bananote
Log in
← All postsGuides

How to transcribe an interview: a step-by-step guide

Pick a transcript style, get consent, record cleanly, then transcribe, format, check and anonymize an interview you can quote and code.

The Bananote teamNovember 19, 202614 min read

To transcribe an interview, decide what the transcript is for, get the interviewee's consent and record close to them in a quiet room. Then turn the audio into text with an app, Microsoft Word or by hand, pick a style (verbatim, intelligent verbatim or clean), and check every name, number and quote against the recording.

An interview transcript is evidence, and transcribing an interview properly means the text holds up when someone checks it against the audio. Someone will quote it in a paper, code it for a thesis, file it in an archive or compare it with the next candidate's answers, so a rough draft isn't enough. For the basics that apply to any recording, such as consent law and setting up for a phone or video call, see how to transcribe a conversation. This guide covers what's specific to interviews.

What you need: the interviewee's consent (plus a signed release if the interview will be archived or published), a phone or laptop to record on, a quiet room, headphones for the checking pass, and an honest block of time (step 5 has the numbers).

The short version

  1. Decide what the transcript is for: research, journalism, oral history or hiring.
  2. Get consent, and a release form if the interview will be published or archived.
  3. Record with the transcript in mind: a list of names, one voice at a time, the original audio file kept.
  4. Choose a style: verbatim, intelligent verbatim or clean.
  5. Transcribe with an app, in Microsoft Word or by hand.
  6. Format it: a header block, speaker labels and timestamps.
  7. Check it against the recording.
  8. Anonymize it if your research needs that.
  9. Code it and pull the quotes for your paper.

Step 1: Decide what the transcript is for

The purpose decides everything after it: how exact the text has to be, whether you need timestamps and who will see it. Settle it before you press record.

Purpose Style to use Timestamps The extra step
Research or a thesis Intelligent verbatim, or full verbatim if you're studying how people speak Every few minutes, so each quote can be traced Anonymize, then code
Journalism Intelligent verbatim for your working copy Beside every line you might quote Check each quote word for word against the audio
Oral history Lightly edited, keeping the narrator's dialect Every five minutes The narrator checks it and signs a release
Hiring Clean, or structured notes instead of a full transcript Optional The same format for every candidate

The oral history row follows Oregon State University Libraries' oral history guide, which asks for a transcript that drops verbal tics but keeps the narrator's dialect, with timestamps every five minutes.

Also ask whether you need a full transcript at all. A transcript costs hours, and the National Park Service's oral history guidance lists timed indexing (a list of topics with the time each one starts) as a cheaper option when an interview doesn't merit full transcription. Oregon State's guide makes the same point: a good interview index can do a lot of the work.

For an interview, consent has two parts: permission to record, and permission to use what you recorded. The second part depends on where the transcript is going.

  • Research and theses. Follow the protocol your ethics board (an IRB in the US) approved, including who may hear the recording and where the files live. If you plan to use a transcription app or hire a transcriber, check the protocol allows it before the first interview, not after.
  • Oral history. The Oral History Association's best practices call for documenting the narrator's informed consent and securing a signed legal release, ideally when the interview is completed, so any sensitive issues that came up can be dealt with before anyone signs. The release is what lets an archive share and reuse the interview, and because an interview is made by both people in it, the National Park Service has interviewers sign the release too.
  • Journalism. Tell the person you're recording and agree what's on the record before the first question.
  • Hiring. Tell candidates you're recording and why, and check your company's policy on how long recordings and transcripts are kept.

Whatever the purpose, ask the consent question again once the recording starts, so the yes is on the audio.

Step 3: Record with the transcript in mind

Most transcript trouble starts in the room. Four habits save hours later:

  • Write a proper noun list first. Names, places, organizations and jargon you expect to hear. The National Park Service recommends handing one to any hired transcriptionist, and it speeds up your own checking pass just as much.
  • Ask for spellings on tape. "How do you spell the name of that clinic?" takes a moment and saves a search later.
  • One voice at a time. Nod instead of saying "mm-hm" over answers. Overlapping speech is hard to transcribe, for a person or for software.
  • Keep the original audio file. You'll need it to check quotes. Recording on your phone's built-in voice recorder and uploading the file gives you both the file and the transcript.

