Interview Cheating Detection: A Recruiter’s Practical Guide
Interview cheating detection works best as layers, not a single trick. Set clear rules for candidates, use questions that are hard to outsource, look at the environment and the answer content instead of body language, and verify with a follow-up before you reject anyone. This guide shows how to do that in live and recorded video interviews, and where the legal limits are in the US.
TL;DR
- Define what counts as outside help in writing before the interview, including AI tools, notes and other people.
- Interviewers are poor lie detectors: research cited below puts detection of faking tactics at roughly 12% to 18.5%. Don’t rely on gut feel about “shifty” behavior.
- Stronger signals come from answer content and the environment, such as extra browser activity or answers that match typical AI output.
- Verify with follow-up questions that demand personal detail, and give the candidate a chance to respond before any decision.
- Notice, consent and bias rules apply to tools that analyze candidates, so check your jurisdiction first.
Table of Contents
- What counts as cheating in a job interview
- Signals that suggest live help or AI use, and how reliable they are
- How detection tools work
- Follow-up questions that verify an answer
- Preventing cheating without making the process hostile
- False positives, bias and legal rules in the US
- Our perspective
- Run structured interviews with Resyme
- FAQ
- Sources
What counts as cheating in a job interview
Before you detect anything, decide what you are detecting. Candidates and interviewers often disagree on this, especially around AI tools.
A useful working definition comes from a company that published its rules for applicants. Zscaler’s candidate AI guidelines prohibit misrepresentation, impersonation, plagiarism and unapproved outside assistance during a live assessment, coding challenge or interview. That covers help from other people, search engines, notes and AI tools, unless the company explicitly authorizes it in writing.
Four behaviors are worth naming in your own policy:
- Impersonation: someone else sits the interview, or a second person is present off camera.
- Live assistance: a coach, friend or AI tool feeds answers in real time.
- Scripted or generated answers: responses written by a language model and read aloud.
- Misrepresentation: claims about experience, titles or skills that the candidate cannot back up.
Some employers allow AI for preparation or for specific tasks. That is a legitimate choice. What matters is that the rule is stated, so a flag means a rule was broken and not that you guessed at intent.
Signals that suggest live help or AI use, and how reliable they are
The honest starting point: people are bad at this. Bill and Melchers cite earlier work reporting interviewer detection rates for faking tactics of about 11.8% to 18.5%. A separate study of asynchronous versus synchronous video interviews found that human interviewers had difficulty detecting deceptive impression management in either format, with the exception of extensive faking in recorded interviews.
That should change how you read a hunch. A candidate who looks away, pauses or seems nervous is behaving the way many honest candidates behave.
Why content beats body language
The research on ChatGPT in recorded interviews is useful here. Canagasuriam and Lukacik (2024) found that candidates using ChatGPT earned much higher overall and content ratings than those who did not. That is the risk: AI-assisted answers look good and can distort scores. The same study found evaluators rated ChatGPT-condition candidates as less honest, which suggests some patterns of AI answers can be noticed.
So the content patterns worth watching are:
- Polished, generic answers that fit any company and any role.
- Textbook structure with no specific project, name, number or mistake.
- Delay after the question, then a fluent, evenly paced answer, as if read.
- Inability to go one level deeper when you ask “what exactly did you do?”
- Answers that change character between questions: sophisticated on the prepared topic, thin on the follow-up.
Treat each of these as a reason to ask another question, never as proof.
Environmental signals
The BPS-style advice, echoed in proctoring playbooks, is to look at evidence about the environment rather than judge demeanor. In practice that means a second voice audible, typing during a verbal answer, a gaze that tracks a second screen consistently, or a change of person on camera between sessions. Note what you observed in neutral language. “I heard a second voice at 14:10” is a record. “They seemed fishy” is not.
How detection tools work
Tools in this category generally combine a few approaches. It helps to understand them so you can judge any product, including ours.
Answer analysis. The tool compares what the candidate said with the output popular AI tools would typically give for the same question, and looks for language patterns that suggest reading from a script. The Canagasuriam and Lukacik paper points to this kind of comparison as a way to strengthen detection.
Environment monitoring. The same authors suggest online proctoring, where legal, to detect whether interviewees open other browsers or applications during a recorded interview.
Identity and presence checks. A 2021 IEEE paper on a cheating detection pipeline describes a staged approach for online interviews: face detection, face recognition, object detection, face tracking and result analysis. These are the building blocks behind many proctoring products.
Voice and text signals. Differences between spoken and typed answers, and patterns that point to dictation or reading aloud, can be flagged for human review.
Whatever the method, use the output as a prompt for a human decision. A flag that triggers automatic rejection is the point where false positives start costing you good candidates.
Follow-up questions that verify an answer
The most dependable verification is a conversation. A candidate who did the work can talk about it from several angles. A candidate reciting a generated answer usually cannot. These follow-ups work in live and recorded formats, and they fit well with a behavioral interview structure.
Probe for lived detail
- “What was the exact situation on the day you started that project?”
- “Who disagreed with you, and what did they say?”
- “What would you do differently now, and what did it cost you the first time?”
- “Which number told you it was working?”
Rehearsed or generated stories tend to be smooth and free of friction. Real ones contain constraints, names, and mistakes.
Change the angle
Ask the same competency from a different direction a few minutes later. If someone described a conflict with a colleague, ask later about a time they disagreed with a manager. Consistent real experience carries across both. For technical roles, ask the candidate to explain why they chose an approach, then change one requirement and ask how the solution would change. Ask them to walk through their reasoning aloud, not just present a result.
Use a personalized prompt
Reference something specific from the candidate’s resume or LinkedIn profile and ask for detail that is not written there: “Your profile mentions the migration at your last role. What was the hardest decision in it?” Generic AI answers cannot draw on the unwritten parts of someone’s history.
