What Is ChatGPT Interview Cheating?
ChatGPT interview cheating is when a real candidate, interviewing under their real identity, uses a generative AI tool during a live video interview or online assessment to generate or improve their answers in real time. That covers a candidate typing a question into ChatGPT off camera and reading the response, using ChatGPT’s voice mode through a hidden earpiece, or running a dedicated overlay tool that listens to the interview and streams suggested answers onto a second screen.
This is a different problem than deepfakes or proxy interviewing, where someone else is pretending to be the candidate. Here the person is genuine. The content of their answers is not.
How Common Is This Right Now?
It is more common than most hiring teams assume, and it is growing fast. Fabric’s analysis of 19,368 AI led interviews found that 38.5 percent of candidates showed cheating behavior, and cheating rates roughly tripled between July and September 2025 alone. CodeSignal has reported that cheating attempts on coding assessments more than doubled, climbing from about 16 percent in 2024 to roughly 35 percent in 2025.
Bloomberg’s July 2026 reporting on tools like Cluely and Interview Coder described a fast growing market of purpose built interview cheating software, some of it raising venture funding explicitly to help job seekers cheat on interviews and coding tests. A Greenhouse survey cited in industry coverage found that 65 percent of hiring managers say they have caught candidates using AI deceptively, while 22 percent of job seekers openly admit to using AI during a live interview. The gap between those two numbers is the real risk: a lot of AI assisted cheating is going undetected.
What Tools Are Candidates Actually Using?
Three categories show up repeatedly across 2026 reporting and vendor research:
- Dedicated interview cheating assistants, such as Cluely and Interview Coder, which listen to the interview audio, transcribe it in real time, and surface suggested answers on an overlay the interviewer cannot see on screen share
- Voice mode general purpose assistants, most often ChatGPT or Gemini, fed through a hidden earpiece or a phone propped just off camera
- Traditional low tech methods, like a second monitor with notes or answers pasted from a browser tab, which still account for a meaningful share of attempts even as dedicated tools grow
Fabric’s data breaks this down further: dedicated assistants accounted for about 45 percent of cheating attempts, voice mode LLM use for about 34 percent, and traditional tab switching or second screen methods for about 18 percent, with human accomplices making up the remaining small slice.
What Are the Warning Signs of ChatGPT Interview Cheating?
The clearest tells are behavioral, not visual. Watch for a consistent pause of three to five seconds before every answer, regardless of question difficulty. That gap is often the time it takes the AI tool to transcribe the question and generate a response.
Other signals recruiters and interviewers consistently report:
- Reading style eye movement, a steady left to right scan, rather than the upward or sideways glance people make when actually thinking
- Answers that sound like documentation, unnaturally structured and polished for spoken conversation
- An inability to explain or defend an answer when asked a natural follow up question
- On coding assessments, code that appears close to complete with almost no trial and error, deleted lines, or debugging pauses
- A tone or vocabulary shift between casual rapport building conversation and the actual answer to a technical or behavioral question
None of these signals alone is proof. Together, and especially when paired with proctoring data, they build a pattern worth acting on.
How Do You Detect Real Time AI Assistance During a Live Interview?
Behavioral observation catches some of it, but consistent detection needs a technical layer working alongside your interviewers. This is exactly what AI proctoring is built for: it monitors gaze direction, tab and window activity, audio anomalies, and device behavior throughout the interview or assessment, flagging patterns that a human interviewer would miss in the moment.
It is worth understanding the mechanics behind that monitoring, which glider.ai has written about in detail on how auto proctoring actually prevents cheating, including how browser lockdown, environment scanning, and activity logging work together rather than relying on any single signal.
One important caveat many competitor guides also raise: screen sharing alone is not reliable protection. Some overlay tools render at a level beneath what standard screen capture picks up, which means a candidate’s screen can look clean on a recording while an assistant is still feeding them answers. Detection has to combine technical monitoring with interview design that makes AI generated answers harder to disguise, which is covered next.
How Is This Different From Deepfakes or Identity Fraud?
ChatGPT interview cheating is a content problem, while identity fraud is a presence problem, and they need different defenses. A deepfake or a proxy interviewer means the person on the call is not who they claim to be at all. A candidate using ChatGPT is genuinely who they say they are; they are simply generating or improving their answers with AI in the moment.
That distinction matters for response. Identity fraud is addressed at the door, through steps like ID verification before or at the start of the interview. AI assisted content cheating is addressed throughout the interview itself, through monitoring, follow up questioning, and assessment design. Glider AI’s broader research on how candidates cheat in hiring assessments covers both categories; this post focuses specifically on the generative AI content layer.
What Should You Do When You Catch a Candidate Using ChatGPT?
Have a written policy before the interview even starts, so a flagged candidate is handled consistently rather than case by case. Most 2026 guidance, including from Incruiter and HeroHunt, recommends a graduated response rather than an automatic rejection, since some roles genuinely allow AI assisted work and the goal is honesty about it, not a blanket ban.
In practice, that looks like:
- Ask a natural, specific follow up question in real time. A candidate reading a generated answer usually cannot defend or extend it under a slightly different framing.
- Flag the session in your proctoring platform and review the recorded behavioral signals alongside the interviewer’s notes before making a decision.
- Separate coached, disclosed AI use, where a candidate says upfront they are using a tool, from hidden, undisclosed use, and weigh them very differently.
- Document the pattern. A single ambiguous pause is not a finding. A cluster of signals across the interview is.
How Can You Redesign Interviews So AI Assistance Matters Less?
The strongest long term fix is designing interviews where generated answers are less useful, not just harder to hide. Live, structured follow up questioning is the single most effective tool interviewers have, since generic AI output falls apart under specific, in the moment probing. Assessment formats also matter: proctored, scored exercises like Candidate 360 combine live monitoring with a structured task, which is harder to game than an open ended conversational answer.
For scaled screening, one way video interviews paired with proctoring let teams review response patterns methodically rather than relying on a single live judgment call. And the case for proctoring generally, beyond just this AI moment, is well established; glider.ai’s research on the benefits of remote proctoring in the modern interview process covers why this layer pays off across the whole hiring funnel, not just for catching AI use.