In Bananote you can record the interview live on your iPhone, iPad, Mac or in your browser, or upload an audio file you recorded elsewhere. Bananote records through AirPods and Bluetooth microphones, so a clip-on Bluetooth mic on the interviewee puts the voice that matters closest to the microphone. For a remote interview, upload the audio file the meeting platform saved.

Step 4: Choose a transcript style

Interview transcription comes in three styles, and the differences are easiest to see side by side. Take one answer from a nurse interviewed for a thesis on night shifts (an invented example) and write it in each of the three styles.

Full verbatim keeps every filler, stutter, repeat and sound (Rev):

"Um, so the, the worst part of nights is, uh, like, around 4 a.m., when you, you've got nobody to call. [laughs] Well, nobody who picks up."

Intelligent verbatim keeps every word but drops the fillers and false starts. SpeakWrite calls it the most common style for interviews:

"So the worst part of nights is around 4 a.m., when you've got nobody to call. [laughs] Well, nobody who picks up."

Clean, also called edited, rewrites for readability and is the style used for publication:

"The worst part of night shifts is around 4 a.m., when there's no one to call who will answer."

Look at what the clean version lost: the joke, and the speaker's own words. That's fine for a published Q&A. It's a problem in a thesis, because a clean transcript can't be quoted as what someone said. If you only have a clean transcript, go back to the audio before you quote.

How to choose:

  • Full verbatim when the way people speak is your data: conversation analysis, discourse analysis, or anything legal.
  • Intelligent verbatim for most research interviews and for a journalist's working copy. Cornell's course guide for student interviewers asks for exactly this: transcribe what you heard, but leave out the "ums" and "uhs" (Cornell University).
  • Clean for a published interview or hiring notes, with quotes checked against the audio.

The names aren't used consistently. Rev calls intelligent verbatim "non-verbatim" or "smooth", and some services call it "clean verbatim". Agree on what the words mean before you hire anyone or hand a transcript to a supervisor.

Step 5: Transcribe the interview

With Bananote

Record the interview live or upload the audio file. Bananote turns the audio into text in 100+ languages and builds a structured note with a summary. It's the same recorder students use for lecture transcription, so an interview works the same way.

Set up a custom template for interviews once and apply it to every one in the project. Headings like "participant ID", "questions asked", "answers in two sentences", "lines worth quoting" and "follow-ups" give each note the same shape, which matters when you're comparing twelve interviews. Whatever tool makes the first draft, step 6 is where you add or check the speaker labels and timestamps your format needs.

Your notes are private by default and shared only if you choose, and the AI service providers Bananote uses may not train on your data (privacy policy). For an interview in another language, keep the transcript in the language it was spoken and translate the note into the language you're writing in. There's more on that in this guide to transcribing in more than one language.

In Microsoft Word

Word has its own Transcribe feature (Microsoft Support):

  1. Open Word for the web in Edge or Chrome, or Word for Microsoft 365 on Windows (on Windows it's limited to some account types). You need a Microsoft 365 subscription.
  2. Go to Home > Dictate > Transcribe.
  3. Choose Start recording, or Upload audio and pick a .wav, .mp4, .m4a or .mp3 file.
  4. Keep the Transcribe pane open while it works.
  5. Word separates the speakers as "Speaker 1", "Speaker 2" and so on. Edit a label and choose Change all to rename every instance, and select Edit transcript section to fix mistakes while you play back the timestamped audio.
  6. Choose Add all to document, or add the sections you want one by one.

Three limits matter for an interview project. A Microsoft 365 subscription covers 300 minutes of uploaded audio a month (30,000 with a Microsoft Copilot license). A document holds one transcript, and starting a new one deletes the old one from the pane. The audio is saved to a Transcribed Files folder in OneDrive, which your ethics protocol may or may not allow. The Microsoft 365 Insider announcement confirms the subscription requirement and the same four file types. Do the math before fieldwork: 300 minutes is five one-hour interviews a month, so a thesis with fifteen of them needs three months of allowance. On a Mac, use Word for the web.

If you'd rather type it yourself in Word, play the audio in a separate player that slows playback and pauses from the keyboard. Oregon State's guide recommends Express Scribe for this, and the National Park Service mentions USB foot pedals for hands-free control.

