How to transcribe an interview: real examples and tools

Software turns an interview recording into text in minutes. Deciding what the transcript should say takes longer: whether the ums stay, what to write when a word is drowned out or two people talk at once, what to do when the conversation wanders off, and how to avoid typing the whole thing yourself.

If you interview people regularly, Wispr Notetaker records and transcribes the conversation on your Mac or Windows computer as it happens, labels who said what, and spells names the way your guest spells them.

What an interview transcript looks like

An interview transcript puts each speaker’s turn on its own line, starts every turn with the speaker’s full name, and keeps the words in the order they were said. Here is the opening exchange from episode 436 of NASA’s Houston We Have a Podcast, in which host Nilufar Ramji interviews Carlos García-Galán, NASA’s Moon Base program manager, as it appears in NASA’s published transcript.

Nilufar Ramji: Carlos, thank you so much for being on Houston We Have a Podcast. First, we’d like to know how you got here. Can you tell us a little bit about yourself and what you do here at NASA?

Carlos García-Galán: Sure, it’s a pleasure to be here with you guys today. So, I’ve always been really inspired by exploration, and pushing the limits, doing things that nobody else has done.

Four conventions make a transcript usable:

• Full names on every turn, so anyone can lift a line without scrolling up to check who said it.
• A new line for each turn.
• Timestamps at the start of each turn or every 30 to 60 seconds.
• Square brackets for anything that isn’t speech, such as [laughs], [long pause], or [inaudible], so nothing you add gets mistaken for something the speaker said.

Decide how verbatim to be before you start

Settle how much of the speaker’s delivery to keep before you type a word, or the transcript will drift between styles halfway through. The same line from the NASA interview, in the three standard styles:

‍

Clean verbatim is the safe default. It reads easily and still lets you put a sentence in quotation marks, and dropping “um” and “uh” is fine as long as you do it everywhere and say so if the transcript will be cited.

AI transcription tends to land between clean verbatim and edited. A tool that quietly fixes “to did” into “to do” has made the line more readable and less exact, which matters the moment you quote it.

How to handle the messy parts

• A word you can’t make out: don’t guess. Write [inaudible] with the timestamp, such as [inaudible 12:04], so anyone checking can find the spot.
• A word you can hear but aren’t sure of: put your best guess in brackets with a question mark, as in [Galán?].
• Two people talking at once: transcribe the voice you can follow and mark the overlap with [crosstalk], rather than inventing an order for words that came at the same time.
• An off-topic stretch: summarize it in brackets, such as [3 minutes on scheduling the follow-up], but only when you are sure nothing in it will matter. The small talk at the start of an interview sometimes holds the best quote.
• Names and jargon you’ve never heard: ask the guest to spell their name and company at the start of the recording, and list the terms you expect them to use so you can check against it afterward. These are the most common errors in any transcript, human or AI.

Transcribing interviews for qualitative research

When the transcript is research data, for a dissertation, a study, or a product research program, your method decides the style:

• Thematic analysis usually works from clean verbatim, because the analysis is about what people said.
• Conversation and discourse analysis need true verbatim with pauses, overlaps, and fillers marked, because how people said it is the data.

Whichever you choose, state it in your methods section and apply it to every interview in the study.

Research transcripts also have to protect the people in them:

• Replace names with pseudonyms or participant numbers, such as P3.
• Remove details that could identify someone, like an employer or a small town, in line with what your consent form promised.
• Keep timestamps or line numbers so every excerpt you quote can be traced to the original.
• Check that your ethics protocol allows the software you plan to use, since some review boards specify where participant audio may be processed and stored.

For product research teams, the guide to an AI notetaker for user researchers covers sessions, consent, and getting transcripts into research tooling.

How long it takes, and the ways to do it

Typing an interview yourself takes several hours for every hour of audio, and longer when the audio is rough or the transcript has to be true verbatim. That is still worth it for a single short clip where every word matters. For anything more, the real choice is between software, which is fast and cheap but needs checking, and a human service, which costs more and takes longer but puts someone accountable behind every word.

