The Four Ways Candidates Hide AI Help (and Which Ones Software Can Actually Catch)

Not all interview cheating is the same, and the differences matter more than they look. Where the AI help is displayed determines whether any detection software can see it at all. Get this map straight and you'll understand both what's catchable and — just as important — what isn't, so you can cover the gaps honestly.

There are four broad ways AI help reaches a candidate during a live interview.

1. Invisible overlays on the candidate's own screen

The most common method. A desktop app shows AI-generated answers in a window engineered to be invisible to screen sharing — the candidate sees it, the interviewer doesn't. This is the Cluely-style approach, and the open-source clones of it are everywhere.

Can software catch it? Yes — this runs on the candidate's machine, so the right kind of detection identifies it and proves it was there. The catch is that detection has to not depend on the tool's name, because these disguise and rename themselves freely.

2. Browser-based helpers

A growing class runs entirely inside a browser tab — listening to the interview, generating answers, displaying them in the same tab — with nothing installed. Others operate as browser extensions.

Can software catch it? Yes. A browser-based helper is active on the machine and leaves traces of its activity. Detection flags that a browser is doing something it shouldn't during the interview — without watching what the candidate is viewing or browsing.

3. Environment manipulation

Some candidates run the interview inside a virtual machine, use remote-control software so a helper elsewhere can assist, or manipulate the video feed itself. This is the most technically involved category, and it overlaps with the most serious fraud — fake and proxy candidates.

Can software catch it? Yes. Remote sessions, virtual machines, and manipulated camera feeds all leave clear, detectable signals on the machine.

4. Off-device assistance

The hardest category, and the one any honest vendor has to be straight about. The candidate keeps the interview machine clean and receives help on a separate device — a second phone or laptop showing answers, or an earpiece feeding them audio. Nothing runs on the interview computer.

Can software catch it? No — and you should be skeptical of any vendor that claims its agent sees a phone sitting off-camera or an earpiece in someone's ear. No software on the interview machine can. This gap is real.

So how is it closed? Not by detection, but by how the interview is run. An off-device assistant depends entirely on receiving the question — by hearing it or seeing it. A visual-first interviewing approach, where key elements are shown rather than spoken and details change live, denies that assistant the context it needs to keep up. The help arrives too slowly or too incompletely to matter. It's a process answer, not a software one.

Putting the map together

Three of the four categories live on the candidate's machine and are catchable with detection that doesn't depend on the tool's name. The fourth lives off the machine and is closed by interview methodology, not software. Any honest approach to interview integrity has to do both — and any vendor claiming a single piece of software catches all four, including the phone on the desk, is overselling.

That two-part split — detect what's on the machine, neutralize what's off it — is exactly how a complete interview-integrity process should be built.

Capifiq detects the on-machine categories with timestamped evidence and equips interviewers to close the off-device gap. Try it free on five interviews.