How to Transcribe and Edit an Interview 2026: A Practical Guide

To transcribe and edit an interview, you record with consent, run the audio through a transcription tool or your own ears to get a first draft, then edit that draft in a fixed order: format, grammar, structure, verification. Sixty minutes of clean audio takes AI a few minutes and a human typist roughly four to six hours, so the editing stage is where the real work sits.

An interview transcript is a speaker-labelled written record of what was said, produced either by a person typing from audio or by an AI tool, then cleaned up and formatted for reading, quoting, publishing or research.

That distinction matters more than it sounds. A transcript that is accurate but unreadable gets skimmed. A transcript that reads well but changes what somebody said gets you a correction email, or worse.

What You Need

Four things, and only one of them is expensive.

• The recording. Two copies in different places before you do anything else. One on the recorder, one uploaded somewhere you can reach from another machine.
• A transcription method. AI software, a human service, or your own keyboard. All three work; they trade speed against accuracy in predictable ways.
• A word processor with audio. Something that can play the recording while you read the text, ideally synced so you can click a sentence and hear it. Scrubbing back and forth by hand is the single biggest time sink in this job.
• A style sheet for the job. Even a one-page note about speaker labels, timestamps and which brackets you use. Decide it once and stick to it.

Tools named constantly in this space include Whisper, Otter, Rev, Reduct, Temi and Trint. All of them handle clear audio and a single speaker well. All of them struggle with the same three things: proper nouns, jargon, and overlapping speech. Plan your time assuming you will fix names by hand.

Method	Speed for 1 hour of audio	Accuracy	Best for
AI transcription	3 to 10 minutes	Good on clean single-speaker audio; unreliable on names and jargon	Drafts, podcasts, first passes on any volume
Human transcription service	4 to 8 hours turnaround	High, especially with a glossary supplied	Legal, medical, executive and research interviews where accuracy is contractual
Manual typing	4 to 6 hours of your own time	As high as your typing, and you catch tone as you go	Short interviews, distinctive speakers, learning the craft

One free workaround keeps surfacing in podcasting forums: upload the audio as a private, unlisted video and use the platform’s auto-transcript, then edit. It costs nothing and the quality on clear speech is serviceable. Delete the video once you have the text.

Step-by-Step: How to Transcribe and Edit an Interview

1. Prepare the recording and define the transcript format

Before the first word gets typed, back up the audio and write down what this transcript has to do. A transcript for a magazine profile, a transcript for a dissertation appendix and a transcript for a podcast episode need almost nothing in common.

Build a speaker list first: full name, role, and the initial or short form you will use after the first mention. In a three-person conversation this takes two minutes and saves you an hour of hunting through “Speaker 2” later.

Then pick your style, because it changes every decision that follows.

	Verbatim	Clean verbatim
Filler words	Kept in full: um, uh, like, you know	Removed
False starts	Kept	Cut
Repetition	Kept	Reduced, unless it carries emphasis
Grammar	Left as spoken, punctuation added for reading	Corrected lightly so the sentence parses
Nonverbal sound	Recorded: [laughs], [pause], [inaudible]	Recorded only when meaning changes
Use it for	Oral history, legal, research data, anything quoted later	Publication, show notes, blog posts, video scripts

Set the punctuation conventions while you are at it. The common ones: use an ellipsis for a skipped word, square brackets for a transcriber’s note, and never a comma to join two unrelated fragments in a clean verbatim.

2. Transcribe the recording accurately

Work in short sections. Five to ten minutes of audio at a time, and stop at a natural break rather than a fixed clock. Long stretches without a pause are where attention drops and errors creep in.

Start each speaker turn on a new line, use the initial you defined, and type a timestamp at least every five minutes. Timestamps are what let a source, an editor or your future self find the exact sentence in the audio later.

Mark what you cannot hear rather than guessing. [inaudible] and [unclear: probable term] are honest and traceable; an invented word is neither. If a name matters and you cannot place it, write [surname unclear] and resolve it against your pre-interview notes.

Capture tone without transcribing every noise. A long pause before an answer, a laugh that undercuts a serious claim, a shift into a different register: those carry meaning. A door closing does not.

3. Complete a first-pass accuracy check

Do this before you touch style, and do it with the audio playing. Once you start smoothing sentences, errors hide inside the smoothing.

Check names, job titles, company names, dates, numbers and technical terms first. These are the highest-risk items because an AI draft gets them confidently wrong and because they are the details a subject will notice.

Then check attribution. Read each line against the audio and confirm the label is right. Speakers cross over constantly in a conversational interview, and a swapped attribution in a published quote is a genuine problem.

Finally, compare the draft with the recording in one pass at faster playback. You are listening for missing stretches, not for style.

4. How to edit a transcript without changing the speaker’s voice

This is where most transcripts get worse, not better. Over-editing produces a piece that is smooth, grammatical and no longer sounds like a human being talked.

Cut the mechanical noise. Filler words, stutters on the first syllable, repeated words, and false starts that lead nowhere all go. When a speaker says “the the reason I left was the deadline,” you write “the reason I left was the deadline.”

Keep the load-bearing words. Odd phrasing, a repeated noun that shows emphasis, an incomplete sentence that trails off, a pause that lets the point land: these are voice. Deleting them makes a speaker sound smoother and less interesting, and it is the point where editing turns into rewriting.

