Ask the AI to Repeat It, and Name the Part You Missed

Can You Ask a Live AI Interviewer to Repeat the Question?

Yes on the live conversational formats two vendors describe, because asking is part of that design. Alex, the AI recruiter formerly called Apriora, publishes candidate testimonials describing the AI summarizing an answer back and asking follow-ups drawn from it. The same testimonials describe candidates asking questions in return ¹. That page is marketing assembled from selected testimonials.

So read it as the intended design, not as evidence the design works.

HackerRank describes the same posture in a technical setting. Its March 2025 AI Day recap says the AI Interviewer it launched conducts first-round interviews, gives real-time hints when a candidate is stuck, and asks follow-up questions to judge the thought process. The post says the interviewer had conducted over 12,000 interviews and was available for initial screening ². That count is the vendor's own, and the passage does not say how many were live employer screens rather than practice runs.

A format built to ask follow-ups does not break when you ask one.

What none of this settles is the configuration in front of you. Both descriptions come from vendors writing about their own products, and what either one does depends on what the employer switched on. The instructions screen is where that shows up, so read it before the timer starts. Four kinds of AI interview, and what each one records sorts the formats apart, because the answer to this question is different for each of them.

A one-way recording and a live AI recruiter are different questions, and only one of them has anyone to ask.

Ask Out Loud Rather Than Waiting in Silence

Say it out loud, because silence is an input the software reads. Alex's employer settings documentation carries an example welcome message for candidates. It says that when the AI finishes speaking it is their turn, and that after 2 seconds of silence the response automatically ends ³. Employers can edit that text, so the exact number will not hold everywhere.

But the shape of it holds: a pause is a decision the system makes for you.

The same page documents an optional add-on the vendor calls Cheat Detection. It monitors for signals including background voices, excessive tab switching and unusual delays before answering, with the employer setting a threshold between 0 and 100 above which an interview is flagged for review ³. It is an add-on, so many interviews will not have it running at all, and the page publishes no accuracy or false-positive figures.

That is a reason to say you need a moment, rather than to sit through it.

So put the sound on your side of the exchange. "Could you repeat the second half of that" takes about two seconds to say and leaves a record of what happened. Thinking about it in silence takes the same two seconds and leaves nothing behind. Set the room up the same way, since a housemate talking in the next room and repeated tab switching are both on the list that page publishes ³.

A question is legible in a way that a long pause is not.

Which Part Do You Actually Want Repeated?

Ask for the part you actually missed, because a bare repeat gives you back the sentence you already did not follow. If a phrase was the problem, name the phrase. If you heard the question and cannot tell which of two readings is wanted, say both and ask which. If you lost the middle, say which part you caught.

• "Could you repeat that from where the client pushed back" names the place the thread was lost.
• "Are you asking how I picked the metric, or how I checked it afterwards" hands over a choice that can be answered in a word.
• "I'll take that as the migration project unless you meant the launch" states a reading and keeps the clock moving.

Restating your reading does a second job a bare repeat does not. On Alex's format the report that reaches the hiring team can include the questions and responses, the video recording and the transcript ³. So the sentence where you named your interpretation sits on the record beside the answer you gave. A person reading it later can see you answered a question you had actually understood.

A misunderstanding you named out loud is a different thing from one nobody caught.

Employer-side guidance pushes in a similar direction. The employer-side guidance on what follow-up questions expose about a candidate's understanding tells interviewers to ask which alternative you ruled out. Another rung asks what would have to be true for the answer to be wrong. A candidate who says out loud which reading they took has started that exchange instead of guessing at it.

That is close to the first rung of the same ladder, offered rather than pulled out of you.

The exchange matters beyond getting one question right. Across two studies in the International Journal of Selection and Assessment, AI-based interviewing was generally viewed as less procedurally and interactionally just than human-based interviewing ⁷. The effect ran through justice dimensions, two-way communication in particular ⁷. Those studies described AI interviewing as it stood in 2020, before live conversational interviewers, and the abstract was read on a citation index rather than the publisher's own page.

The two studies are no reason to stay quiet in the one format built to answer you.

What Do You Do When Nobody Is on the Call?

Spend the preparation time settling your reading, then say it in your first sentence. HireVue's 2025 explainability statement makes preparation time per question an employer setting, from 0 to 5 minutes or unlimited, sitting beside a yes-or-no switch for question retries ⁴. Unlimited retries is the vendor's recommended setting rather than a promise about any particular employer.

The same statement says how a candidate experiences the assessment is configurable by the company deploying it ⁴.

Then state the assumption and answer the question you named. "I'm reading this as the technical side rather than the client side" costs a few seconds at the top of an answer and makes everything after it legible. A reviewer who wanted the other reading can see exactly where the two of them parted. Without that sentence, an answer to the wrong question looks the same as an answer from someone who did not know the material.

An assumption on the record is cheaper than a guess nobody can see.

Saying it clearly matters more than it sounds, because HireVue's AI-scored interview runs on a machine transcript. HireVue names Rev.ai as the speech-to-text supplier behind that transcript. It publishes its own internal testing: English word error rates in the United States under 10% on average for Rev.ai, against 15% to 25% for the other systems it tested ⁵. Those are a vendor's own figures about a supplier it chose, with no sample size and no independent replication published.

On that platform, a transcription error is a scoring input.

Employers reading the other side of this get a version of the same warning. The employer-side guidance on judging communication after an AI-translated interview tells hiring teams that a garbled sentence belongs to the pipe, not to the candidate. The fix there is to mark it unreadable and ask again. That is what you want a reviewer doing with a line the transcript mangled.

When the barrier is the format rather than the question, ask for a change to the format. HireVue's sample instructions screen carries a Request Accommodations button for a qualified disability that relaxes the time limits, with the request routed to the employer, who decides ⁶. The employer-side guidance on whether ADA accommodations apply to an AI-based assessment says an accommodation should change the conditions, never the skill being measured.

On that platform, relaxed time limits also drop the game-based section, so it is a different assessment rather than a longer one ⁶.

See how it works >