A candidate answers every question in a smooth, complete paragraph. The answers are correct, well-structured and oddly generic. Three weeks later, in the client's office, the same person cannot explain a basic concept from their own CV.
If that sounds familiar, you are not alone. AI help during online interviews has moved from rumour to routine, and the usual video call was never built to catch it.
How common is AI cheating in interviews?
The numbers have climbed fast:
- 35% of proctored test sessions showed cheating or fraud attempts in 2025, up from 16% in 2024, according to CodeSignal data compiled by HeroHunt.
- 22% of US job seekers said they use AI during live interviews in a 2026 Resume Genius survey, reported by Newsweek and compiled by HeroHunt.
- 65% of hiring managers said they had caught applicants using AI deceptively, in a Greenhouse survey of 4,136 people (HeroHunt).
- Gartner found that 6% of 3,000 candidates admitted to interview fraud — posing as someone else or having someone pose as them (HR Dive).
Junior and technical roles are hit hardest, because the questions are predictable and the answers are easy for an AI model to produce.
The four ways candidates use AI in an interview
- A hidden overlay on the same screen. Tools such as Cluely and Interview Coder listen to the interviewer, send the question to an AI model and show the answer in a window on the candidate's own screen. They are sold as undetectable, and the window is built to stay out of a screen share. We cover these tools in detail in Can interviewers detect Cluely and Interview Coder?
- A second device. A phone or tablet below the webcam, running a voice-mode AI assistant that hears the question through its own microphone.
- A second screen. An extra monitor with a chatbot open, out of the webcam's view.
- A person. Someone in the room, on an earpiece, or a stand-in who takes the whole interview. That last one is a proxy interview — see how to detect proxy interviews.
Warning signs to watch for
None of these proves anything on its own. Together, and repeated, they are worth a closer look.
- Reading eyes. The eyes sweep left to right in short steps and jump back to the start of the next line — the pattern of reading, not of thinking. Thinking eyes drift up or away and come back.
- Flat answer timing. A pause of three to five seconds before every answer, whether the question is "What is your name?" or "Explain eventual consistency". Real people answer easy questions fast and hard ones slowly.
- Fluent, then lost. A perfect first answer, then confusion at a simple follow-up about their own example.
- Repeating the question aloud. Buying time while the tool transcribes.
- Text that appears without typing. A whole function or paragraph lands in the code pad at once.
- A partial screen share. The candidate shares one window or one browser tab instead of the whole screen.
- Answers that don't match the CV. Textbook knowledge, but no story about their own project.
Why screen sharing doesn't catch it
Operating systems let an app keep its window out of screen capture. That was built for good reasons — so a password manager or a banking app does not appear in a recording. Interview copilots use the same capability, so the interviewer's view of the shared screen shows a clean desktop while the candidate reads an answer from it.
A webcam pointing at the face does not help either: it cannot see the screen, the desk, or the phone below it.
Checks that work
1. Add a second camera. The candidate's own phone, placed beside them, shows their desk, keyboard, hands and screen from the side. An overlay hidden from screen share is still light on a monitor, and a camera sees it. So does a phone on the desk or a second screen.
2. Watch the eyes, on the candidate's device. Eye-movement analysis can separate reading from thinking. Doing it in the candidate's browser means no video has to leave their machine — only the signal and its time.
3. Time the answers. Measure the silence between the end of the question and the start of the answer, and how unbroken the answer is. A flat delay on every question, followed by a fluent monologue, is a strong pattern.
4. Ask follow-ups an AI can't predict. "You said you cut the query time in half — what was the slowest query, and what did you change first?" A candidate who did the work answers with detail. A candidate reading a summary does not.
5. Require the whole screen. Block a single-window or single-tab share, and check how many displays are connected.
6. Check typing in code rounds. Record whether code was typed or pasted. A shared code pad, run in the room, makes this simple.
7. Close the loop in the final round. Build a short in-person check from the candidate's own online answers. If they could explain it online, they can explain it across the table.
What not to do
- Don't reject on one signal. Poor lighting, a slow connection, a disability or simply the way someone thinks can trigger an automated check. A signal is a reason to look, not a verdict.
- Don't monitor in secret. Tell candidates what is watched and kept, before they start. It is fairer, it is what India's DPDP Act expects, and it deters most cheating on its own.
- Don't rely on AI-text detectors. They misfire on clear, correct human writing.
How Oximus Pro handles it
Oximus Pro turns the candidate's phone into a second camera, reads eye movement and answer timing on the candidate's device, flags text that appears without typing and partial screen shares, and gives your panel a timeline of signals with pictures. The interviewer decides. For the final round, it writes a five-minute check from what the candidate said online. See how it works or book a demo.
Frequently asked questions
Can interviewers tell if you are using AI?
Often, yes. Reading eye movements, a flat delay before every answer and weak follow-up answers are noticeable to a trained interviewer, and a second camera shows what a screen share hides. No single check is perfect, which is why several are used together.
Is using AI in a job interview cheating?
If the employer has not allowed it and the candidate presents AI-generated answers as their own, most employers treat it as misrepresentation. Some companies now allow AI in parts of a technical round; the rules should be stated to every candidate in advance.
What is the best software to detect AI cheating in interviews?
Look for a tool that adds a second camera, analyses eye movement and answer timing, detects pasted text and partial screen shares, and leaves the decision to a person. Avoid tools that only analyse the text of answers.
Do candidates have to agree to interview monitoring?
They should be told clearly what is monitored and kept, and asked to agree, before the round starts. Under India's Digital Personal Data Protection Act, 2023, the hiring company is responsible for giving that notice.
