Live Captions in an Interview: Ask Before the Round

Which Tool Are You Asking About?

Two different products get called the same thing, and only one of them writes your answers. A live caption track turns the interviewer's speech into text you read while they talk. An interview copilot listens to the question, drafts a response, and puts it on your screen for you to say. The published lists quoted below were written about the second one.

The UK Civil Service publishes a candidate's guide that draws the line in public. Refining and clarifying your own ideas, researching the organisation, and checking spelling, grammar and clarity sit on its acceptable list. Inflating or inventing skills and experiences, pasting generic AI responses into an application, and using AI to complete personality and ability assessments sit on the unacceptable one ¹.

None of those items is about hearing the question.

That guide binds one employer and nobody else, it carries no date on its page, and it names no method for noticing any of this ¹. Read it as a worked example of how a large employer thinks about the question, rather than as a standard being applied to you. It turns up twice here: once for what it permits, and once for what it rules out.

The line it draws is about whether the ideas and the claims are yours.

Read the Wording You Were Given

The rule in your invitation decides this, and it is usually one or two sentences long. The same Civil Service guidance draws one absolute line: do not use AI during any live assessment at any stage of the selection process ². The same page says applications may be rejected where AI is used inappropriately ². Written that broadly, the line reaches a caption tool on its face.

That is a stated power rather than a measured outcome. The page gives no rate, no examples and no appeal route, and it says nothing about how anyone would decide a tool had been used inappropriately ².

A question before the round is cheaper than an argument after it.

So send one line through whatever channel you already have with the recruiter: "I'd like to run live captions during the call so I can follow the questions. Is that allowed?" A written yes turns a judgment somebody else would make later into a permission you already hold.

The same message can carry three more questions:

• what tools are allowed during the live round, and whether the answer changes for a recorded round
• whether reading a transcript of the interviewer counts as one of them
• what to do if the captions drop in the middle of an answer

Get the answer in the thread, in writing, before the round starts.

When Hearing Is the Barrier, Ask for an Accommodation

If a hearing disability is the reason you want captions, this stops being a question about tools. In the United States the ADA makes it discrimination for an employer with 15 or more employees to fail to make reasonable accommodations for a qualified applicant with a known disability, unless the employer can show undue hardship ³. That duty sits on the employer, and your request is what starts it.

The threshold for making the request is low. Under the EEOC's 2022 technical assistance on algorithmic assessments, telling an employer that a medical condition may make a test difficult to take is already a request for reasonable accommodation ⁴. You do not have to mention the ADA or use the phrase ⁴. That document was archived and removed from the agency's site in January 2025, so pair it with the statute rather than resting on it.

Saying the reason is what makes it a request.

On one widely used platform the route is a button, though it is narrower than it looks. HireVue's 2025 statement reproduces a Request Accommodations prompt shown before an AI-scored interview, offered to candidates who need extra time for a qualified disability ⁵. Accommodation requests go to the relevant department of the employer managing the hiring process, and the employer decides whether an exemption is granted ⁵.

Taking the relaxed time limits also removes the game-based section, so that is a different assessment and not merely a longer one ⁵. The document calls that wording a sample its customers can alter ⁵.

The button is a request, and the employer answers it.

So ask what changes as well as what is relaxed, and say plainly that captions are what you need rather than extra time.

Wording the request is its own small problem. How to ask for an accommodation on an AI interview covers the phrasing and where to send it. The employer-side page on whether ADA accommodations apply to an AI-based assessment is what a hiring team reads about a request like yours.

All of that is United States law, so check what applies where you are applying.

What If the Barrier Is Language, Not Disability?

Then no accommodation statute in this article stands behind the ask, and it becomes an ordinary request. Nothing stops a recruiter from saying yes. But the ask carries none of the legal weight the disability route carries, so make it early and in plain words rather than asserting a right you do not have.

What US anti-discrimination guidance does reach is the fluency bar itself. The EEOC's 2016 enforcement guidance on national origin discrimination says an English fluency requirement is generally permissible only where it is required for effective performance of the position ⁶. Applying one uniform standard across dissimilar jobs, or demanding more fluency than a job needs, may violate Title VII ⁶. That is guidance rather than law, and it was written about explicit fluency requirements rather than about a caption tool.

That is a frame for the conversation, not a card to play in it.

Translating the interviewer's questions into your language sits with the captions rather than against them. It helps you understand what was asked, and it composes nothing you say.

Translating your own answers runs the other way, and it is a different decision. It changes what the interviewer hears, because the sentences reaching them are a machine's rendering of what you said. What a hiring team is told to do after an AI-translated interview is worth reading first. The advice there is to judge the record for the decisions you made, and to treat a garbled sentence as the pipe's fault rather than yours.

Name it before the round rather than running it quietly.

Don't Trust a Caption Over the Question

A caption is a machine transcript, and a machine transcript can contain words nobody said. OpenAI's model card for Whisper says the models perform unevenly across accents and dialects, and may show a higher word error rate across speakers of different genders, races and ages ⁷. That is a vendor describing its own system, not an audit of it.

The failure can be larger than one wrong word. A 2024 study ran clinical research recordings, including speech from people with aphasia, through Whisper and found transcriptions containing whole invented sentences ⁸. Of those hallucinated transcriptions, 38% carried at least one harm the authors coded, such as a made-up name ⁸.

Keep that proportionate. On the same audio segments the authors found no comparable invented sentences from the other major services they tested, including RevAI, the supplier HireVue names for transcription ⁸. A re-run after later model updates found the problem in far fewer of the same segments ⁸. The Whisper runs date from 2023, and the recordings were clinical interviews rather than job interviews.

So do not answer the caption. Answer the question, and when the caption and your ear disagree, say so and ask for the sentence again.

Asking somebody to repeat a question is an ordinary thing to do.

None of these sources measured a live caption tool in a job interview. What they measure is the recognition step underneath, which is the same kind of step at least one AI interview vendor scores your answers through. The record on AI interviews, accents and disability covers that ground for the answers you give, rather than the questions you read.

See how it works >