TL;DR
A short summary of where ChatGPT cheating detection in interviews stands in 2026:
- One in five US professionals admit they have secretly used AI tools in job interviews. 55% of surveyed candidates agree AI use in interviews has become "the new norm." This is no longer a rare problem.
- In an experiment by interviewing.io, candidates using ChatGPT solved verbatim LeetCode questions correctly 73% of the time. Interviewers did not suspect misuse in any of the sessions. Human detection on its own has failed.
- Most articles list the wrong warning signs. Eye movement, tab switching and "polished answers" are easy for practiced cheaters to get around within a day. The checks that catch ChatGPT use look at other things. They look at how long a candidate pauses, speech that is too smooth, words that do not match the resume, how many people are speaking, lip-sync, and follow-up questions the candidate cannot prepare for.
- Software that only scans text for AI writing fails. OpenAI shut down its own AI text detector because it wrongly flagged too much human writing. Detection has to combine several checks on the conversation, the audio and the candidate's behavior.
- Recorded video interviews are now easier to cheat on than live ones, because the candidate controls the recording. Voice-based AI interviews that adapt to each answer remove the rehearsed answers that ChatGPT is good at.
- The setup that works best combines four layers of checks: proctoring of the screen and room, audio analysis, AI plagiarism analysis, and interview questions that adapt to each candidate.
- Goodfit's voice interviews ask follow-up questions based on each answer and flag suspicious moments. Each flag links to that moment in the recording, so one odd pause does not sink an otherwise strong interview. Your team makes the final call.
Why ChatGPT cheating in interviews became a real problem
For about 15 years, the main cheating worry was someone whispering answers off-camera or holding up a printout. Both were rare and obvious when they happened. ChatGPT changed the problem in three ways at once.
The first is access. Anyone with a phone, a second monitor or a Bluetooth earpiece can run ChatGPT live during an interview. The candidate does not need a friend, a script or technical skill. They need a free OpenAI account.
The second is fluency. Before 2023, a candidate cheating with a search engine read out answers that sounded like Wikipedia. The phrasing was stiff and formal. ChatGPT gives answers in a casual, spoken style, with natural pauses and transitions when prompted well. An interviewer listening for stiff answers will no longer catch it.
The third is volume. The same candidate can apply to 50 jobs in a day with AI-written cover letters. They can then walk into 50 first-round interviews with AI help ready. The cost of trying to cheat has collapsed. The cost of catching it has not.
A 2025 survey by Sherlock AI found that 20% of US professionals have secretly used AI in job interviews. 55% of surveyed candidates agreed AI use in interviews has become "the new norm." A Resume Builder study put the number at 41%. The interviewing.io experiment shows the problem most clearly. Candidates using ChatGPT solved verbatim LeetCode questions correctly 73% of the time, and interviewers did not suspect cheating in any of the sessions.
What candidates are actually doing
Detection has to start from a real picture of how candidates cheat. The methods have changed.
- Live answers on a second screen. This is the most common method. The candidate has ChatGPT open on a phone, a second monitor or a hidden browser tab. Basic proctoring catches it.
- Answers relayed by a friend. A friend listens to the interview on a phone call, types the questions into ChatGPT, and feeds answers back by earpiece or chat. Tab-switch checks miss it. Audio analysis catches it.
- Rehearsed answers written in advance. Before the interview, the candidate gets ChatGPT answers to every likely question and reads them out. This beats most recorded video interview platforms entirely.
- Impersonation. Someone other than the real candidate takes the interview. This is more and more common in entry-level remote technical hiring.
- Bluetooth earpiece coaching. A hidden earpiece relays answers from a coach or an AI tool. Catching it takes lip-sync analysis, a count of how many people are speaking, and checks for speech that is too smooth.
- Live AI interview tools. A new kind of tool, such as Cluely, reads the screen during the interview and shows AI-written answers on top of it. These tools are built to get past traditional proctoring.
Why most cheating detection fails
Most basic guides on catching ChatGPT cheating repeat the same five tips. Ask follow-up questions. Watch for eye movement. Listen for answers that sound too polished. Require screen sharing. Ask candidates to think out loud. These tips work against a first-time cheater. They fail against anyone who has read the same article.
