Interview transcription example: verbatim and clean
An interview transcript needs to preserve what the person meant, including the parts they were uncertain about. Compare the same short passage in two formats, then choose the conventions your project needs.
InterviewerWhat happened next?
Participant“I went back to the first screen.”
A verbatim transcript can retain fillers, repeated words, and self-corrections. A clean reading version removes selected speech clutter while preserving meaning. Define your convention, keep speaker labels and source timestamps consistent, and mark words you cannot establish. Never turn an uncertain answer into a confident quotation.
Start with the same source passage
One passage from a user-research interview, run through both conventions below. It contains a phrase nobody could make out, which is the case every real transcript hits and most examples leave out.
00:12 — Maya, interviewer: “What happened after you opened the page?” 00:15 — Noor, participant: “Um, I—I tried the date filter twice. It took, like, three minutes—no, maybe two. Then [words deliberately unspecified in this example]. I didn’t finish booking.”Download text file
The useful detail is the self-correction: Noor does not settle on an exact duration. That uncertainty belongs in both versions. The passage also says the booking was unfinished; it does not establish why.
Verbatim and clean transcription, side by side
| Format | The participant’s response |
|---|---|
| Verbatim example | [00:15] Noor: Um, I—I tried the date filter twice. It took, like, three minutes—no, maybe two. Then [unclear]. I didn’t finish booking. |
| Clean reading example | [00:15] Noor: I tried the date filter twice. It took three minutes—no, maybe two. Then [unclear]. I didn’t finish booking. |
[00:15] Noor: Um, I—I tried the date filter twice. It took, like, three minutes—no, maybe two. Then [unclear]. I didn’t finish booking.
[00:15] Noor: I tried the date filter twice. It took three minutes—no, maybe two. Then [unclear]. I didn’t finish booking.
Here, “verbatim” retains the scripted filler words and repeated start. The clean version removes “Um,” “I—I,” and “like,” but keeps the estimate, correction, unfinished phrase, and negative. Transcription services and research projects use these labels differently, so agree on the rules before editing a whole interview.
“The filter failed, so Noor abandoned the booking after two minutes” would be an interpretation with added facts. Neither version supports that sentence as a quotation or a confirmed observation. Keep possible explanations in a separate analysis note.
Choose speaker, timestamp, and uncertainty rules
| Element | A workable convention |
|---|---|
| Speakers | Use one stable label per person. If identity is unknown, use “Speaker 2” until verified. |
| Timestamps | Use elapsed time from the source recording at each speaker turn, and around passages needing review. |
| Unclear speech | Use [unclear] and a nearby source timestamp. Leave the words unresolved until you can establish them. |
| Overlapping voices | Use [overlapping speech] where separation is unreliable; do not assign guessed words to a speaker. |
| Meaningful sounds | Use a restrained label such as [laughs] when it affects interpretation. Do not infer an emotion from a pause. |
| Editorial additions | Put supplied context in brackets and keep it distinct from the speaker’s words. |
These are suggested conventions, not a universal transcription standard. For a project with an established style guide, follow that guide. If you cut an excerpt from a longer recording, retain its original time references or document the offset so another reader can find it.
Do not make a speaker’s grammar more formal just because you would write the sentence differently. A repeated word may be disposable in a reading copy, but hesitation can matter when your research concerns confidence, recall, or how someone makes a choice.
If you already have the interview as an SRT or VTT subtitle file, use the SRT to TXT converter to make a readable text copy. Keep timestamps when you need to check a quote against the recording.
Check quotations before using them
- Find the passage in the recording and listen to the surrounding question and answer.
- Confirm the speaker, wording, negations, names, and numbers. Check any apparent correction against what was actually said.
- If words remain unclear, leave the marker. Do not ask AI to complete a quote from context.
- Keep a reading copy separate from the source transcript. Document substantive edits and follow your project’s quotation and consent rules.
The Oral History Association recommends reviewing draft transcripts for accuracy and making editorial differences apparent. The GOV.UK user-research guide likewise separates what was observed from what researchers think it means. These principles help keep evidence usable when a quote moves into a report.
Before the reading copy goes anywhere, the transcript cleanup tool will rename Speaker 1 and Speaker 2, merge turns that were split mid-sentence, and take out the hesitation sounds, without touching a word anyone said.
Give AI a limited editing job
Review the transcript below without inventing missing speech. Keep speaker labels and timestamps unchanged. Preserve negations, uncertainty, and self-corrections. Suggest a separate clean reading version, removing only fillers or repeated starts that do not change meaning. Leave [unclear] markers in place. List each edit and any passage that needs checking against the recording. Do not add explanations, identify unknown speakers, or create direct quotations from a summary. Transcript: [paste the passage you are permitted to process]Download text file
An assistant can propose edits to text; it cannot verify an unheard recording from that text alone. In Phrase, record or import permitted interview audio, review the transcript, and open the attached source to check important passages. Speaker labels can be available, but names and assignments still need your review.
Media processing uses cloud services. One media-to-note conversion is free, and Pro includes the rest along with the AI features, at the allowance shown in the app. Keep the unedited source available when you prepare a reading copy or a research summary.
Sources & editorial notes
Sources were checked on September 8, 2026. Steps for Apple features follow Apple’s current documentation, linked below. The conditions change with each iOS release.
- Oral History Association: Archiving Oral History Transcript review, consistent project conventions, and disclosure of editorial differences.
- GOV.UK: Analyse a research session Check observations against recordings and distinguish what participants said from interpretation.
Published by Eversight, the makers of Phrase. Our editorial approach · Suggest a correction