How to Clean Interview Transcripts for Publication

A rough interview transcript is not ready for readers just because the words are on the page. Speaker labels drift. Paragraphs run too long. Filler words pile up. Automated punctuation can make a half-finished thought look polished, or make a clear answer look confused.

The cleanup pass is where an editor turns that rough text into something a reader can follow while protecting what matters most: the speaker’s meaning, voice, and context.

Quick answer: Clean an interview transcript by fixing transcript errors, improving readability, labeling uncertainty, and removing clutter that does not carry meaning. Do not rewrite a speaker into a smoother version of themselves. If a passage needs substantial rewriting to work in publication, paraphrase it instead of presenting it as a direct quote.

This is different from finding the best quote or building a full article workflow. The job here is narrower: prepare a reader-facing transcript or excerpt that is clear, accessible, and honest about the recording it came from.

Decide what kind of published text you are making

Before cleaning a line, name the deliverable. A full transcript, a lightly edited Q&A, and a direct quote inside an article should not be cleaned to the same standard.

Deliverable	What the reader expects	What you can clean	What you should not do
Reader-facing transcript	A readable version of the interview	Speaker labels, paragraphing, punctuation, obvious filler, repeated false starts, inaudible labels	Turn speech into polished essay prose
Published excerpt	A selected passage that still represents the conversation	Shorten setup, remove irrelevant side trails, label omissions, clean light clutter	Remove context that changes the speaker’s point
Direct quote in an article	The source’s words, not your rewrite	Only the publication’s allowed quote treatment, such as rare ellipses or bracketed clarification	Correct grammar, smooth awkward phrasing, or combine separate thoughts as if continuous
Paraphrase	The reporter’s accurate rendering of the point	Rewrite for clarity with attribution	Put quotation marks around cleaned-up wording

The safest rule is simple: the more you rewrite, the less it belongs inside quotation marks.

What to clean in an interview transcript

A good cleanup pass removes friction without removing the speaker.

Fix transcript errors against the audio

Start with words that are wrong, missing, or assigned to the wrong speaker. Automated transcripts can be useful first drafts, but they are not truth-verifying records. Accents, dialects, low-resource languages, overlapping speech, and long pauses can all produce errors.

Fix:

• misheard names, places, organizations, and technical terms
• speaker labels that switched mid-answer
• punctuation that changes the meaning of a sentence
• words inserted by the transcript that the source did not say
• words omitted where the audio is clear

If you cannot confirm the wording, do not guess. Label the uncertainty.

Make speaker turns readable

Long blocks of speech are hard to scan. Break them where the speaker changes topic, answers a new question, qualifies a claim, or moves from story to explanation.

Readable paragraphing is not cosmetic. It helps the reader understand the structure of the answer.

Weak transcript cleanup:

INTERVIEWER: What changed after the audit? DIRECTOR: The biggest change was staffing we had been using temporary coverage and the night team was doing intake and follow-up at the same time so after the audit we split those roles and added a second review on Fridays.

Better cleanup:

INTERVIEWER: What changed after the audit?

DIRECTOR: The biggest change was staffing. We had been using temporary coverage, and the night team was doing intake and follow-up at the same time.

After the audit, we split those roles and added a second review on Fridays.

The words are still the speaker’s. The structure is now readable.

Remove clutter that does not carry meaning

Most reader-facing transcripts do not need every um, uh, repeated article, or abandoned first syllable. Oral history guidance often treats transcription as more than copying every sound; the transcript should remain readable while maintaining the interviewee’s integrity.

Usually safe to remove:

• isolated filler words: um, uh, ah
• accidental repeated words: we we started
• abandoned starts that add no meaning: I was going to— no, the main issue was...
• interviewer backchannels that only show listening: mm-hmm, right, yeah
• obvious transcription artifacts from the tool

But do not remove clutter automatically. A hesitation can matter if it shows reluctance, emotion, uncertainty, or a careful correction.

Keep or label:

• a long pause before a sensitive answer
• a self-correction that changes the point
• laughter that affects tone
• a repeated phrase used for emphasis
• a stumble that is part of the story, not just noise

What to label instead of clean away

Some material should not be silently removed because the reader needs to know something happened in the recording.

Use labels sparingly and consistently.

Situation	Better label	Why it helps
You cannot hear a word	[inaudible] or [inaudible at 12:14]	Shows uncertainty instead of inventing text
You are unsure of a name	[name unclear]	Prevents a wrong proper noun from becoming published fact
People talk over each other	[crosstalk]	Explains why wording may be incomplete
The source laughs while answering	[laughs]	Preserves tone when it affects meaning
A long pause carries meaning	[pause] or [long pause]	Shows hesitation without dramatizing it
You remove private information	[redacted]	Makes the omission visible
You add a clarifying reference	[the March audit]	Helps the reader follow without pretending the speaker said those words

Do not use brackets to smuggle in your interpretation. [angrily], [obviously lying], or [does not answer] are editorial judgments, not transcript labels.

What to leave alone

Cleaning should not make every source sound like the same person.

Leave alone:

• word choices that show region, profession, age, or personality
• informal contractions that are natural to the speaker
• careful hedges such as I think, as far as I know, or from what I saw
• grammar that is understandable and part of the speaker’s voice
• repetitions used for emphasis
• emotionally important pauses or restarts

Be especially careful with dialect and non-native speech. A transcript should help the reader understand the speaker. It should not exaggerate pronunciation, mock usage, or flatten a person’s voice into a generic newsroom style.