AI text detectors have been shown not to work. OpenAI shut down its own AI text classifier in July 2023 because of low accuracy. Independent studies keep finding false positive rates above 10%. These tools often label human writing as AI-written.
Recorded video interviews are easy to cheat on by design. The candidate controls the camera angle, the lighting, what sits off-screen and when they answer. On platforms that allow retakes, they also control the number of takes. HireVue, Spark Hire and similar tools are especially exposed.
Live interviews run by people miss smooth cheating. When a candidate uses ChatGPT well, the answer sounds natural. Interviewers are not trained to tell a well-spoken person from someone reading well from a screen. The interviewing.io experiment proved this.
Tab-switch detection has limits. It catches a candidate who moves to another browser tab. It does not catch a candidate using a second device. ChatGPT on a phone is invisible to browser checks.
Relying on one warning sign flags too many honest people. A candidate who pauses 6 seconds before answering might be cheating. They might also be thinking, nervous or on a slow connection. Treating any single sign as proof rejects honest candidates. Treating it as proof of nothing lets cheaters through.
Layer 1: Environmental proctoring
This is the layer most platforms say they have. Done well, it catches the bottom 50% of cheating attempts.
- Tab switch detection. Logs each time the candidate leaves the assessment tab. It should record when they left and how long they were away, as well as the count.
- Fullscreen enforcement. The assessment runs in fullscreen, and each exit is logged.
- Copy-paste blocking. Stops the candidate from pasting outside text into the assessment.
- DevTools detection. Flags when the candidate opens browser developer tools.
- Face detection. Flags when no face is visible, when the face changes during the interview, or when more than one face appears.
- Second device detection. Some platforms can spot a second screen in the webcam image, either in the room or reflected in the candidate's eyes.
Layer 2: Audio forensics
Most platforms do not have this layer. It is also where the most reliable warning signs show up.
- Silence ratio. Candidates using ChatGPT pause longer before they answer, and they do it every time. Across the whole interview, the share of silence tends to be 30 to 50% higher than normal.
- Speech that is too smooth. Candidates reading from a script or screen speak more smoothly than people do in real talk. Their answers lack the "uhs," "ums" and restarts of live speech.
- Speaker count (diarization). Audio analysis counts how many different people speak. Two speakers, the candidate and the AI interviewer, is normal. A third voice is a strong sign of fraud, especially one that lasts more than three or four seconds.
- Lip-sync analysis. The software checks whether the candidate's lip movements match the audio. This catches impersonation, pre-recorded audio and earpiece relays.
- Voice comparison. Some platforms compare the candidate's voice with a sample recorded earlier, during identity checks. This catches cases where someone else takes the whole interview.
Layer 3: AI plagiarism analysis
This layer looks at the words in the candidate's answers. The other layers look at the room and the audio.
- Signs of reading. Stiff sentence structure that does not sound like speech. Unnatural keyword density. Formal linking words like "furthermore," "consequently" and "in conclusion," which people rarely use when speaking off the cuff.
- Vocabulary mismatch. The resume says the candidate is a recent graduate from a little-known college. Their answers use the words and frameworks of a senior consultant. That gap between stated background and how they speak is a red flag.
- Rigid structure. ChatGPT answers follow set templates, such as "Three key reasons...," "First, second, third..." and "In summary..." Real spoken answers rarely follow such a strict shape.
- Repeated stock phrases. ChatGPT tends to repeat certain phrases, such as "It's worth noting that..." and "This is a critical consideration because..." Many of these phrases in one interview is a red flag.
Layer 4: Adaptive interview design
The most underrated layer is not technical at all. It is the interview itself. Verbatim LeetCode questions are easy to cheat on, because ChatGPT saw them in training. Changed questions are harder. New questions that test reasoning under pressure are hardest. The interviewing.io experiment showed this clearly. When interviewers asked unchanged questions from a known set, candidates using ChatGPT passed 73% of the time. When the same questions were changed, the pass rate dropped sharply.
