Often not by watching alone. In an interviewing.io experiment with 32 analysed interviews, no interviewer raised a concern about cheating although every candidate was secretly using ChatGPT. Employers rely instead on question design, follow up questions, screen sharing and in person rounds. Hidden use is a minority habit: about one in five people admit to it in self reported surveys.
What did the experiment find?
interviewing.io ran 37 interviews in which candidates were told to use ChatGPT without the interviewer knowing, and analysed 32 of them. Candidates passed 73% of interviews built on verbatim LeetCode questions and 67% on modified ones, against 53% for a control group of ordinary mock interviews. On custom questions the pass rate fell to 25%.
No interviewer mentioned concerns about cheating. The authors list the limits themselves: audio only, groups of 9 to 12, and candidates chosen for their willingness to cheat. The result that holds up best is the gap between familiar and custom questions.
That gap explains much of what interviewers changed. A question a model has seen many times can be answered from memory, word for word; a question written for the interview cannot, and the follow up questions about why a step works expose an answer the candidate did not produce.
How common is hidden AI use?
Self reports put it at about one in five. In a Blind poll run from 10 to 20 April 2025, 20% of the 2,510 verified professionals who answered said they had secretly used AI during an interview. Resume Genius's 2026 survey of 1,000 active US job seekers, reported by Newsweek, found 22% using AI during real time interviews.
Vendor figures run higher and should be read as such. CodeSignal, which sells assessments with cheating detection, says flagged cheating and fraud attempts on proctored assessments rose from 16% in 2024 to 35% in 2025; it states that these are attempts it detected and flagged, not successful cheating.
Interviewers suspect more than they catch. In interviewing.io's 2025 survey, 81% of interviewers at large tech companies said they suspected candidates of using AI to cheat, and about a third said they had caught someone.
What do employers do instead of guessing?
They change the conditions. Meta, per interviewing.io's survey, asks candidates to share their whole screen and switch off background filters, and has interviewers flag suspected cheating. Amazon's reported guidance tells recruiters to watch for typing during questions and answers that sound read. Google, Cisco and McKinsey brought back in person interviews for part of their hiring, and Deloitte UK did the same for graduate recruitment.
Question design does the rest. Custom problems and follow up questions about why you chose something expose borrowed answers far better than watching a face. One startup, The Pragmatic Engineer reported, hid an instruction in its take home brief that only an AI would follow; all four completed submissions contained it.
Does detection software work?
There is little independent evidence either way. In interviewing.io's survey only 11% of big tech interviewers said their company uses cheating detection software, and nearly all of them were at Meta. The research behind this site found no peer reviewed study of how often AI cheating happens in hiring, and the figures that exist come from surveys and from vendors that sell detection.
What is the safer approach for a candidate?
Ask what is allowed and follow it. Where AI is banned, the downside is disqualification at companies such as Amazon. Where it is allowed, use it openly; interviewers in those rounds want to see how you work with it, and hiding it wastes the signal they are looking for.
The company policies page lists which employers allow AI in which rounds, and our practice rounds let you rehearse the open version.