Tighten structure without rewriting sentences. Long interviews loop. The same point arrives at minute four, minute nineteen and minute forty-one, slightly differently each time. Pick the best version, cut the other two, and keep a note of where each survivor came from.

Stitching quotes from different parts of the interview is standard practice in journalism, and it is the point of a transcript. Two rules keep it honest: never join two fragments that contradict each other or change the sequence of events, and never make a speaker’s answer follow a question they were not asked.

Some transcripters and writers send the edited version back to the subject for a read-through. It is a small courtesy, and it catches things you cannot hear.

5. Format the transcript for its intended use

Formatting is where a draft turns into a document somebody can use.

Use full names on first appearance, then initials or short forms. Break paragraphs at a change of topic, not at a change of speaker alone. Keep each speaker turn to a readable block rather than one line per sentence. Put timestamps in the margin or at the start of the line, consistently, and never mix styles.

For academic work, APA 7th edition sets out the expected conventions: a title and the speaker’s name on the first page, an italicised label for each speaker, and a bracketed transcriber’s note for anything editorial. Chicago and AP differ on labelling and on how they handle in-text quotation of an interview, so check the guide your institution or publication actually uses rather than assuming.

Match the export to the destination. DOCX or PDF for editors, reviewers and archives. TXT or Markdown for search and analysis work. SRT or VTT for video captions, which need strict timing and no speaker labels in the text layer.

A copy-ready template, if you want a starting point:

INTERVIEW TRANSCRIPT
Project / Topic:
Date and location:
Duration:
Interviewer:
Interviewee:
Transcript style: verbatim / clean verbatim
Transcribed by / date:

[00:00]
INT: Opening question.

[04:12]
DR. NAME: Answer, with [inaudible] where speech is unclear and [pause] where the silence means something.

6. Proofread and verify the final transcript

Final pass, and it is a checklist rather than a read-through:

• Every name, title and number checked against the recording one last time.
• Every quotation you plan to publish matched word for word with the audio, including the quote marks and the ellipsis.
• Speaker labels consistent throughout, including the places where two people talked over each other.
• Consistent brackets, timestamps and file naming.
• Consent and permission on file, plus whatever handling instructions came with it.
• Unsensitive details removed from any copy that leaves your machine.

Deliver the final file plus the rough version if the transcript will be quoted. Editors and fact-checkers want to see what was removed, and a record of the edits protects you if a subject disputes a quote weeks later.

Common Mistakes and How to Fix Them

Over-editing until the speaker sounds like a press release

The fix is a rule you write down before you start: you may remove disfluency, never diction. If the speaker said “gave a damn,” that stays. If the speaker said “really made an effort,” that stays too, awkward as it reads.

Mislabelling speakers

Auto-diarization guesses, and on overlapping speech it guesses wrong. Fix it by voice and by content in the accuracy pass, then confirm with a full playback. In a final transcript, use [inaudible] for an unattributable remark rather than guessing a name.

Losing the question that made an answer make sense

Condensing an interview often strips the setup and leaves an answer floating. Decide in advance whether questions appear in the published version, and if they do, keep them short and cut the rambling preamble rather than the question itself.

Publishing quotes nobody verified against the audio

Verify every quote character by character before it ships. This is the number one complaint in writing forums about AI-assisted transcripts, and it is entirely avoidable. Treat every pull-quote as a separate verification task, not a copy-paste from your draft.

Trusting auto-diarization with proper nouns

Personal names, company names, product names and any jargon are the first casualties of automated transcription. Keep a glossary of expected names before you start and search for each one in the draft. Getting them wrong is the fastest way to lose a subject’s trust.

Editing the same document for six hours straight

A common working method among freelance writers is to transcribe the whole interview, take a short break, then re-read with fresh ears. Fatigue hides more errors than carelessness does, and the break costs ten minutes.

Tips for a Clean, Publication-Ready Interview Transcript

Keep the rough transcript. The edited version is the deliverable; the raw version is your evidence. Store them together, dated, and never overwrite one with the other.

Write a glossary before you start and keep it beside you. Names, companies, product terms, local references: a two-minute list up front prevents thirty minutes of hunting later.

Use consistent terminology across the whole transcript. If you introduce a concept in italics, keep it italicised every time, and do not alternate between a term and its synonym in the same passage.

Handle sensitive material deliberately. Interviews with anonymous or protected sources should be stored somewhere access-controlled, with identifying details stripped from any working copy. If the recording is the only evidence of what was said, say so in a header note.

Disclose the method when it matters. Readers and subjects both respond better to a transcript that states whether it was machine-generated, human-typed, or generated and then edited by a person.

Finally, think about what the transcript becomes next. Quotes, pull-outs, show notes, video captions, a research codebook: the finished document is usually raw material for at least three other things, and keeping it clean makes all of them faster.

Frequently Asked Questions

Conclusion

Knowing how to transcribe and edit an interview is mostly discipline about order. Get the recording backed up and the style chosen before you type a word, produce a first draft fast, run the accuracy pass with the audio playing, and only then edit for readability. Formatting and export come last, and the rough copy gets kept alongside the final one.

Start with those three things today: back up the file, decide verbatim or clean verbatim, and schedule the accuracy pass. Everything else gets easier once those are settled.