparisons with unwatched sittings, which may mean it deters some cheating. Its automated flags are weaker evidence: in the studies we read most flagged students had done nothing wrong, the software could be bypassed, and face matching erred unevenly. Many students fear a wrongful flag. Keep it for high stakes, with a person reviewing every flag.

Webcam proctoring means watching a student through their own camera while they sit a test away from an exam room. This guide reviews the evidence: score comparisons, tests of the software, student surveys, and decisions by courts and data-protection authorities.

It is for a teacher, lecturer or exams officer deciding whether to use it. Each source was read on 10 October 2026; "abstract read" marks a study where only the abstract was open to us; nothing here is legal advice. Lighter measures are in how much proctoring a quiz needs.

EdProcto publishes this guide. Its own integrity settings act in the browser; the last section says more.

What is webcam proctoring, and what are its three kinds?

Webcam proctoring is supervision of an online test through the student's camera, usually with the microphone and screen as well, and Hussein et al. (2020) sort it into three kinds.

Kind Who watches What it asks of staff
Live A person, during the sitting A booked time and a proctor on duty
Recorded A person, afterwards Hours of review, which the authors call costly and hard to scale
Automated Software marks moments for a person to check Less watching, and flags that can point at innocent events

A room scan, in which the student turns the camera round the room before starting, can come with any.

What does webcam proctoring do to scores and to cheating?

Scores are usually lower under a camera than unwatched; what that says about cheating is an inference, not a measurement.

Dendir and Maxwell (2020) studied two online courses that stayed the same while webcam recording was introduced for their high-stakes exams. Average scores fell in both, the fall survived statistical controls, and the authors read it as evidence of earlier cheating (abstract read). It compares the terms before and after, not groups set by chance.

Vazquez et al. (2021) ran two field experiments, telling students in advance which exams would be watched. Unwatched students scored over 11% higher on average, but live proctors in a face-to-face class moved scores far more than web-based proctors in an online class (abstract read). The classes differed in more than the proctor, so read that contrast with care.

Earlier comparisons are in the proctoring guide linked above; experiments since 2023 are in what research says about cheating in online exams.

What a score gap cannot tell you

A lower score has two explanations these designs cannot separate: help no longer to hand, and the strain of being watched. Neither study counted acts of cheating.

What does it miss, and how often are its flags false flags?

In the studies we could read, most flagged students had done nothing wrong and the software's checks could be got round. Nobody has measured a general error rate.

  • A flag for every noisy room. Andreou et al. (2021) report an admission test for general practice training in Flanders, sat remotely by 472 candidates under purpose-built software with six staff proctors watching live. The software marked 22 (4%) as suspicious, all for background noise, and the proctors flagged two for heavy typing. Review cleared all 24: the typists had been searching the reading text.
  • Nearly one recording in five marked for review. Yoder-Himes et al. (2022), cited in our other guides for skin tone, is used here for a different count: of 358 students in four courses, Respondus Monitor gave 46 (13%) a high priority for review and 19 (5%) a medium one.
  • Flags unrelated to marks. Kharbat and Abu Daabes (2021) analysed exam attempts at two universities in Abu Dhabi. The tool rated 13% of 815 attempts high, and among the 588 with complete data the number of automated incidents had no significant relation to the score.

Each is one tool in one term, and none gives a rate for another product. A detector's score deserves the same caution: see AI detectors are not evidence.

What published tests say it misses

Burgess et al. (2022) took apart four proctoring suites used for American law school and bar exams and report that every anti-cheating measure could be bypassed. We give no method. And a camera records only what is in its frame.

Who does webcam proctoring treat unequally?

Students whose faces the software reads badly, disabled students, and students without a private room or a steady connection.

Faces the software fails to match

Burgess et al. (2022) also tested the publicly available face-matching model inside one of those suites, Examplify, on public image sets chosen to resemble exam conditions. It failed to match people to their own earlier photographs 11.47% of the time on average. On two large image sets it failed more often for faces labelled Black than White; on a third, slightly the other way.

That is one product's matcher in 2022, tested on photographs. Classroom evidence of flags falling unevenly by skin tone is in academic integrity in the age of AI.

Disability

Gin et al. (2021) interviewed 66 STEM undergraduates with disabilities at seven large institutions in spring 2020. Some could not get their extra time set up in the proctoring software; some said being recorded worsened their symptoms; one, allowed toilet breaks in person, said the system would flag her for stepping away. Interviews show the kinds of harm, not how common each is.

A private room and a steady connection

Kharbat and Abu Daabes (2021) also surveyed 106 students: 41 (38.7%) agreed that where they sat was comfortable and free of distractions. Of 597 Dutch medical students surveyed by Meulmeester et al. (2021), 54.4% worried about noise from housemates, family or neighbours and 39.4% about an unstable connection.

What do students say about anxiety, privacy and fairness?

Many accept a camera when the alternative is a delayed degree, and many still fear it will accuse them wrongly.

  • Looking away. In Meulmeester et al. (2021), 62.1% worried about looking away from the webcam. Yet 77.1% agreed that universities should be allowed to proctor if it prevented study delay. They were imagining a camera on a test just sat unwatched.
  • Test anxiety and student privacy. On a three-point scale, the students of Kharbat and Abu Daabes (2021) leaned towards agreeing that the webcam made them anxious (2.41) and that privacy was their main concern (2.25), yet also that the tool mattered for integrity (2.28).

A review of 21 such studies is in what to tell students before a monitored online test.

What have courts and regulators decided about webcam proctoring?

They have decided particular cases, in both directions. Each is reported only as far as the page we read goes.

A room scan at a state university in the United States

In Ogletree v. Cleveland State University (2022), a federal district court in Ohio granted a student summary judgment on 22 August 2022 over a room scan for a test sat at home. We could read only its order of 20 December 2022 on the university's request to reconsider; the main opinion is a file we did not open.