By hand

Typing an interview yourself is slow. Here's how long five guides say it takes:

Source Time per hour of audio
Rev About 4 hours
Minnesota Historical Society About 4 hours (one hour per 15 minutes of audio)
Cornell University course guide 4 to 6 hours
National Park Service 6 to 8 hours
Oregon State University Libraries 8 hours for a clean, finished transcript

Plan on 4 to 8 hours for every hour of audio. A 45-minute interview is 3 to 6 hours of work. The low end is a first draft from clear audio. The high end includes formatting and checking, and Rev notes that muffled audio or background noise takes longer still.

The method that works: listen once all the way through and note the hard parts, type in short chunks on the second listen, then listen a third time to fix what you missed (SpeakWrite's guide, linked in step 4). Where you can't make out a word, mark it with the time, like [inaudible 00:12:40], or type your best guess with a question in brackets, as Cornell's guide suggests, and ask the interviewee later.

Hiring a transcriptionist

A professional costs more than an app and takes longer to come back, but makes sense when every word has to be exact. Give them a proper noun list, a sample transcript showing your format, your style choice from step 4 and a confidentiality agreement (National Park Service).

Step 6: Format the transcript

Start every transcript with a header block. Oregon State's guide asks for the narrator, interviewer, date, location and total length. Cornell's adds the names and date on the first page and a short version in the page header of every page. A research version looks like this:


Project: Night shift nursing (MA thesis)
Interview ID: P04
Interviewee: P04 (real name in the identifiable copy only)
Interviewer: Sam Chen (SC)
Date: 2026-03-12
Location: Staff room, by agreement
Length: 47:12
Audio file: P04_2026-03-12.m4a
Style: intelligent verbatim
Checked against audio: yes, 2026-03-15

Then the body. Start a new paragraph each time the speaker changes, with their initials or role at the start. Rev suggests writing a full name the first time someone speaks and initials or a role after that. Add a timestamp at regular points, such as every five minutes, and beside anything you might quote:


[00:05:00]
SC: What's the hardest part of a night shift?

P04: So the worst part of nights is around 4 a.m., when you've
got nobody to call. [laughs] Well, nobody who picks up.

SC: Who would you call, if someone picked up?

P04: The charge nurse on [inaudible 00:05:41], or the doctor on call.

Keep sound notes like [laughs] for moments where they change the meaning, as here, where the laugh marks a joke. Cornell's guide says tone of voice and body language mostly belong in your own interview notes, not in the transcript.

Step 7: Check the transcript against the recording

Every transcript gets one full pass with the audio playing, whether a person or software made it. The National Park Service says plainly that transcripts from voice recognition software always need checking for accuracy, and Oregon State's guide asks you to re-read the finished transcript while listening to the recording. Work in this order:

  1. Proper nouns. Check every name against your list from step 3.
  2. Numbers. Dates, doses, amounts and years.
  3. Every [inaudible] marker. Listen again, slower, and fill in what you can.
  4. Who said what. Fix any line that sits under the wrong speaker.
  5. Every line you'll quote. Word for word against the audio, with its timestamp.

In oral history, the narrator often checks the transcript too. Send it with the passages and names you struggled with highlighted, and agree on limits first, because some narrators want to rewrite their spoken words into polished prose (Oregon State University Libraries). In journalism, ask your editor before sharing a transcript with a source.

Step 8: Anonymize the transcript

If your ethics approval promises anonymity, the transcript is where you deliver it. The UK Data Service's guidance on anonymizing text data sets out the practice:

  • Replace names and other direct identifiers with pseudonyms or descriptive tags, like [colleague 1], and use the same replacement across every transcript.
  • Watch for "jigsaw identification": a hospital, a ward and a start year can identify someone together even when no name appears.
  • Don't strip so much that the meaning goes. Replace a detail with a meaningful description rather than deleting it.
  • Mark every replacement clearly, and keep an anonymization log of what you changed.

Here's a line from the same invented nurse interview, before and after:

"I trained at Northgate General in Leeds with my cousin Joan, and we both moved to the cardiac ward in 2015."

"I trained at [a large teaching hospital] in [a city in northern England] with [relative], and we both moved to [a specialist ward] in [mid-2010s]."

The log entry: "P04, 00:12:30: Northgate General to [a large teaching hospital]; Leeds to [a city in northern England]; Joan to [relative]." Keep the identifiable transcript and the log wherever your protocol says, separate from the copies you share. Remember the header block too: the shared copy says P04, not a name.

