How to tell if a candidate is using AI in an interview

You can usually tell a candidate is using AI in an interview from how an answer arrives, not from what it says. Watch for fully formed answers to questions nobody could prepare for, a sudden jump in polish partway through the call, and delivery that runs a beat behind the question. No single sign settles it. A follow-up built on the candidate's own answer usually does.

This isn't a fringe worry any more. In the Institute of Student Employers' 2026 survey, reported by Eastern Eye, 66% of UK graduate employers reported at least one case of a candidate using AI improperly in an interview. In a Resume Genius survey of 1,000 US job seekers, reported by Newsweek, 22% said they use AI during live interviews.

How can you tell if a candidate is using AI in an interview?

Look for a pattern across several answers rather than one odd moment. AI-assisted answers tend to be complete, structured and fluent on questions that should need thinking time, and they often sit a little apart from the rest of the conversation. These five signals are the ones worth noting:

Fluency where you'd expect thinking. Unprepared speech has false starts, corrections and the odd "um". A polished answer to an unpredictable question is worth a second look.
A shift in register. The candidate talks casually, then switches to textbook-precise language for one answer. The shift is the signal, not the polish.
Too much structure for an open prompt. Ask "walk me through how you'd approach this" and get a numbered, three-part answer with no pause. That isn't how most people think out loud. Many copilots now show answers as short bullet points, so the candidate can put them in their own words and not sound read. The wording changes, but the points still arrive in a tidy order, each one complete.
Answers that don't match earlier ones. They struggled with a basic concept ten minutes ago, then answer a harder one flawlessly.
Reading cadence and a lag. Copilots such as Cluely, Final Round AI and LockedIn AI have to hear the question before they can answer it. Watch for a stall ("great question, so…"), then evenly paced speech with pauses in odd places, eyes fixed on one spot.

What isn't a sign of AI use?

Some habits look suspicious and aren't. Treat these as normal unless they come with the signals above:

Rehearsed answers to common questions. "Tell me about yourself" should sound prepared. That's preparation, not AI.
Strong communicators. Some people are fluent. Fluency on its own isn't evidence.
A formal register from non-native speakers. If it's their style all the way through, it's their baseline, not a shift.
A pause before answering. One pause proves nothing. People think.

What matters is a change partway through the call, across more than one answer.

Why can't you just see it on screen?

You can't see the copilot on the candidate's screen, because it's built not to show. Most interview copilots run as an overlay that stays out of a screen share, and a candidate can always read from a phone or a second laptop instead. Your video platform only shows you what the candidate chooses to share. Our guide to the stealth AI interview tools in use today covers how they work.

Lockdown browsers and desktop proctoring apps can check what's running on the candidate's computer, but they mean installs for the candidate and IT, and they still miss a second device. Whatever the setup, the answer still has to be spoken out loud. That's why the conversation is the most reliable evidence you have. For platform specifics, see how to detect AI cheating on Zoom, Teams and Google Meet.

The conversation is also where the evidence can reach the interviewer in time. In the live interviews we run to vet candidates, our interview tech reads the conversation as it happens and flags answers worth probing, with the exact line and the timing behind each flag, while the candidate is still there to answer a follow-up, not in a recording afterwards.

What should you do if you suspect a candidate is using AI?

Probe, don't accuse. A copilot answers the question you asked; it struggles when the next question depends on the candidate's own words and experience.

Build on their answer. Pick one detail they just gave and go deeper: "Why did you choose that approach over the alternative?" or "What would break if the load tripled?"
Ask for the specifics only they'd know. Names of systems, the trade-off they lost an argument about, what they'd do differently now.
Interrupt gently. A short clarifying question mid-answer is normal in conversation. Someone reading tends to lose their place.
Write down the exact line and the time. If you raise it later, you'll want the evidence, not a feeling.
Ask openly at the end if it matters. "Did you use any tools during this interview?" is a fair question. A defensive non-answer tells you something too.

Don't make a hiring decision from one polished moment. Treat it as a reason to dig one level deeper, not a verdict. If you want fewer awkward conversations, set the rules before the interview: tell candidates whether AI help is allowed, and when.

For more on delivery, see signs a candidate is reading a script on Zoom.

Frequently asked questions

Do you need to know which copilot a candidate is using?

No. What you're judging is how each answer is delivered, not what's on their screen. That includes the bullet-point format copilots use so answers don't sound read: the pause while the points appear, points delivered in list order, and written phrasing in a spoken answer. A pattern is a reason to ask a follow-up, never proof on its own.

Can Zoom, Teams or Google Meet detect if a candidate is using AI?

Not by themselves. The platform shows you the candidate's camera and whatever they choose to share, and most copilots are designed to stay out of a screen share. You have to judge the answers themselves, as they're spoken.

Is it OK to ask a candidate whether they're using AI?

Yes. Ask it neutrally, the way you'd ask about any other part of their process, and ideally after you've set expectations at the start. How they answer is often more useful than the answer itself.

Should you reject a candidate if you think they used AI?

Not on a suspicion alone. Use it as a prompt for follow-up questions, in the call or in a second round. If the answers hold up under specific follow-ups, the candidate probably knows their stuff.

How we handle it

When we vet candidates, every live interview is read for the signals in this guide on every answer: the timing before it, a shift from how the candidate talks the rest of the call, written phrasing in speech, and points delivered in list order. Our interview tech reads the conversation rather than the screen, so nothing is installed and a copilot on a second device doesn't help. Anything flagged becomes a follow-up question in the call, and a person makes every decision.

We vet every candidate down to the interview before you meet them. See how we vet, or book a call.