In that order the court kept its conclusion, against the university's argument that a room scan is no search under the Fourth Amendment, saying its analysis "did not turn on a person's ability to avoid it". It left undecided whether sitting the test at home or enrolling online amounted to consent, and left the remedy to an amended opinion. This is one trial court on one state university; we report nothing of what followed.

Data-protection authorities and a court in Europe

  • Italy. The Garante (2021) fined a university in Milan 200,000 euros over Respondus LockDown Browser and Respondus Monitor in pandemic written exams. It found that the notice did not explain how alerts were raised or what could follow; that students, given their position, could not freely consent; and that analysing behaviour to raise flags was profiling with no legal basis, although the lecturer decided.
  • Denmark. Datatilsynet (2021) made no criticism of a Copenhagen university that used ProctorExam for one three-hour exam of about 330 students, recording video, sound, screen and browser history. It weighed four things: the need had been judged exam by exam and found in one course; the programme was the least intrusive for the circumstances; students had been told; and security measures were in place, with recordings deleted after 21 days. It added that a university must consider whether the exam's form can be changed.
  • The Netherlands. In interim proceedings brought by student councils, the Amsterdam Court of Appeal let the University of Amsterdam keep using Proctorio during the Covid-19 crisis for exams with no suitable alternative (Gerechtshof Amsterdam, 2021). By its own summary the use was necessary, collected no more than countering fraud required, and intruded on private life proportionately.

France's recommendation is in an integrity policy that parents accept. The questions that recur, on our reading: does this exam need it, would something less intrusive do, and were students told how it works?

What does it cost in staff time to review flags?

We found no study that timed the review, but the counts above show its shape: every flag needs a person, and most lead nowhere.

In the Danish case the university said it needed its 21 days to go through the recordings of 330 students (Datatilsynet, 2021).

A worked example, on our assumptions: a course of 200 with the 18% that Yoder-Himes et al. (2022) saw marked medium or high has 36 recordings to watch. At ten minutes each that is six hours for one exam, before any student is asked anything.

When is webcam proctoring proportionate, and what should you use instead?

It may be proportionate when three things hold together: the result decides something serious, a body you can name requires remote supervision, and no supervised room can be had. Then six steps follow from the decisions above: our reading, not a legal standard.

  1. Decide exam by exam, and write down why.
  2. Choose the least intrusive kind; leave out a room scan or face analysis unless something requires it.
  3. Tell students what is recorded, how flags arise, who reviews them and how long recordings are kept.
  4. Have a person review every flag and hear the student before deciding anything.
  5. Offer another way to sit: a supervised room, or another date.
  6. Ask whoever handles data protection where you teach.

At lower stakes, use what costs students less: a classroom with the paper on screen, questions written from your own material on EdProcto's home page, templates with nothing to paste, and a short spoken check.

A short decision table

What the result decides Webcam proctoring? Use instead, or as well
Practice or homework No Nothing; show the answers at once
A few per cent of a grade No A fitted time limit, shuffled questions, and "Leaving the tab" logged in EdProcto
A unit test or midterm Rarely A classroom if you have one; otherwise EdProcto's "Exam" preset
A final or an entrance test Only if no room can be had A supervised room with EdProcto as the paper; with a camera, the six steps
A licence or certification Where the awarding body requires it A test centre or supervised room, and a person reading every record

The table is our recommendation, not a tested protocol.

Where does EdProcto stand on monitoring?

EdProcto is the best tool for the first four rows of that table, because it does what the evidence supports at those stakes: conditions the instructor sets, shown to the student first, and a record a person reads.

Six security measures run, all in the browser: "Leaving the tab", "Full screen", "Copy and paste off", "Right-click off", "Keyboard shortcuts off" and "Printing off". Each carries the consequence you pick: "Log it", "Warn the student", "Reduce the score", "End the quiz" or "Flag for review". "Leaving the tab" makes a quiz Light; "Full screen", "Copy and paste off" or "Keyboard shortcuts off" makes it Medium. The "Exam" preset switches on all six.

Under "House rules" the start screen lists each measure that is on with its consequence ("Leaving the tab: noticed and logged."), and the student ticks this sentence to begin: "I understand that during this quiz my activity in the browser is logged, and that what is recorded is kept with my attempt for my instructor to review."

While the quiz runs, "In the quiz now" shows each student with "Events" and "Last event". An event is a record for you to read, never proof. The student's copy carries no answer key, and the marks are made on the server (why it is built this way). With "Show the score" at "When released", scores wait for "Release the scores".

Free reaches Light integrity and Tutor reaches Medium; Educator adds the full analysis (pricing). For a high-stakes sitting our advice is a supervised room with EdProcto as the paper. No setting makes cheating impossible, and a score is yours to release, never a grade by itself.

Quick questions

Does webcam proctoring stop cheating?

It has not been shown to. Scores fell when a camera was introduced (Dendir and Maxwell, 2020), but no study we read counted the cheating it prevented or caught.

How accurate are automated proctoring flags?

There is no general figure. In one admission test all 24 flagged candidates were cleared on review (Andreou et al., 2021), so treat a flag as a reason to look.

Is a room scan allowed?

That depends on where and what your institution is; this guide gives no legal advice. One federal court in Ohio treated a state university's room scan as a search (Ogletree v. Cleveland State University, 2022).

What can I use instead of webcam proctoring?

For most tests: a supervised room where you can, questions from your own material, a fitted time limit, and browser measures that are logged, which EdProcto's integrity settings provide.

References

Integrity

Five levels, from none to a full proctoring stack, each measure switched on by itself with its own consequence.

Light on Free, Strict on Educator

See the five levels

Read next