- Build questions from the candidate's resume. Skip "tell me about a time you handled a difficult customer." Ask "you mentioned in your application that you handled a refund escalation at [specific employer]. Walk me through what happened."
- Ask follow-up questions based on the last answer. Follow-ups make the candidate build on their own answer in real time. A pre-written answer cannot do that.
- Change a constraint. In coding interviews, change one rule in a known problem ("now solve this, but the input has duplicates").
- Ask candidates to explain their answer. Have them talk through their reasoning out loud. ChatGPT answers often sound right, but the candidate cannot defend them when asked why they chose that approach.
What a real detection report looks like
A useful proctoring report does three things. It shows what was flagged, rates how sure the system is, and leaves the final call to the recruiter.
Interview integrity summary. A short summary at the top showing which checks flagged something. It includes the audio proctoring score, the AI plagiarism score and a count of screen and room incidents.
Audio breakdown. Silence ratio compared with normal. A fluency score, with the normal range for real speech marked. The number of speakers over the interview. Any part where the lip-sync match drops below the threshold.
AI plagiarism breakdown. Flags for each answer, with a severity score. The transcript highlights the flagged parts and names what triggered each flag, such as a vocabulary spike, rigid structure or a stock phrase.
Incident timeline. Every tab switch, fullscreen exit and face detection problem, with a timestamp. The recruiter can click any incident and jump to that moment in the recording.
Recommended action. The report avoids a yes-or-no verdict on cheating. It gives a risk level instead: "low risk, recommend advance," "medium risk, recommend a follow-up question in the next round," or "high risk, recommend manual review."
What does not work, and why
Several detection methods get recommended often and mostly produce noise.
"Watch for eye movement." This worked in 2023, when candidates were new to ChatGPT. By 2026, most cheaters have practiced reading from a second screen placed just below the camera. Their eyes look like they are on the interviewer.
"Ask candidates to think out loud." This helps in technical interviews where the reasoning matters. In most interviews, the candidate can simply read out reasoning written by AI.
"Use plagiarism detection software." Turnitin and similar tools were built for written essays and papers. They struggle with spoken interview transcripts.
"Ask candidates not to use AI." This is a request. It does not detect anything. Surveys show candidates use AI tools even when told not to.
"Trust your gut." The interviewing.io experiment proved this fails. Trained interviewers, told nothing about the experiment, did not suspect cheating in any of 32 sessions.
"Use AI text detectors after the interview." OpenAI shut down its own. Independent studies show false positive rates above 10% on human-written content.
How to actually build a cheating-resistant hiring process
Use voice AI interviews with follow-up questions instead of recorded video. No two interviews follow the same path, so rehearsed answers stop working. The candidate cannot prepare answers for questions that do not exist yet.
Run all four layers at once. Use proctoring of the screen and room, audio analysis, AI plagiarism analysis and adaptive questions. Each layer catches different cheating methods. Together they cover the main ways people cheat.
Rate each part of the interview instead of giving a yes-or-no verdict. One odd pause does not cancel an interview. A pattern of flags across several layers does. The recruiter sees the flagged moments and makes the final call.
Build questions from the real job and the candidate's resume. Candidates can rehearse generic questions. They cannot rehearse specific questions about the experience they claimed.
Check identity before the interview. Match a photo ID to the face in the interview. Compare a voice sample with the interview audio. This catches impersonation that no check during the interview can.
Treat detection as a set of layers. Vendors that list "AI fraud detection" as a single bullet point are usually matching keywords in the transcript. Real detection uses four separate systems, each with its own models and confidence scores.
How Goodfit handles ChatGPT cheating detection
Goodfit's cheating checks come with every web interview by default. You do not pay extra for them.
An AI interviewer runs the interview. It writes questions from the job description and asks follow-up questions based on what the candidate just said. This is the main defense against rehearsed answers. The candidate cannot prepare for a question that does not exist until they answer the one before it.
In web interviews, Goodfit flags a second person in the room, earbuds, reading from notes, a photo held up to the camera, and AI assistant tools in the browser. It also logs tab switches, fullscreen exits and pasting. Each flag has a timestamp.
On voice interviews, Goodfit runs a lip-sync check to see whether the voice matches the person on camera. These checks do not run on plain phone calls, so use the web interview when cheating is a concern.
