How to Clean an Interview Transcript

What a Messy Interview Transcript Looks Like

Raw interview text often mixes timestamps, caption fragments, generic labels, filler, and uncertain facts. Improve formatting without manufacturing confidence.

Speaker 1:
so when did you actually
leave the company

Speaker 2:
uh I think it was late 2021
or maybe january 2022

Interviewer:
When did you actually leave the company?

Guest:
I think it was late 2021, or maybe January 2022. I would need to check.

Decide Whether You Need Verbatim or Clean-Read Text

Verbatim text preserves filler, false starts, interruptions, and incomplete sentences. Clean-read text removes distracting artifacts while preserving meaning and voice. An edited interview is a separate stage: raw transcript → cleaned transcript → verified transcript → edited interview.

Make Speaker Labels Consistent

Choose one label system and keep it throughout. Interviewer and Guest work well; verified names can be clearer. Consistency matters more than complex labels.

Speaker 1:
...
Host:
...
Interviewer:
...

Interviewer:
...
Guest:
...

Do Not Guess Speaker Identity

Automatic speaker numbers can be wrong. Do not replace them with names until the audio confirms who spoke. Wrong attribution is more serious than an unattractive label.

Remove Interview Timestamps Selectively

Remove caption-level timestamps for a readable transcript when they are not needed, but keep meaningful points for quote checks, clips, research, sensitive claims, and editorial review. See the remove timestamps guide.

07:14 Leaving the Company
18:30 First Product Failure
31:45 Hiring the Early Team

Join Broken Caption Lines

Join fragments that belong to one sentence, but preserve speaker changes, question-and-answer boundaries, real paragraphs, lists, and quotes. See the remove line breaks guide.

Raw:
we did not have a sales team
at the time so every founder
was answering customer calls

Clean:
We did not have a sales team at the time, so every founder was answering customer calls.

Keep Questions and Answers Easy to Scan

Keep dialogue as dialogue instead of compressing it into prose. The content stays the same, but separate turns make the interview easier to scan.

Handle Filler Words Carefully

Some um, uh, false starts, and low-value repetition can be removed in clean-read text. Do not remove kind of, probably, maybe, or I think when they communicate uncertainty or tone.

Preserve Uncertainty

A polished answer must not become factually stronger than the source. Keep qualifiers around dates, revenue, employee counts, percentages, timelines, and recollections.

Source:
I think we had around 40 customers. It may have been closer to 30.

Clean-read:
I think we had around 40 customers, though it may have been closer to 30.

False Starts and Self-Corrections

Remove an explicitly corrected word when the correction is clear. Do not choose between “September—maybe October” when the speaker did not resolve the uncertainty.

Remove Obvious Duplicate Fragments

Reduce export duplication such as “the biggest problem / the biggest problem was hiring,” but preserve natural emphasis such as “really, really difficult.”

Handle Interruptions and Overlapping Speech

Preserve the interaction instead of rewriting it as a summary. If overlapping audio is unclear, use [overlapping speech] or [unclear] rather than inventing a quote.

Interviewer: So the reason you—
Guest: No, the pricing was not the reason.
Interviewer: Okay. What was?
Guest: Retention.

Verify Names, Proper Nouns, Dates, Numbers, and Money

Check people, companies, products, places, technical terms, dates, employee counts, prices, and percentages against the recording or reliable production material. Do not correct them just because a likely spelling looks right.

Direct Quotes Need an Extra Verification Pass

Before publishing a prominent quote, check wording, negation, numbers, qualifiers, names, and context against the audio. A cleaned transcript is a working source, not an automatically verified quotation database.

Do Not Improve the Guest’s Answer Too Much

Fix formatting, not personality. Turning a guest’s conversational answer into corporate prose is rewriting. A good cleaned interview still sounds like the people in the recording.

Full Before-and-After Interview Example

This cleanup removes timestamps and caption breaks while preserving I think, maybe, I do not remember exactly, and under twenty.

Before:
Speaker 1:
how many people were on the team

Speaker 2:
uh I think fourteen
maybe maybe sixteen
I do not remember exactly

After:
Interviewer:
How many people were on the team at that point?

Guest:
I think fourteen—maybe sixteen. I do not remember exactly, but definitely under twenty.

Cleaning Research Interviews

Use a conservative standard when wording, pauses, hesitation, emotional emphasis, or terminology may be evidence for analysis. Decide the transcription standard before cleanup.

Cleaning Customer Interviews

Preserve the customer’s own language for product, UX, sales, or positioning research. Classification and interpretation belong later, not in the transcript itself.

Cleaning Interviews for Publication

A publication workflow can clean formatting, verify quotes and facts, lightly edit readability, and send sensitive changes for review when required. Significant restructuring should be treated as an edited interview.

Interview Transcript to Article, Notes, and Clips

A cleaned transcript can support an article, internal notes, or clips, but each is a separate transformation. Verify facts before paraphrasing, keep source timestamps for clips, and keep the cleaned transcript separately.

Interview Transcript Cleanup Checklist

• Keep an untouched original.
• Choose verbatim or clean-read text.
• Remove only timestamps you no longer need.
• Keep question-and-answer structure visible.
• Make labels consistent and verify identity before using names.
• Remove only obvious duplicate fragments.
• Preserve meaningful repetition, uncertainty, and interruptions.
• Mark genuinely unclear overlapping speech.
• Verify names, dates, numbers, money, and direct quotes.
• Keep the interviewee’s voice and separate cleanup from rewriting.

Frequently Asked Questions