Candidates are using AI that listens to the interview and feeds them answers, and the interviewers on the other end of the video call are catching some of them by ear. The giveaways are a delay before each answer, a filler phrase that praises the question while the app works, and language that is too polished for someone speaking off the cuff. In UK graduate hiring, 66% of employers have reported at least one candidate misusing AI in an interview.
The giveaways belong to the current tools
Think about what the app has to do. It hears the question, turns the speech into text, sends the text to a language model, waits for the answer and puts it on the candidate's screen, out of sight of screen sharing. Each step takes time, and the pause an interviewer hears is all of those steps added together. Each step is also getting faster, so the pause shrinks toward the length of an ordinary breath. The stiff wording goes the same way: it is the model's default voice, and one line of instruction to speak plainly and briefly changes it.
That has a cost on the other side. A recruiter who ends a call because the candidate paused and sounded formal will, before long, be ending calls for the wrong reason. People who are nervous, or who are interviewing in their second language, pause to find words and speak carefully. They look exactly like the tell, and the cheats will not.
A question with a stock answer tests the app
The more lasting fix is in the question. "How do you handle conflict in a team?" has a well-known good answer, and a model produces it in seconds. It tells the interviewer how good the app is.
Take a shop owner hiring a shift supervisor. She asks instead: "Think of the last time you had to change the rota with a day's notice. Who was unhappy?" Then she follows what the candidate says. "What did you say to that person?" "What did they do next?" "What would you change now?"
Every follow-up depends on the sentence before it, so the app has to hear the answer, guess at details it cannot know, and keep the story consistent across three layers. Someone who lived it answers quickly and adds things nobody asked for, like the name of the colleague who covered. An invented story gets thinner with each question. This takes no more of her time than six stock questions would: two deep questions with follow-ups fill the same forty-five minutes, and she learns more.
What it means for candidates
If you are the candidate, the app's real value is before the interview. Use AI to rehearse the stories only you have: ask it to question you about a project you ran until you have remembered the numbers, the awkward person and the decision you got wrong. That prepares you for the follow-ups a good interviewer asks, and nobody can accuse you of anything. Interviews that allow AI openly already grade judgment instead of answers, and that kind of preparation is what they reward.
There is a second reason to avoid the shortcut. Someone who talks their way into a job on borrowed answers ends up doing work they were never suited for, and both they and the employer pay for it.
An interview is a test of whether a person can talk about their own work. AI has made general answers free, so only specific ones now tell the interviewer anything.