In brief
The signs interviewers list describe the candidate’s behaviour, not the program’s operation. No technical traces are visible from the interviewer’s side of the call: the video stream carries only what the candidate chooses to show.
That is why companies’ countermeasures aim at making the gap between the pause and the answer more visible, rather than at finding software.
The pause and what fills it
A pause before answering means nothing on its own: everyone thinks. Suspicion comes from what fills it and what is said after it.
Interviewers name the same set of signs: eyes drifting to the side or down, keyboard clatter in the silence, cursor movement where the candidate is not showing anything. Each sign on its own has an everyday explanation — together they add up to a picture.
Something else matters more. The pause repeats on every question and is the same length each time, including questions an experienced person would answer instantly. An even rhythm gives more away than the length of any single pause.
The speech changes register
A specialist’s spoken language is uneven: it has hedges, backtracking, “well, more like this.” Text generated by a model is even and structured — it shows when a person suddenly starts speaking in written-language constructions.
The gap widens when the register shifts within the conversation. The candidate talked freely about their previous job in ordinary words, then switched to textbook language on a technical question.
An answer with no experience in it
A model answers the question; it does not tell a story. The answer contains a correct definition and not a single detail of the kind only someone who has done the work by hand would have.
This is checked in one move: instead of “what is it,” the interviewer asks how it went on the last project. Where there is experience, details surface — a library version, a disagreement with a colleague over the approach, an outage on a Friday. Where the answer was read, the conversation runs into generalities.
A second sign of the same kind is the lack of a position. A person with experience usually has an opinion and can defend it, while a read-out answer stays neutral, because the model avoids being categorical.
What companies change in their rounds
Employers hardly try to detect the program. They change the format of the round so that a prompt no longer gives an advantage.
| What they do | Why |
|---|---|
| The task as an image instead of text | The problem statement cannot be copied into a chat: it has to be recognized, and recognition makes mistakes |
| A blurred or crossed-out fragment in the code | A model reconstructs such a fragment inaccurately and answers beside the task, while a person sees it correctly |
| Watching the solution process rather than the result | The path is what gets assessed: drafts, dead ends, backtracking. A finished correct answer with no path raises questions |
| Follow-up questions that dig deeper | A prompt answers the question asked. Three follow-ups in a row require an understanding the reader does not have |
| Talking about experience instead of theory | Details of past projects cannot be pulled out of a model: they never happened |
Detectors do not exist
It is worth noting separately what the market does not have. There is no program that reliably determines from a video stream that a candidate is reading a prompt, and promises of that kind deserve the same scepticism as promises of complete invisibility from the other side.
A person makes the decision, based on the overall impression. This explains why the story has no technical resolution: the candidate is not exposed, they are simply not invited to the next stage.
Frequently asked questions
Can an interviewer technically detect a prompter program?
Not from their side of the call. Zoom or Meet only shows the video stream and the shared screen, and a window excluded from capture does not get into the stream. Detection happens through the candidate’s behaviour, not through technical traces.
What gives it away most often?
The combination of the pause and the quality of the answer. A pause on its own is normal — the person is thinking. Suspicion arises from a pause followed by a structured, bookish answer without a single hedge, on every question in a row.
Does turning off the camera help?
Quite the opposite. Refusing to turn on video without a clear reason is itself on the list of signs interviewers name. That is exactly why technical rounds usually ask for the camera to be on.
Why do tasks include blurred or crossed-out code?
Such a fragment breaks recognition: a model reading the screenshot reconstructs the text inaccurately and gives an answer that does not match what is on the screen. A person looking at the task with their own eyes will not notice the discrepancy, while someone reading a prompt will answer beside the point.
How common is the use of prompts?
There is no verifiable data for the Russian market. Estimates such as “one in four candidates in the technical round” do circulate, but their authors do not disclose the sample or the counting method, so such figures cannot be relied on.
How the prompting technology itself works is covered in the article on the stealth assistant, and the formats of technical rounds in the live coding breakdown.