Proxy interviews: how to spot a stand-in before you hire

On 30 October 2026, while researching this post, I pulled Google’s results for proxy interview in the US and in India. Google’s AI answer at the top defined it plainly as job interview fraud. The results under it told a different story. In the US, two of the top results were social media posts from people looking for someone to sit their interviews, or selling the service. In India, the top results included offers of “full proxy interview support” and a page of job listings for proxy interview support work. The fraud is not hidden. It is advertised next to the definition.

If you hire for remote or hybrid roles, especially in IT, you will meet it. This guide is for the recruiter who has to spot it: how common it is, the four forms it takes, the signs in a remote round, what screening technology can and cannot catch, and the process checks that actually stop a stand-in. The closely related problem of candidates getting answers from ChatGPT mid-interview is covered in what to do about ChatGPT during an AI interview.

What is a proxy interview?

A proxy interview is interview fraud in which someone other than the real applicant takes the interview, or feeds the applicant answers during it. The person assessed is not the person who will do the job. It thrives in remote rounds, where the interviewer sees one face on one screen and cannot see who else is in the room.

The harm is not just one bad hire. A proxy hire into a role with system access can be a security problem, which is why the FBI’s Internet Crime Complaint Center warned employers as early as June 2022 about deepfakes and stolen personal information used to apply for remote work positions, mostly in IT, programming, database and software jobs.

How common are proxy interviews?

Nobody can measure it exactly, because a successful proxy is by definition not caught. The best public number is self-reported. Gartner surveyed 3,000 job candidates in 2025, and 6 percent said they had participated in interview fraud, either posing as someone else or having someone else pretend to be them. The same Gartner work predicts that by 2028, one in four candidate profiles worldwide could be fake. That is a forecast, not a measurement, but it shows the direction.

Our own data measures a neighbouring problem: help being fed to a candidate during a live voice interview. Across 2,582 assessed interviews in our hiring data from India, 1.9 percent showed elevated or high signals of outside assistance, and 2.6 percent of candidates switched tabs at all. That is lower than the panic suggests. It is also not a count of proxies, and I want to be careful not to present it as one.

The four forms a proxy interview takes

1. The full stand-in. A more skilled person attends the whole interview as the candidate. The real candidate joins on day one. This is the hardest to catch in a single round and the easiest to catch across rounds.
2. The off-camera feeder. The real candidate is on screen; someone else listens and types or whispers answers. It shows up as timing: long pauses, then polished answers.
3. The lip-sync or deepfake. The face on camera is the applicant’s, the voice belongs to someone else, or the video is altered. The FBI’s warning described exactly this: lip movement on camera that did not fully match the audio, and coughs or sneezes that did not line up with the picture.
4. The round switch. The real candidate does the HR round; a proxy does the technical round. Each interviewer sees a consistent person, and nobody compares them.

Signs of a proxy interview in a remote round

None of these proves fraud. Each is a reason to ask one more question.

• Audio and video out of step. Lips moving slightly off the words, or sounds that do not match what you see.
• Long pause, then a perfect answer. Real people thinking out loud hesitate, restart and correct themselves. An answer that arrives after fifteen seconds of silence, fully formed, is a classic sign of someone being fed.
• The camera fails at the important moment. It works for introductions and freezes for the technical question.
• They cannot discuss their own resume. Ask about a specific project, a manager’s name, a tool they listed. A stand-in knows the subject but not the candidate’s life.
• The voice changes between rounds. Different accent, pace or pitch between the screening call and the panel.
• Small surprises go badly. Ask them to share their screen, turn the camera to the side, or answer a question that changes the scenario. Rehearsed setups struggle with the unexpected.
• Pressure to keep things remote. Every proposal of a quick in-person or live video step is met with a reason it cannot happen.

What a screening tool can and cannot catch

I would rather be precise here than reassuring.

HireQwik’s voice interviews compute seven signals of possible outside help after every call: tab switches, long answer latency, a “slow but fluent” pattern (a long wait followed by a hesitation-free answer), reading from a script, essay-style structure, zero self-correction in speech, and camera off. They are banded low, elevated or high, and they are flags only: they never change a score, a verdict or an automatic decision. The panel tells the recruiter they are “signals, not proof” and to verify in the recording. We calibrated them on real interviews before switching them on; a naive “eight seconds is suspicious” rule would have flagged about 30 percent of honest candidates. The interview recording is kept, and a job can be set to ask each candidate questions built from their own resume, which are harder for a stand-in than generic ones.

What it does not do: it does not verify identity. There is no face matching and no voiceprint, and anyone adding those should know that biometric data carries its own legal risk (we wrote about voiceprint laws and AI interviews). So a skilled stand-in who takes the entire call can pass our screen, just as they can pass a human phone screen. Any vendor who tells you their screening alone stops proxies should be asked exactly how they verify who is speaking.

Five process checks that stop a stand-in

Proxy fraud is beaten by process across rounds, not by one clever detector.

1. Check identity at a live round before the offer. Government ID held up to the camera at the start of a live video round, matched against the application. Simple, and it removes the casual cases.
2. Ask a continuity question. In the final round, refer to something specific the candidate said in round one: “In your first interview you mentioned the payment bug in your last project. What was the fix?” A stand-in in either round cannot answer it.
3. Compare voices across rounds. If the first round was recorded, have the final interviewer listen to two minutes of it beforehand. A different voice is obvious to a human ear.
4. Make one step live and unscripted. A short in-person meeting or a live screen-share task near the end of the process. Most proxy setups depend on everything staying predictable and remote.
5. Check again on joining day. The person who joins should match the ID, the photo and the voice from the interviews. Background verification also catches stolen identities: the FBI found that some applicants’ personal information belonged to another person.

What not to do

Do not reject automatically on a signal. A long pause can be a nervous fresher, a slow network or someone translating in their head. A camera that drops can be a cheap phone. Ask for a short re-check instead and decide on what you see then. Do not accuse in the interview; say you would like a quick second conversation, which honest candidates accept easily. And write down what you saw and when, so the decision can be explained later.

The most useful shift is to stop treating any single round as proof of who someone is. A screen, ours included, tells you how a person answered. The process around it tells you whether the same person will show up on day one. If you want to see how the assistance signals and the recording sit next to a verdict, book a short demo and we will walk you through a real scorecard with sample data.