Can interviewers detect an AI interview assistant?

The two detection surfaces

Technical detection is what software can observe: processes on the machine, windows in a captured frame, network activity, browser focus events. Behavioural detection is what a human notices: pauses in the wrong places, an answer that does not match the resume, prose that sounds nothing like the candidate speaking a minute earlier.

These are not equally important. In an ordinary video interview — no proctoring agent, no lockdown browser — the technical surface is close to empty, because a standard conferencing client has no visibility into what else runs on your computer. The behavioural surface, meanwhile, is wide open and is where essentially every real-world case of "we could tell" comes from.

What a standard video interview can and cannot observe

For a normal Zoom, Teams, Meet or Webex call with no additional assessment software installed:

• Cannot see your other applications

  A conferencing client is a sandboxed application. It does not inventory your running processes and has no mechanism to report what else is open.

• Can see exactly what you share

  When you share your screen, the client sends the captured frames. Anything in those frames is visible, which is the entire problem capture exclusion addresses.

• Can see your camera

  Eye movement, and the reflection in glasses on a high-resolution feed, are observable by a human. This is the most common way people are noticed.

• Can hear pauses

  Audio is the richest signal in the call. Silence in the wrong place is more conspicuous than a wrong answer.

What proctored assessments add

A locked-down coding assessment or a proctoring platform is a different environment, and the honest answer for these is different too.

• Process and window inventories

  Proctoring agents run with elevated permissions specifically to enumerate what is running. This is a category of observation an OS window flag has no answer to.

• Focus and clipboard monitoring

  Assessment platforms commonly log tab switches, focus loss and paste events, and flag them for review.

• Continuous camera and screen recording

  Recorded for later human review, often with automated flagging of gaze direction.

• Terms you agreed to

  Proctored assessments come with explicit rules you accept before starting. CueNova does not support this use case.

The behavioural signals that actually give people away

Across the reports that circulate publicly, almost none involve software detecting software. They involve a human noticing something.

• The four-second silence

  A question lands, then nothing for four seconds, then a fluent and complete answer. The gap is the tell, not the answer. This is why an assistant that streams the opening of a response immediately is more useful than one that returns a perfect paragraph late.

• Register that does not match

  Conversational speech for small talk, then a sudden shift into written-essay vocabulary. Read structure, speak normally.

• Answers unmoored from the resume

  A generic best-practice answer where the interviewer expected the specific project on page one of your CV. Grounding answers in your own material is what prevents this.

• Collapsing on the follow-up

  The single most common failure. A strong first answer followed by an inability to defend any part of it. If you cannot explain a point, do not make it.

• Visible reading

  Eyes tracking left to right on camera. A glanceable overlay near the camera is read differently from a wall of text in a corner.

What this means in practice

The candidates who get caught are, overwhelmingly, the ones who tried to read a script. The ones who use assistance as a prompt — a structure, a reminder of the trade-off they had forgotten, the name of the pattern they were reaching for — are working within the part of the interview that no detection surface addresses, because thinking with a reference is not a detectable event.

It also means the most valuable thing a tool in this category can do is not hiding. It is shaping the answer so you can say it in your own voice, from your own experience, and defend it when the follow-up comes.

Whether to use assistance at all depends on the rules of the interview you are in, and knowing those rules is your responsibility. Many people use tools like this exclusively for practice rounds and mock interviews, where the question does not arise.