Step 9: Code the transcript and pull quotes for your paper

Coding starts after the transcripts are clean. In Braun and Clarke's reflexive thematic analysis, the first phase is reading and re-reading the transcripts and making notes, the second is coding them with short labels in two or more rounds, and only then do you start building themes (Braun and Clarke).

  • Give every transcript an ID (P01, P02) and keep the timestamps, so any quote traces back to a point in the audio.
  • Read once with no codes. Just margin notes on what stands out.
  • Code in rounds, and keep a codebook: each label and what it means, so "isolation" means the same thing in transcript 3 and transcript 12.
  • Build a quote bank as you go: interview ID, timestamp, the quote and its codes. Check each quote against the audio and against the anonymized copy before it goes into a draft.

Bananote helps with the finding, not the thinking. Keep each project's interviews in one folder, search them for a word, and ask the chat on an interview's notes a question like "Did this participant talk about calling for help at night?" to find where it came up. Then go back to the transcript to read it in full. The codes and themes are your analysis, and Braun and Clarke describe qualitative analysis as subjective work by the researcher. For the wider research workflow, see how Bananote works for researchers handling interviews, talks and papers.

Common mistakes when you transcribe an interview

  • Quoting from a clean transcript. Clean text is a paraphrase. Quote from verbatim or intelligent verbatim, checked against the audio.
  • No timestamps. Without them, nobody can trace a quote back to the recording, including you.
  • Leaving the release form for later. An oral history without a signed release is hard for any archive to share.
  • Removing names but leaving the jigsaw. A ward, a city and a year can identify someone just as well as a name.
  • Planning for typing time only. Budget 4 to 8 hours per hour of audio by hand (Rev puts it at about four, Oregon State University Libraries at eight for a finished transcript), and leave room for the checking pass.
  • Running out of Word's allowance mid-fieldwork. 300 minutes a month goes quickly in a busy interview week.

Frequently asked questions

How long does it take to transcribe an interview?

By hand, plan on 4 to 8 hours for every hour of audio. Rev and the Minnesota Historical Society put it at about four hours, Cornell's interviewing course guide at 4 to 6, the National Park Service at 6 to 8, and Oregon State University Libraries at 8 for a clean, finished transcript. Muffled audio, background noise, hard-to-follow speakers and unfamiliar jargon push it higher.

How do you transcribe an interview in Word?

In Word for the web or Word for Microsoft 365 on Windows, go to Home, then Dictate, then Transcribe, and record or upload an audio file. It needs a Microsoft 365 subscription, which covers 300 minutes of uploaded audio a month. Rename the speakers, correct mistakes against the audio, then add the transcript to your document.

What is intelligent verbatim transcription?

Intelligent verbatim keeps every word the speaker said but removes filler words, stutters and false starts. SpeakWrite calls it the most common style for interview transcripts, and it reads cleanly while still keeping the speaker's words. Full verbatim keeps the fillers, and clean or edited transcription rewrites for readability.

Should you include "um" and "uh" in an interview transcript?

Only if how people speak is part of what you are studying, as in conversation analysis, or if the transcript has a legal use. For most research and journalism, leave them out and use intelligent verbatim.

How do you format an interview transcript?

Start with a header block: project, interview ID, interviewee, interviewer, date, location, length, audio file name and transcript style. Begin a new paragraph each time the speaker changes, with their initials or role first, and add timestamps at regular points and beside any line you might quote.

Do you have to transcribe the whole interview?

No. If the interview is only for your own notes, a timed index (a list of topics with the time each one starts) can be enough, and oral history programs use one as a cheaper alternative to a full transcript. Transcribe in full anything you will quote or code.

How do you anonymize an interview transcript?

Replace names and other direct identifiers with pseudonyms or descriptive tags, use the same replacement in every transcript, and watch for details that identify someone in combination, such as a workplace, a city and a year. Mark each replacement and keep a log of what you changed, stored apart from the shared copy.

Can Bananote transcribe an interview in another language?

Yes. Bananote transcribes 100+ languages and can translate the note into another language, so you can keep the interview in the language it was spoken and work from a translation.


Your next interview is a recording away from a transcript and a structured note you can search, quote and code. Record or upload your first lecture free.

Get started free

Try it on your next lecture

Hit record, stay present, and let Bananote handle the notes, flashcards, and quizzes.