• Typing it yourself suits one short interview where every word matters, at the cost of several hours of work per hour of audio.
• Your video platform’s built-in transcript covers occasional remote interviews on one platform. It depends on host settings and your plan, and does nothing once the interview moves to another platform or into a room.
• An automated transcription service handles interviews you’ve already recorded, but it adds an upload step after every interview, and speaker names usually have to be added by hand.
• An on-device notetaker such as Wispr Notetaker fits interviews you run regularly, remote or in person. It records live, so it has to be running during the interview.
• A human transcription service is the choice when the exact words carry legal or editorial weight. It costs money per minute of audio and takes longer than software.

If the interview is already sitting in a file, an upload service or a human transcriber is the route, because Wispr Notetaker captures audio live and cannot process a recording made elsewhere. For the next interview, Notetaker’s transcript builds while you talk, with no upload step and no speaker names to fill in afterward. It is included in Wispr Flow’s free plan.

Record the next interview so it transcribes well

Most transcription errors start in the recording, and in an interview you control only your half of the audio.

Remote interviews

Ask the guest to join from their own computer, wear headphones, and sit somewhere quiet. Headphones matter more than people expect: a guest on laptop speakers feeds your voice back into their microphone, and the transcript picks up both of you at once.

In-person interviews

Set the laptop or recorder between you, angled toward the guest. A room with soft furnishings beats a café or a bare meeting room that echoes.

Phone interviews

A call on your mobile never passes through the computer doing the recording, which makes it the hardest format to transcribe. Take the call on your computer instead. FaceTime, WhatsApp, and browser calls play through the machine.

Get consent before you record

Tell the guest you’re recording before you begin, then ask again once the recording is running. That way their agreement is on the transcript itself.

In the U.S., recording laws differ by state. Some require the consent of only one person in the conversation, others require everyone’s, and the rules can differ between phone calls and in-person conversations.

When you and your guest are in different states, follow the stricter rule. The Reporters Committee for Freedom of the Press keeps a state-by-state Reporter’s Recording Guide covering both.

Outside the U.S., check the rules where each of you is. When the stakes are high, talk to a lawyer.

If a guest is wary of how long their words will be kept, Wispr Notetaker can delete transcripts automatically after 1, 7, 30, 90, 180, or 365 days. You set the window in Notetaker’s settings.

How to transcribe an interview with Wispr Notetaker

Wispr Notetaker records on your computer instead of joining the call, so it works the same on Zoom, Google Meet, Teams, a Slack huddle, a browser call, or across a table, and nothing extra appears in the guest’s window.

Before the interview

• Put the interview on your calendar with the guest on the invite.
• Add the guest’s name, their company, and any terms they’re likely to use to your personal dictionary.

Notetaker reads both, so an unusual surname or product name comes out spelled correctly instead of as the nearest common word.

During the interview

• Click the notification that appears when the interview starts to begin recording. For a call that was never on your calendar, Notetaker detects it and asks whether to take notes, including FaceTime and WhatsApp calls on your computer.
• For an in-person interview, open Notetaker and record from your laptop’s microphone.
• Glance at the live transcript, which separates you from the guest, to check an answer without losing your place.

After the interview

• A second AI pass runs through the recording, fixes mistakes, and adds full name labels.
• If anyone is unlabeled or labeled wrong, name them once and the fix applies across the whole transcript.
• Share the transcript by link or email.

Notetaker transcribes 21 languages, including Spanish, French, German, Japanese, Hindi, and Arabic, so interviews in another language follow the same steps.

What a real interview test showed

In our podcast test, Wispr Notetaker made 11 word errors on a 730-word interview segment and Zoom’s built-in transcript made 19. The bigger difference was speakers: Zoom put every line under one name, while Wispr Notetaker separated the host from the guest.

We used the same NASA interview excerpted above, because NASA’s published transcript gave us an answer key for a real two-person conversation with casual phrasing, technical vocabulary, and an uncommon name on each side. We played the episode into a Zoom call, transcribed it with both tools, and scored each against NASA’s transcript from the host’s introduction to the end of García-Galán’s answer about NASA’s Ignition initiative. Fillers, punctuation, capitalization, and numerals versus spelled-out numbers were ignored on both sides.

