AI proctoring — automated exam monitoring at scale

What AI proctoring actually does

AI proctoring watches the webcam feed for the length of the exam and, where screen monitoring is turned on, the shared screen too. It looks for the kind of anomaly that would make a proctor look twice — a different face appearing on camera, the participant leaving the frame, or a break in the normal audio and video pattern of an exam session. It also handles identity verification at the start of the session, matching the participant against an ID document — the same check available with a live proctor. When nothing unusual happens, nothing happens: no one is watching in real time, and the exam runs uninterrupted.

That's also its limit. AI proctoring flags; it doesn't judge. It can't tell a participant reaching for a glass of water from one reaching for a phone, so it flags both and leaves the distinction to a person — it never removes someone from an exam or decides on its own that something was wrong. That call sits with the exam provider, informed by a human reviewer's read of the flagged moment. Because it has no situational understanding of a room the way a live proctor watching in real time does, it fits high-volume, routine-stakes testing better than exams where a wrong call is expensive. For those, human proctoring, or a mix of both, is the better starting point.

How flagging and human review actually work

Flagging happens automatically, during the exam, without anyone watching live — that's the whole point of the cost saving. When the system raises a flag, it doesn't reach a conclusion on its own: it adds the moment to a queue that a trained member of PRUEFSTER's support staff works through. That's the same support team behind human proctoring — support doesn't change depending on which product you buy, only whether a proctor is also watching live.

Review happens after the exam, not during it, which is why the comparison above marks AI proctoring's post-exam review effort as higher: someone still has to look at what was flagged, just later rather than in the moment. A reviewer watches the flagged clip, decides whether it's worth escalating, and, where it is, writes up what happened for the exam provider to act on. Most flags turn out to be nothing — someone stretching, a shadow crossing the webcam, a brief connection drop — and are closed without further action. The ones that aren't get the same trained-staff attention a live proctor would have given them in the moment, just on a delay.

This is why AI proctoring pairs well with occasional human proctoring, or with the 360° room scan for exams where the stakes justify it: the flagging catches what happened, but someone still has to decide what it means.