Ask for the fragment you lost, not the whole sentence. "I caught 'the service handles' and then lost the number, could you give me that figure again?" repairs the gap and costs you nothing, where a bare "sorry, could you repeat that?" makes the interviewer simplify the question and strip out the detail you needed. Keep generic repeats to about two an hour, turn captions on before the call rather than hunting for the button during it, and know that captions and on-device transcription both still lose words to crosstalk, poor microphones and room noise.
Three scripts, and the budget on each
All three are politeness-safe. What separates them is how often you can spend one before the interviewer starts adapting to you, which is the part nobody warns you about.
The targeted repair
"I got 'the service handles about' and then lost the number. Could you give me that figure again?" You are asking for six words, not sixty. It proves you were following, it takes two seconds to answer, and it does not trigger the reflex where they rephrase the whole question in simpler words. Name the last thing you heard clearly, then the shape of what went missing: a number, a name, a verb.
The read-back
"Let me check I have this. You want the design to survive a whole region going down, and you care more about consistency than latency. Is that the priority order?" This is not a request, it is a summary, which is why it is effectively free. It also repairs sideways: if you heard "capacity" for "consistency", they correct it now rather than four minutes into the wrong answer. Convert clarifications into read-backs and you will barely touch the other two budgets.
The one-time line request, said early
"Before we get going, your audio is coming through a little compressed on my end. Would you mind speaking a touch slower? I would rather catch the detail than guess at it." Blame the connection. Not their accent, not your ears. Say it inside the first five minutes, while it is still housekeeping. The identical sentence at minute forty reads as an excuse for a round that has already gone badly.
| What you say | Safe budget in a 45 to 60 minute round | What overspending costs you |
|---|---|---|
| "Sorry, could you repeat that?" | Two | On the third, most interviewers switch to short sentences and plain vocabulary. It sounds kind, and it removes the constraint that made the question interesting. |
| Targeted repair ("I lost the number after...") | Three to four | Past that they slow down pre-emptively and start filling your pauses, so the silence you need for structuring an answer stops being available. |
| Read-back ("Let me check I have this...") | Effectively unlimited | Only costs you if the read-back comes back wrong twice running, which reads as not tracking rather than as a hearing problem. |
| One-time slow-down request | Once, in the first five minutes | A second reads as a complaint. After a third, some interviewers move detail into chat, which helps you read and hurts the conversational signal they are scoring. |
| "Would you mind dropping that in the chat?" | Twice, and only for a number, name or constraint | Beyond two it turns into dictation, and the round slows past the ground they planned to cover. |
When the audio degrades in the middle of a question
The instinct is to wait politely for the sentence to finish. That is the mistake: you lose the whole question, and what comes back is a paraphrase that has quietly dropped a clause.
- Interrupt at about two seconds. "Sorry, you cut out right after 'the service handles'. Could you pick it up from there?" Two seconds of broken audio is a repairable gap. Twenty seconds is a lost question.
- Check your Bluetooth profile, because on Windows this is usually the culprit. A Bluetooth headset shows up twice in Sound settings: a stereo output, often listed as "Headphones", and a hands-free one, often listed as "Headset". The moment a meeting app claims that headset microphone, Windows drops the link to the hands-free profile and everything you hear collapses to roughly telephone bandwidth. Consonants blur, which is exactly what wrecks comprehension across an unfamiliar accent. Set output to the stereo entry and input to the laptop's built-in or a wired mic, and the link stays in the high quality profile. Newer LE Audio hardware can avoid the downgrade, but do not assume yours does, test it beforehand.
- Turn your own camera off, and say why. It frees your upload and your machine. It does not repair audio that is already arriving broken. It is still worth doing because it is the only lever on your side, and narrating it signals that you are handling the problem calmly.
- Move numbers into text. "Would you mind pasting that constraint in the chat?" Figures, service names and function signatures survive text. Tone and follow-up questions do not, so spend this on specifics only.
- Two failed attempts, then propose the dial-in. "Shall I dial in on the number in the invite and keep video here?" Most Zoom and Teams invites carry one. One trade-off: once their voice arrives through a phone rather than your PC, transcription running on the computer stops hearing them.