Before and after: cleaning a transcript for readers

Here is a fictional example from a local health interview. The goal is a reader-facing transcript excerpt, not a polished article quote.

Rough transcript

INTERVIEWER: Can you explain why the clinic cut evening hours?

PATEL: Uh, yeah, so we we didn't close the clinic. That's important. We paused evening hours in March because two nurses got reassigned to the shelter and if we kept doing nights we'd have unsafe staffing. And I don't mean unsafe like bad care, I mean not enough staff in the building. We told people, or we tried to tell people, by text and posters but some of those didn't go out until, I think, the second week.

INTERVIEWER: So patients were not notified before the change?

PATEL: Some were. Some weren't. That's on us.

Reader-facing transcript cleanup

INTERVIEWER: Can you explain why the clinic cut evening hours?

DR. PATEL: We didn't close the clinic. That's important. We paused evening hours in March because two nurses got reassigned to the shelter, and if we kept doing nights, we'd have unsafe staffing. I don't mean unsafe like bad care. I mean not enough staff in the building.

We tried to tell people by text and posters, but some of those didn't go out until, I think, the second week.

INTERVIEWER: So patients were not notified before the change?

DR. PATEL: Some were. Some weren't. That's on us.

What changed

Edit	Why it is acceptable in a readable transcript
Removed Uh, yeah, so	Filler did not carry meaning
Removed repeated we	Accidental repetition, not emphasis
Added paragraph break	Separates the staffing explanation from the notification issue
Added punctuation	Helps the reader follow the sentence without changing words
Standardized speaker label to DR. PATEL	Gives the reader a clear source identity
Kept That's important	Shows emphasis and source intent
Kept I think	Signals uncertainty about timing
Kept Some were. Some weren't. That's on us.	The short phrasing is the point; smoothing it would weaken the answer

This is cleaned, but it is not rewritten.

If you were writing an article, you might quote only the strongest exact sentence:

“Some were. Some weren’t. That’s on us,” Patel said.

You might paraphrase the rest:

Patel said the clinic paused evening hours after two nurses were reassigned to shelter work, creating what she described as a staffing risk.

That distinction matters. The transcript can be lightly cleaned for readability. The article quote should not become your improved version of the source’s sentence.

When to paraphrase instead of quote

Some passages are true, useful, and still bad direct quotes. That is normal.

Paraphrase when:

• the speaker’s meaning is clear but the sentence is tangled
• the passage needs grammar repair to make sense
• the usable point is spread across several minutes
• the quote would require several bracketed clarifications
• the wording is too murky to represent fairly
• the transcript is uncertain and the audio does not resolve it
• the speaker makes an important factual claim in a way that needs context around it

Quote when the source’s exact wording matters: accountability, emotion, memorable phrasing, a clear admission, a precise explanation, or a sentence the writer cannot improve without changing its force.

AP guidance is strict on this point: quotations should not be taken out of context, and flawed or unclear quotations should be paraphrased if the paraphrase remains true to the original. If the meaning is too unclear to paraphrase accurately, do not use it.

A publication cleanup checklist

Use this pass after the rough transcript exists and before the transcript or excerpt goes to a reader, editor, producer, or audience.

1. Source and access

• The original audio is preserved.
• The rough transcript is preserved or versioned.
• The published transcript has the interview date, speaker names or roles, and any necessary context.
• If the transcript accompanies audio, it provides a real text alternative for people who cannot hear the recording.
• Important non-speech information is included when it affects meaning: [laughs], [pause], [crosstalk], or relevant visual context for video.

2. Accuracy

• Proper names, titles, places, acronyms, and numbers are checked.
• Speaker labels are correct.
• Uncertain words are labeled instead of guessed.
• AI-generated punctuation has been reviewed for meaning.
• Any suspected transcript insertion has been checked against the audio.

3. Readability

• Speaker turns are separated cleanly.
• Long answers are broken into logical paragraphs.
• Filler words are removed only when they do not carry meaning.
• False starts are removed only when the final sentence makes the abandoned start irrelevant.
• Repetitions are kept when they show emphasis, emotion, or uncertainty.

4. Integrity

• The speaker still sounds like themselves.
• Dialect, accent, and non-native speech are not exaggerated or mocked.
• Brackets clarify; they do not editorialize.
• Ellipses, if used, do not change meaning.
• A cleaned transcript passage is not automatically treated as a direct quote for an article.

5. Publication notes

• If the transcript is lightly edited, say so.
• If sections are omitted, make the omission visible.
• If material is redacted, label it.
• If the excerpt has been condensed, explain the basis for condensation.
• If the published piece uses paraphrases, keep quotation marks only around words the source actually said.

Where DraftCut fits

DraftCut is useful when the transcript is the working surface, but the original interview still needs protection.

For an interview excerpt, you can work transcript-first: read the speaker turns, make edit decisions in text, preview the derived playback, and export the result. The important product habit is non-destructive editing: the original audio and original transcript stay unchanged, while edits create derived playback and export state.

That does not mean DraftCut fixes a transcript automatically or replaces listening. It means the cleanup and excerpting process can stay close to the source while giving the editor room to compare, revise, and recover decisions without overwriting the original interview.

The practical standard

Clean interview transcripts for readers, not for vanity.

Remove what blocks comprehension. Label what the reader needs to know. Preserve the source’s voice when the wording matters. Paraphrase when clarity requires rewriting. And keep the original recording close enough that every cleaned line can be defended against what was actually said.

On your next interview, run one cleanup pass with this question in front of you: Am I making the transcript easier to read, or am I making the source easier to misread?