Interview cheating in the age of AI: what actually stops it

Every remote interview now has a possible third participant. A candidate can run a model in another window that listens to your question and writes an answer in seconds. Some products are sold for exactly that purpose. Most hiring teams have met at least one candidate whose answers were fluent, fast and strangely empty.

The instinct is to fight software with software. That rarely works. This article explains why, and what does.

How common is it?

Common enough to plan for. In a 2025 survey for the insurer Hiscox, 29% of applicants said they had used AI tools to help with remote interviews, and 45% had used them for online tasks and assessments. Gartner found that 6% of candidates admitted to interview fraud, meaning someone else posing as them or them posing as someone else.

Large employers have reacted. In early 2025 CNBC reported that Google's chief executive told staff it was worth making "some fraction" of interviews in person, in a story about a product built to help engineers cheat in technical interviews.

Those numbers measure different things. Some of what is counted as "AI use" is preparation, not cheating. That is the first problem to solve.

Step one: decide what cheating means

Most teams never write this down, so candidates guess. Some will assume any AI is forbidden. Others will assume anything goes because the employer uses AI too.

Be explicit, in writing, before the interview:

• Allowed: using AI to prepare, research the company, practise answers, polish a CV.
• Allowed if you say so: using AI during a take-home task, if you want to see how candidates work with it. Many jobs now involve AI daily.
• Not allowed: receiving live answers during a conversation, having someone else answer, or submitting generated work as your own when told not to.

Clear rules do two things. Honest candidates stop worrying, and dishonest ones lose the excuse that nobody told them.

Why detection disappoints

Detection products promise to catch hidden windows, unusual eye movement, typing rhythms or "suspicious behaviour". In practice they run into three problems.

1. It is an arms race you lose. Cheating products update weekly. Your interview process does not.
2. False positives fall on the wrong people. Candidates who look away to think, read slowly, are neurodivergent, or are working in a second language can all trip behavioural flags. You end up penalising the honest and the different.
3. Some of it is unlawful. Systems that infer emotions from faces, voices or body movement in the workplace have been prohibited under Article 5 of the EU AI Act since February 2025, and the Commission's guidance treats candidates in hiring as within that workplace ban. "Nervousness" or "honesty" scoring falls straight into that category.

Some technical controls are reasonable, such as a locked-down coding environment for a timed test. But no control replaces good interview design.

What actually stops it: design, not surveillance

The goal is not to make cheating impossible. It is to make it useless. AI is good at producing a plausible first answer to a known question. It is weak at the things below, so build your interview around them.

Ask about their work, not the textbook

"Explain the difference between X and Y" has a perfect answer one prompt away. "Tell me about the last time X went wrong on your project, what you tried first, and why that did not work" does not. Real experience produces specific, slightly messy detail.

Go three questions deep

The first answer can be generated. The third follow-up, built on what the candidate just said, is much harder to fake in real time. Train interviewers to ask "why?", "what would you do differently?" and "what did your manager say?" before moving on.

Make the work sample a conversation

Instead of judging a submitted task on its own, discuss it live. Ask the candidate to change one thing, explain a choice, or walk through an edge case. People who did the work can do this easily. People who pasted it cannot.

Use structured scoring

Agree what a strong answer looks like before the interview, and score every candidate on the same criteria. Structure makes generic answers obvious, because they never hit the specific points you are listening for.

Keep one step live, and in person where it matters

For roles where the risk is high, one in-person or closely supervised step is a fair ask. Gartner's research found 62% of candidates said they were more likely to apply to a role that required an in-person interview. Many candidates want proof the process is real too.

Check references properly

Call references on numbers you find yourself, and ask about specific claims the candidate made in interview.

How MetaDay handles this

At MetaDay, our AI recruiter runs the early and middle stages of hiring, from outreach and follow-ups to screening calls, interviews and scheduling, inside boundaries the company approves. Two design choices matter here.

First, every candidate is told they are speaking with an AI, and can choose to speak with the team instead. We think transparency has to run both ways: if employers expect candidates to be honest about AI, employers should be honest about it too.

Second, the final interview and the decision stay with people. The run hands the hiring team a ranked top three, so their time goes into the step where depth matters most: three questions deep, live, about the candidate's own work. That is the part no cheating product can sit for you. Run this way, World Mobile hired a QA specialist in one working day.

FAQ

Can employers detect AI use during a video interview?

Sometimes, but not reliably, and detection that reads faces or emotions creates legal and fairness risks. Interviews built on follow-up questions and the candidate's own work are more dependable.

Is it cheating to use AI to prepare for an interview?

Usually not. Preparing and practising with AI is widely accepted. Getting live answers during the interview, or letting someone else answer, is cheating unless the employer explicitly allows it.

What is the best way to stop interview cheating?

Publish clear rules, ask about the candidate's own work, go several follow-up questions deep, discuss work samples live, and keep at least one step live with people.

The best defence against a borrowed answer is a question only the real candidate can answer.