‍

The six errors both tools share are mostly places where NASA’s transcript keeps a stutter such as “the the,” so the third row is the fairer comparison. Each tool’s other notable errors:

• Zoom wrote the host’s name as “Nova Ramji,” heard “Ignition” as “Ignatia” and “diluted” as “deluded,” and dropped a short phrase entirely.
• Wispr Notetaker wrote “Nilofer” for “Nilufar,” a one-letter slip on an unusual name.

Zoom labels speakers by which participant’s microphone the audio comes through, and because the whole episode played through one connection, Zoom had no way to split it. On a call where the guest joins from their own device, Zoom would label them by name. Wispr Notetaker told the two voices apart from a single audio stream, which is the situation you are in whenever an interview happens across a table.

Find and check quotes before you use them

You rarely need a whole interview transcript again. You need one line, or an answer you half remember from three weeks ago.

Finding the line

• Inside one interview, Notetaker’s Ask anything box answers questions about that conversation, so you can ask what the guest said about pricing instead of scrolling for it.
• Across your history, Notetaker searches every meeting you have recorded and links each answer back to the conversation it came from.
• When you write, connect Notetaker to Claude, ChatGPT, or Gemini through MCP and draft from the guest’s actual words rather than your memory of them.

Checking it against the audio

Even a 1.5% error rate means a mistake every few sentences, and the mistakes cluster in exactly what a quote depends on. Before any line goes inside quotation marks, jump to its timestamp, play it while reading along, and confirm:

• the spelling of every name
• every figure and unit
• the exact wording, including any slip the tool tidied for you

Published transcripts need the same check. NASA’s own transcript of the interview above introduces it as episode 439, while the audio and the episode page both say 436.

Match the setup to the kind of interview

• Hiring interviews: the guide to an AI notetaker for recruiters covers intake calls, candidate screens, and panel debriefs.
• Face-to-face interviews: how to record and transcribe an in-person meeting goes deeper on microphones, rooms, and labeling voices that share one device.
• Remote interviews: the guides for Zoom and Google Meet cover each platform.

Limits to know

• Long interviews: recordings stop at 2 hours by default. You can raise the cap to 3 hours in Notetaker settings, and a warning before the cap lets you extend it.
• HIPAA-covered interviews, such as conversations with patients, need a different route for now. Wispr Flow is HIPAA-ready for dictation under a signed BAA, but Wispr Notetaker isn’t available on a BAA account yet.
• Compliance: Wispr Flow is SOC 2 Type II and ISO 27001 compliant on every plan, including the free one.

Spend the interview listening, not typing

The best interviews come from paying attention to the person in front of you, and that is hard to do while you take notes. Wispr Notetaker handles the record so you can follow up on the answer you just heard instead of scrambling to write it down. It starts with your calendar invite and dictionary, so your guest’s name is spelled right from the first line. It tells host and guest apart even when both voices come from one microphone, then runs a second pass after you hang up to fix what it missed. Weeks later, you can ask for a line and get it back with a link to where it was said.

Notetaker runs on Mac and Windows, works on any platform your guest picks or across a table, and is included in Wispr Flow’s free plan. Get started with Wispr Notetaker before your next interview.

Frequently asked questions

Can I transcribe an interview when the guest sends the meeting link?

Yes. Wispr Notetaker records through your computer’s audio rather than through the meeting platform, so you don’t need to be the host, and the account doesn’t need cloud recording turned on.

Will my guest be notified that I’m transcribing?

Not automatically, since nothing joins the call. You can turn on a setting that drops a note into the meeting chat as soon as recording starts: “Hi — quick note that I’m using Wispr Notetaker to transcribe this meeting so I can stay present. Let me know if you’d like me to turn it off.” You can change the wording, and so can your organization.

What happens to the audio of my interviews?

Interview audio is encrypted and stored only temporarily, on your device and in some cases in secure cloud storage, so Notetaker can create the transcript and verify its quality. After that it is deleted automatically. Notetaker doesn’t create voiceprints of you or your guests, your content is never sold, and it isn’t used to train AI models without your consent. You can delete any interview, transcript, or note at any time.

Can I bring interview transcripts over from Otter or Granola?

Yes. Notetaker imports your Otter or Granola history in a few steps, including transcripts and summaries, so older interviews stay searchable next to new ones.

‍