When you suspect unauthorized help
A public example of a fair process is the State of Connecticut’s General Letter 195, which covers executive-branch vacancies. It prohibits unauthorized AI-generated answers and real-time assistance. If a hiring manager suspects it, the guidance is to pause, describe the observed behavior, repeat the instructions and let the candidate respond. That order protects honest candidates and gives you a defensible record.
A short script helps: “I want to check that I’m assessing your own thinking. I noticed [observation]. Can you tell me how you’re approaching this question?” Then listen. Some candidates will simply say they were reading their notes, which you can handle under your stated rules.
Preventing cheating without making the process hostile
Detection catches some cases. Design reduces the rest.
Explain the rules and the reason. British Psychological Society guidance on online assessments, published in 2006, advises explaining the mutual benefit of accurate assessment and telling candidates they may be reassessed under supervision later, which lowers the payoff of deception. The age of that document shows how long this problem has existed.
Don’t count on warnings alone. Bill and Melchers tested several countermeasures, including warnings and certain algorithms focused on nonverbal and paraverbal cues, and found they did not reduce faking intentions or faking in selection interviews. Warnings are still worth giving for clarity, but pair them with better question design.
Ask questions that resist outsourcing.
- Anchor questions to the specific job: realistic scenarios drawn from the role, not generic prompts that any chatbot answers well.
- Require behavioral evidence: situation, action and result, with detail.
- Add a follow-up step on every key answer.
- Keep recorded-interview time limits reasonable, so legitimate candidates are not pushed into reading from notes.
Keep it accessible. Candidates with slow connections, shared living spaces, speech differences or non-native English can all trip naive detectors. Offer alternatives such as a rescheduled slot, and review flags with a human before you act on them.
Tools built for structured interviewing can help with the question-design side. Our guides on turning job descriptions into interview questions and analyzing soft skills with traceable evidence cover how to keep every score tied to something the candidate actually said.
False positives, bias and legal rules in the US
Any tool that analyzes candidates carries legal duties. This is not legal advice, so involve your employment counsel before you deploy monitoring.
A Loeb & Loeb summary from 2022 highlights several points:
- Anti-discrimination law can apply to “neutral” AI tools that have a discriminatory disparate impact.
- New York City has required, since 1 January 2023, bias audits and advance notice before certain automated employment decision tools are used.
- Illinois’s Artificial Intelligence Video Interview Act took effect on 1 January 2020 and was amended effective 1 January 2022. It requires notice that AI may analyze video interviews, an explanation of how it works and what characteristics it uses, and applicant consent. Employers that rely solely on AI to pick candidates for in-person interviews must collect and report race and ethnicity data.
The practical response is simple: tell candidates what is analyzed, get consent where required, keep a human in the decision, and review outcomes by group. For a wider look at the privacy side of AI interviews, see this piece on AI interview data and privacy. If you hire in Europe, our guide to GDPR recruiting compliance and DPIAs covers the equivalent obligations.
One more caution on false positives. Detectors that read writing style can mislabel honest candidates, particularly non-native speakers. A flag should start a verification conversation, never end a candidacy.
Our perspective
Most advice on interview cheating treats it as a surveillance problem. We see it as a question-quality problem first.
If an interview asks generic questions, a generic answer, human or machine, will score well. So we build interviews around the role: questions tailored with domain knowledge, anchored in behavior, and followed up so that depth is tested, not just fluency. Our platform is designed to validate honesty and integrity, with voice and text detection, and to produce reports that let you compare candidates on the same footing.
We also believe the candidate experience matters. An accessible, in-depth conversation gets more honest answers than an interrogation, and it gives recruiters evidence they can actually explain to a hiring manager.
Detection should support the recruiter’s judgment and not replace it. That is how we design it.
Raul
Run structured interviews with Resyme
If you screen high volumes or hire for specialized roles, manual verification does not scale. Resyme runs automated, behaviorally anchored interviews tailored to the role, validates candidate honesty, and delivers comparable reports, using the candidate’s resume and LinkedIn profile as input.

Pricing is published on our site: pay-as-you-go interviews at €4.50 to €7.99 each, a Company plan at €599 per month and an Agency plan at €4,500 per month. There is also a Free plan and an Enterprise plan, which have no published price. Small teams and startups without an ATS can begin with single interviews and add a plan as volume grows.
FAQ
Can candidates use ChatGPT during an interview?
Only if the employer allows it. Many employers prohibit unauthorized AI answers and real-time assistance, as Connecticut’s General Letter 195 does for its executive-branch hiring. State your rule in the invitation so candidates know where the line is.
How do you tell if someone is cheating on a video interview?
Don’t rely on how they look. Watch for generic answers, an inability to give specific detail, a second voice or signs of reading, then ask follow-up questions that require personal specifics. Document what you observed in neutral terms before making a decision.
What should I do if I suspect cheating?
Pause, describe what you saw, restate the rules and let the candidate respond. Then verify with a different-angle question or a short supervised follow-up. Decide only after that conversation, with a human review of any tool flag.
Is it legal to use AI to monitor video interviews?
It depends on where you hire. New York City and Illinois, for example, have notice, audit or consent requirements for certain AI tools in hiring. Check with employment counsel in each jurisdiction before you deploy monitoring.
Sources
- Ethical AI Guidelines for Zscaler Job Applicants
- Comparing job applicant deception in asynchronous vs synchronous video interviews
- Thou Shalt not Lie! Exploring and testing countermeasures
- ChatGPT, can you take my job interview?
- Cheating Detection Pipeline for Online Interviews
- General Letter No 195 Candidate Use of AI
- Using online assessment tools for recruitment
- A Privacy and Employment Law Primer