What has changed
Proctoring used to mean checking that the webcam was on and nobody else was in the room. That was enough in 2019. It is not enough in 2026.
Here is how people cheat now. The candidate has ChatGPT open on a second device or in another browser tab. They copy the question, get an answer and read it back. The webcam looks normal the whole time. Webcam-only proctoring catches none of it.
A less obvious kind of cheating is live coaching. A friend or paid helper listens on a second device and whispers answers. The candidate is alone in the room, so eye tracking and webcam photos do not flag it. To catch this, you need to check the audio and how the candidate behaves.
The warning signs that matter now
Browser checks show what the candidate's device is doing, and they catch cheating as it happens today. A good proctoring system tracks 12 types of violations in three groups: browser behavior, what the camera sees and what the microphone hears.
- Leaving the tab or window, which flags anyone switching to ChatGPT mid-answer
- Copy and paste, in both directions
- Unusual keyboard or input-device activity
- Screen sharing, which catches candidates streaming to a helper
- Face checks: missing candidate, more than one face, or a different face
- Lip-sync checks, which catch a candidate reading an AI answer aloud
Flag does not mean fail
A flag means a person should look at that moment. It is not an automatic rejection. Good candidates trigger false alarms all the time. Their room is noisy, they clicked away to check a detail, or their cat walked across the keyboard. Proctoring should point reviewers to the right moments quickly.
Every flag should have a time stamp and a link to that point in the recording. A reviewer can then jump to it in seconds. A 30-minute test with 4 flags should take 3-4 minutes to review.
Give flags severity levels. One tab switch is low severity, since the candidate may have checked the time. Three tab switches during one coding question is high severity. Let reviewers focus on clusters of serious flags instead of reading every small alert.
What to tell candidates
Before the test, tell candidates what is monitored, what a flag means and what happens if one fires. Candidates who know the rules do better and feel respected. Hidden monitoring feels hostile, and fewer people finish.
One short paragraph works: "This assessment records your screen and webcam. Our system flags unusual activity, like new tabs or copy and paste, for a person to review. If your cat walks across the keyboard, that is fine. Do not use ChatGPT or a second device."
Telling candidates also protects you legally. In many places, monitoring people without telling them creates compliance risk. A clear statement up front removes doubt. It also builds trust with candidates who have nothing to hide.
What not to proctor
Do not proctor casual take-home tasks, portfolio reviews or reference checks. Save proctoring for timed tests taken on the platform, where cheating would change the result.
Do not proctor interviews with people either. In those, the conversation itself tells you what you need. Proctoring belongs on coding tests, timed skill assessments and standard cognitive tests. Leave it off panel interviews.
Too much proctoring makes candidates feel watched. If every step feels like surveillance, strong candidates will drop out, because they have other options. Use proctoring only on the steps where cheating would change the outcome.
Frequently asked questions
Why is webcam-only proctoring no longer enough?
Cheating now happens on the device, not in the room. A candidate with ChatGPT in a second tab or on a second device looks normal on webcam. To catch it, you need browser checks for tab switches, copy and paste and screen sharing. You also need audio checks to catch live coaching.
How do you detect ChatGPT use during an online assessment?
Track three things. Browser behavior covers tab switches, copy and paste in both directions and unusual input. The camera check looks for a missing face or more than one face. The audio check includes lip-sync analysis, which catches candidates reading AI answers aloud. A good system tracks 12 violation types across these three groups.
Should a proctoring flag automatically reject a candidate?
No. A flag means a person should review that moment. Good candidates trigger false alarms all the time, like a noisy room or a quick look at the clock. Every flag should have a time stamp and a link to the recording, so a reviewer can check it in seconds. Focus on clusters of serious flags, not every alert.
Should you tell candidates they are being proctored?
Always. Candidates who know the rules do better and feel respected. Hidden monitoring feels hostile, lowers completion and creates legal risk in many places. One clear paragraph before the test is enough. Say what is monitored, what a flag means and what happens next.
Which assessments should be proctored?
Proctor timed tests taken on the platform, where cheating would change the outcome. That means coding tests, timed skill tests and standard cognitive tests. Do not proctor take-home tasks, portfolio reviews or interviews with people. Too much proctoring drives strong candidates away.