Skip to content

Catch a rehearsed answer with the second question

In short

A live call shows pauses, gaze and polished wording, and each has honest causes, so no single sign proves a candidate is reading ChatGPT answers. Tracy AI Interviewer asks every applicant a follow-up in their own field, waits for the answer and shows a reviewer integrity signals with their times, beside a recording and transcript. A signal is a reason to look, never proof, and a person decides.

Candidates using ChatGPT in interviews: what a live call can show

A live call does show things. A steady delay before answers, wording that sounds like documentation and a gaze that stays on one spot are the signs people mention for candidates using ChatGPT in interviews. Each has honest causes, so no single one proves anything; look for a few together. A follow-up about the answer the candidate just gave is a check you can run. The cheating detection guide covers what a signal can and cannot show.

Watching a call is a weak defence on its own. In an interviewing.io experiment of 32 mock coding interviews with senior candidates (the post is undated; read 14 Nov 2026), professional interviewers ran interviews with candidates who had been told to use ChatGPT. Not one interviewer raised cheating as a concern, and 81% of the candidates had no worry about being caught. The setup was audio only, so a video call gives you more to see. Even so, a feeling is not a check.

Hirers do suspect it. In Checkr's September 2026 survey of 3,000 US managers, 67% had suspected a candidate of using AI to misrepresent themselves. Checkr sells background checks, and these are managers' suspicions, not measured rates.

What a call cannot do is ask the second question of everyone. A signal tells you which minute to open; only a follow-up, asked of every candidate the same way, shows whether the person did the work.

Which signs suggest a candidate is reading ChatGPT answers?

These are signs people commonly mention, and the right-hand column shows why none is a test.

SignWhat you seePossible honest explanation
Delay before answersA similar pause before every answer, whatever the questionThinking time, a slow connection, a second language
GazeEyes fixed on one spot off cameraNotes, a habit of looking away to think
Answer styleTidy structure, formal links between points, no concrete detailA prepared answer, a careful speaker
Echoed questionThe question repeated back before each answerNerves, buying time
Thin second answerA fluent first answer, then vague detailNerves, or the topic is not theirs
Window changesTab changes or leaving full screenA note, a notification

Six signs people commonly mention, with an honest explanation for each. Read each as a prompt to ask a question.

Leave accent, name, connection quality and camera angle out of it, because they say nothing about whether someone used a tool.

Why does a follow-up question work better than watching?

A pasted or memorised answer is built for the question as it was asked. The next question depends on details inside that answer, and the detail is harder to supply from a script. A candidate who did the work can go one level deeper. A candidate reading a screen may have to pause, ask again and read again, which a reviewer can see in the words as well as the timing.

In the written sample below, the first answer is smooth and says almost nothing. The follow-up asks for what the candidate saw, which only someone who did the work can describe. None of the 32 interviewers in the interviewing.io experiment raised cheating, so a call watched in real time left them with a feeling and no evidence. A follow-up chain leaves a transcript line to point at.

Sample transcript: a general answer about faster queries, then Tracy asks what the slowest query showed.
Tracy follows a general answer with a question about what the candidate actually saw, with times on each turn. Illustration with sample data

A smooth answer is not evidence; the question underneath it is.

What follow-up questions expose a rehearsed answer?

Use the same few moves on every candidate, so the comparison is fair.

  • Ask for the detail behind the claim: "What did you see before you changed it?"
  • Change one constraint: "What if the data were ten times larger?"
  • Ask for the road not taken: "What else did you consider, and why not that?"
  • Ask for the first mistake: "What did you try first that did not work?"
  • Pick up their own word: "You said the cache went stale. What went stale, and when?"

Listen for specifics, for an answer that moves when the constraint moves, and for a plain "I don't know". For engineers, the same chain works on code; see coding interview follow-up questions.

What Tracy adds when candidates use ChatGPT

  1. Every applicant starts on click, at any hour, with no scheduling and no recruiter on the call, so the same questions reach the whole pool.
  2. Tracy asks the next question in the applicant's own field and waits for the answer, so a smooth first answer meets a question about the detail behind it.
  3. Every result keeps a recording, a timestamped transcript and what the person said to each question, with integrity signals and their times. A signal is a reason to look, never proof.

What does Tracy AI Interviewer show you?

Tracy AI Interviewer runs those follow-ups for every applicant. When the CV check is on, she reads the CV and asks about it. She then follows up in the person's own field and waits for the end of the thought before she asks for the missing detail. Each applicant starts on click, at any hour, so the same questions reach the whole pool, including people your team had no time to call. For engineers she can set a coding task and ask about approach and trade-offs, as AI technical interviews explains.

On video interviews she also captures tab changes, abnormal eye tracking (gaze), live face detection, delay before answers, leaving fullscreen and external screens, unless you turn them off. A reviewer sees each one in the report with its times, beside a recording of both sides and a timestamped transcript. The candidate is told on the join screen that the conversation is recorded and transcribed. Tracy gives no detection rate and decides nothing.

Illustration of a signals panel with sample data: gaze seen twice, one fullscreen exit, other signals not shown.
Tracy lists each signal with its times, so a reviewer opens the transcript at that minute and reads the answer. Illustration with sample data

Signals arrive with their times, so the reviewer knows which minute to open.

What should you do when you suspect a candidate used ChatGPT?

Start with your own rule. If candidates may use AI to prepare or to draft, say so, and say where it stops. A candidate who reads answers from a tool during a live interview has broken a rule only if you wrote it down.

Then run the same routine each time. Read the transcript at the minute of the signal. Compare the answer with the CV, one level deeper. Ask the person to explain it in a short live call, with the questions you would ask anyone. Record the decision and the evidence so a second person can review it. A person decides, every time, and one signal is never enough to reject someone.

When the worry is wider than ChatGPT, such as a stolen identity or a proxy, candidate fraud detection covers the layers. For definitions, see interview fraud and deepfake candidates.

Four steps: a signal appears, open the minute, ask one level deeper, a person decides.
Tracy lists each signal with its time, so the reviewer opens that minute, asks one level deeper and decides. Illustration, no data.

A signal sends the reviewer to one minute and one question.

Which path fits you?

Common questions

Can you tell if a candidate is using ChatGPT during an interview?
Not from one sign. Delays, a fixed gaze and polished wording all have honest causes. A follow-up that asks for the detail behind an answer is a check you can run, and a reviewer should weigh any signal with the recording and transcript.
Is using ChatGPT in an interview cheating?
It depends on your rule. If you tell candidates that no AI help is allowed, reading answers from a tool breaks it. If you never said, state the rule in the invitation next time.
What should I do if I think a candidate used ChatGPT?
Read the transcript at that moment, compare the answer with the CV one level deeper, and ask the person to explain it live. Record the decision and the evidence. A person decides, never the tool.
Can software detect ChatGPT use in an interview?
No tool should be read as a verdict. Tracy asks follow-ups on the work and shows a reviewer signals such as gaze and delay before answers with their times, beside the recording. She gives no detection rate, and a person decides.
Why do candidates use ChatGPT in interviews?
Some want help with a hard question and some use it to polish answers they would give anyway. Motive is a guess, so judge the evidence in the answer.

Read one interview's evidence yourself

Create one interview for your role and read the follow-ups, signals and transcript yourself. The first 3 results are free.

Try it free (opens in a new tab) Book a demo (opens in a new tab)

Your first 3 results are free. No card needed.

Sources