Captions compared: Zoom, Teams, Meet and on-device transcription
Every option below is something you switch on for yourself, and none of them are announced to the room. The thing that announces itself is recording, or starting a saved transcript, which is a different feature entirely.
| Zoom automated captions | Teams live captions | Google Meet captions | On-device Whisper (OfferPilot AI) | |
|---|---|---|---|---|
| How you turn it on | Show Captions in the meeting controls | More actions, then Language and speech | The CC button in the call bar | On your own PC, before the call |
| Anyone told or shown | No | No | No | No. Nothing joins the call, nothing in the participant list |
| Can it be missing on the day | Yes. Account-level feature, so on a locked-down corporate account the button can be absent | Possible under an admin policy | Possible under an admin policy on managed Workspace accounts | No. It sits outside the meeting |
| Headphones or AirPods | Yes | Yes | Yes | Yes. It reads system playback, so what you hear is what it hears |
| Platform with no caption feature | No | No | No | Yes, as long as the audio plays through the PC |
| What you read | A running line that lags a second or two and rewrites displayed words as more context arrives | Punctuated sentences, plus a drafted answer grounded in your resume | ||
| Where audio is processed | By the platform's own service | On your machine | ||
| Biggest weakness | Shared by all four: accuracy falls on the accents least represented in the training data, which is exactly the situation you wanted help with | |||
Meet the revision behaviour before the interview, not during it. A word appears, then changes a beat later once the model has heard the rest of the phrase, and watching that while composing an answer is disorienting the first time. Third-party note-taker bots are a separate category: they join as a visible participant, so they are not a quiet option here.
What on-device transcription changes
The gap the platforms leave is the one-off tool. Plenty of interviews now run inside an assessment platform or a browser-based room with no caption button at all. OfferPilot AI captures system playback loopback, so headphones are fine, transcribes on-device with Whisper, and puts a speakable draft on screen in about a second, grounded in your resume and the company and role you set, and you can rehearse beforehand with a bank from the company Q&A generator. On this problem the transcript line is the useful half: you get the sentence in text while it is still fresh enough to ask a targeted repair about. Ctrl+Alt+H hides the overlay and holding P makes it see-through, which matters when you want to glance rather than read. The overlay is excluded from screen capture at the operating-system level, so it does not appear in Zoom, Teams, Meet or a recording. That covers screen sharing only. It does nothing about a webcam, a lockdown browser, or someone watching you in the room. The Interview Copilot page has the rest.
What none of this fixes
- Overlapping speakers. Two interviewers finishing each other's sentences, or someone talking over your answer, degrades every system in that table. Separating simultaneous voices is the hardest part of the problem and none of these solve it.
- Bad microphones and background noise. A laptop mic in a hard-surfaced room, an open-plan office, a fan behind them. Nothing downstream recovers information that was never captured. If their audio is genuinely poor, the fix is at their end or on a phone line, not in software at yours.
- Confident invention. Speech models, Whisper included, can emit fluent text during silence or noise that nobody said. If a line surprises you, treat it as unconfirmed and ask, rather than answering the sentence you read.
- Reading competes with listening. Captions repair the fragment you missed. They are not a channel to read instead of listening, and people who try to read every word answer worse.
- Where this is the wrong tool. OfferPilot runs on Windows 10 and 11 only, no macOS, Linux or mobile, so a phone screen where the audio never touches the computer is out, as is an in-person interview. If the real issue is hearing loss rather than an unfamiliar accent, the accommodation route beats any of this: ask the recruiter for a captioned platform, an interpreter, or the questions in writing. That request is routine, usually granted, and costs you nothing inside the round.
Ten minutes the day before
Open a meeting with yourself on the exact platform named in the invite and confirm the caption button exists on that account. Fix your Windows sound devices so a headset cannot drag you down to hands-free quality. Put the dial-in number in your phone. Write the three scripts on a sticky note by the monitor, because under pressure you will not compose them. Then rehearse against the accent you actually struggle with, using a recorded talk rather than a friend who already speaks the way you are used to, and pull prompts from a question list so the phrasing is unfamiliar too. The free tier is five minutes a day with no card, and the clock only advances while the copilot is receiving audio, so a rehearsal run costs close to nothing.
Related guides
- Deaf Interview Accommodations: CART vs Captions vs Interpreter
- Non-Native English Speaker Interview Tips: Recovery Phrases
- Technical Interviews in English as a Second Language: 40 Phrases
- Panic Attack During an Interview: What to Say in the Moment
- Interview Accommodation Request Emails: 6 Copy-Paste Templates