ResourcesRequest a Demo
AI interview cheating

How To Catch Candidates Using AI In Interviews.

This is the plain-English version hiring teams actually search for. Candidates can use ChatGPT, Claude, Gemini, browser AI, copilots, local LLMs, hidden answer overlays, remote helpers, or paste-in answers during a live interview. The reliable way to catch it is to correlate what the computer is doing with what is happening in the interview.

SignsAI methodsProofFAQ
ChatGPT / Claude / GeminiAI
Copilot or local LLMAI
Hidden answer overlayHigh risk
Large paste after questionReview
Pause-then-burst typingReview
What to look for

Common signs a candidate is using AI during a live interview.

AI tool opens near a hard question

An AI app, browser tab, AI sidebar, or local LLM appears seconds after a difficult question or coding prompt.

Focus keeps leaving the meeting

The candidate switches away from the call or IDE at moments when they should be thinking or explaining.

Answers arrive in unnatural bursts

Long pauses followed by polished paragraphs, uniform typing, or large paste events can indicate generated answers.

Hidden overlay is present

Some tools show answers on the candidate's screen while hiding from normal screen sharing.

Copilot or coding assistant appears

Code suggestions may be allowed in some processes, but they should match the stated interview policy.

Second-screen or remote help

AI can be paired with another person, another monitor, or remote-control software, so review the whole environment.

How candidates use AI

Match the trick to the signal.

ChatGPT, Claude, Gemini, Perplexity

Browser title, app process, DNS activity, focus change, answer timing.

InterviewWatch signal

AI tool detection plus timing and focus correlation.

GitHub Copilot, Cursor, coding copilots

IDE extension activity, assistant process, paste or code-generation timing.

InterviewWatch signal

Tool presence and coding-round context in the final report.

Local LLMs

Local model runners, AI desktop apps, unusual compute activity, nearby paste bursts.

InterviewWatch signal

Known process detection and correlated answer insertion.

Hidden answer overlays

Answer window visible to candidate but excluded from screen capture.

InterviewWatch signal

Hidden overlay detection.

Remote helper using AI

Someone else controls the machine or sends answers to the candidate.

Evidence standard

Do not accuse someone because one thing looked strange.

A fair process needs timestamps and correlation. One paste can be normal. One tab switch can be harmless. But an AI browser tab, a focus change, a hidden overlay, and a large paste within seconds of the same question is a review-worthy pattern.

InterviewWatch approach

Detect machine-level integrity metadata without recording screen, audio, video, or keystroke content.

Set interview rulesTell candidates whether AI, docs, copilots, or web search are allowed.
Before
Collect metadataWatch tools, focus, windows, devices, clipboard size/source, and timing.
During
Review correlated signalsUse the report to decide whether follow-up is needed.
After
FAQ

AI use in interviews, answered.

How do you know if a candidate is using AI during an interview?
Look for correlated machine-level signals: AI app or browser activity, focus changes after hard questions, hidden overlay windows, large paste bursts, local LLM processes, or answer timing that does not match normal typing behavior.
Can candidates use ChatGPT without sharing their screen?
Yes. A candidate can use ChatGPT, a browser AI sidebar, a local LLM, a coding copilot, or a second-device workflow while the meeting window looks normal. That is why interview detection needs system-level metadata, not only screen share.
Should employers ban all AI in interviews?
Not always. Some interviews allow documentation or tools. Define the rule up front, disclose monitoring clearly, and review evidence in context instead of failing candidates for one isolated signal.

Need to catch AI use without turning interviews into surveillance?

InterviewWatch detects AI assistance, overlays, paste bursts, remote help, and timing patterns with consent-first metadata.

Request a DemoSee all cheating methods