Every answer is saved in the recording and the transcript. Recruiters can read the candidate's own words and judge whether an answer sounds read out or rehearsed.
The recruiter sees all of this in the candidate report. Each flag links to that moment in the recording, next to the transcript and the score for each skill.
Pricing is per assessment (see pricing). The first 20 are free on every account. The cheating checks are included and are not charged separately.
Frequently asked questions
How common is ChatGPT cheating in interviews?
Several 2024 and 2025 surveys put the share of candidates using AI in interviews between 20% and 41%. A Sherlock AI survey found that 55% of candidates believe AI use in interviews has become normal. In the interviewing.io experiment, interviewers failed to spot ChatGPT use in all 32 technical interviews. Whichever source you trust, the behavior is too common to ignore.
Can interviewers detect ChatGPT cheating without specialized tools?
Sometimes, but not reliably. In the interviewing.io experiment, trained interviewers using a standard interview format missed cheating in every session. Skilled cheaters put a second screen just below the camera, type prompts silently and read answers in a casual tone. Interviewers working without tools miss most well-run AI cheating.
What is the most reliable way to detect ChatGPT use in interviews?
Use several layers of checks together. Proctoring of the screen and room covers tab switches, fullscreen exits and face detection. Audio analysis covers silence ratio, speech that is too smooth, speaker count and lip-sync. AI plagiarism analysis covers signs of reading, vocabulary mismatch and rigid structure. Adaptive interview design means questions built from the candidate's resume and follow-ups based on earlier answers. No single check works on its own.
Do AI text detectors like GPTZero work for interview transcripts?
Not reliably. OpenAI shut down its own AI text classifier in 2023 because its accuracy was too low. Independent research has found false positive rates above 10% on human-written content. Interview transcripts are especially hard for these tools, because spoken answers are full of filler words and casual phrasing. If you rely on AI text detectors alone, honest candidates will get flagged by mistake.
How do you detect a candidate reading ChatGPT answers off a second screen?
Use several checks together. Eye movement analysis catches some cases. Silence ratio analysis catches the steady pause while the candidate types into ChatGPT and waits for the answer. Checks for speech that is too smooth catch a reading pattern that real talk does not have. Reviewing the words and structure in the transcript catches the rigid template of ChatGPT answers. Each check has gaps. Together they catch most cases.
What is the difference between cheating in live interviews vs recorded video interviews?
Recorded video interviews are easier to cheat on because the candidate fully controls the recording. They can pause, retake, place notes off-camera or read answers written in advance. Live interviews are harder to cheat on in real time. Candidates can still use earpieces, second devices or friends sending answers by chat. AI voice interviews that ask live follow-up questions are the hardest to cheat on, because no two interviews follow the same path.
Can voice biometrics catch a candidate getting someone else to take the interview?
Yes, when it is part of identity checks. The candidate records a voice sample before the interview. The interview audio is then compared with that sample, and a mismatch flags possible impersonation. This catches cases where someone else takes the whole interview. Voice matching does not catch a candidate who uses ChatGPT themselves.
How do you balance detection with candidate experience?
Follow two rules. First, run checks in the background where you can. Face detection, audio analysis and transcript review can all run without interrupting the candidate. Avoid stopping the interview to say "you switched tabs, explain yourself" unless you must. Second, rate each part of the interview instead of giving a yes-or-no verdict. A candidate with one minor flag is different from a candidate with three major flags. Matching your response to the flags keeps the process fair.
Should you ban candidates outright if cheating is detected?
It depends on how serious the flags are and how sure the system is. Strong flags across several layers, such as audio, plagiarism and screen checks together, probably justify disqualifying the candidate. One weak flag might call for a follow-up question in the next round instead. The platform's job is to show what it found. The recruiter's job is to make the call.
Does Goodfit support custom proctoring rules per role?
Goodfit's cheating checks run on every web interview. You decide how to act on the flags for each role. A senior leadership role might only need a light review. A high-volume frontline role might need a closer look at every flagged candidate. Automations can sort candidates by their flags, but any rule that would reject someone asks your team first.