Verbatim vs Clean Verbatim Transcription: Which Style to Use

Verbatim transcription records speech exactly as spoken, including fillers such as "um", stutters, false starts and repetitions, while clean verbatim (also called intelligent verbatim) keeps the speaker's words but removes those stumbles so the text is easier to read. A third style, edited transcription, goes further and rewrites for clarity. Choose true verbatim when how something was said matters (legal work, conversation analysis), clean verbatim for most interviews, research and subtitles, and edited text for publishing.

Below you will find each style defined, the same short passage written all three ways, a comparison table, the situations that require each style and a checklist for choosing.

What is verbatim transcription?

"Verbatim" means word for word, but transcription providers and researchers use it for three different levels of detail. The names vary between style guides, so it pays to define them before you order or produce a transcript.

True verbatim (also "strict" or "full" verbatim) captures everything audible: every word, every "um" and "uh", stutters, false starts, repeated words, unfinished sentences, and usually non-speech sounds and pauses such as [laughs], [coughs] or (pause). Grammar is left exactly as spoken.

Clean verbatim (often called intelligent verbatim) keeps the speaker's own words and sentence structure but leaves out fillers, stutters, false starts and repetitions that add nothing to the meaning. It should not paraphrase or "improve" what the person said.

Edited transcription (sometimes "edited clean" or "light editing") also corrects grammar, tightens rambling sentences and may reorder or condense phrasing so the text reads like written prose. It is a faithful record of the meaning, not of the words.

Some research traditions add a fourth level: detailed transcription with notation for timing, overlaps and intonation, such as the Jefferson system used in conversation analysis. Our guide to transcription for qualitative research covers those conventions.

True verbatim vs clean verbatim: the same passage three ways

Here is a four-line excerpt from an interview, transcribed in each style.

True verbatim

I: So, um, how did you— how did you end up in nursing?
P: Uh, well, I mean, it was— it was kind of an accident, you know? [laughs]
P: My mum was a nurse and, like, I— I swore I'd never do it.
P: But then, um, after uni I was... I was just sort of, like, drifting. (pause)

Clean verbatim

I: How did you end up in nursing?
P: Well, it was kind of an accident. [laughs]
P: My mum was a nurse and I swore I'd never do it.
P: But then after uni I was just sort of drifting.

Edited

I: How did you become a nurse?
P: Almost by accident. My mother was a nurse, and I had sworn
   never to follow her, but after university I was drifting.

Notice what changes. The clean version removes "um", "uh", "I mean", "you know", "like" and the repeated "how did you" and "I", but keeps the speaker's informal phrasing ("mum", "uni", "kind of"). Whether to keep [laughs] is a style-guide decision; many clean verbatim guides keep non-verbal sounds only when they change the meaning. The edited version is shorter and smoother, but it is no longer the participant's voice.

Comparison table

When each style is required

Legal and compliance

Depositions, hearings, police interviews, insurance claims and recordings used as evidence normally call for true verbatim. A pause before an answer, a self-correction or a "yes... no, sorry" can change how a statement is read, so removing it is an editorial decision a transcriber should not make. Exact requirements depend on the court, jurisdiction and whoever orders the transcript, so follow their instructions or formatting rules.

Academic and qualitative research

The method decides. Conversation analysis, discourse analysis and linguistic studies need true verbatim, often with notation. Thematic analysis, applied research and most policy work typically use clean verbatim, because the focus is on what participants say rather than how. Whichever you choose, write the rules down before transcribing and report them in your methods section.

Journalism and media

Reporters usually work from clean verbatim and go back to the audio before quoting. Direct quotes should match what was said; cutting fillers is generally accepted, but changing words is not. For long-form features, an edited transcript can be a useful reading copy, as long as quotes are checked against the recording. Our interview transcription guide covers the recording and quoting workflow.

Subtitles and captions

Subtitles are almost always clean verbatim or lightly edited. Viewers have a few seconds to read each cue, and broadcasters' style guides set limits on reading speed and line length, so fillers and false starts are dropped unless they matter to the scene or a character. Captions for deaf and hard-of-hearing viewers add sound descriptions such as [door slams], which is a different kind of detail from verbatim fillers. See our SRT and VTT subtitle generator guide for formatting.

Business, meetings and podcasts

Meeting notes, podcast show notes and customer interviews are usually clean verbatim, often followed by a summary. True verbatim adds reading time without adding value here.

How AI transcripts behave

Automatic transcription does not map neatly onto these styles, so it helps to know what to expect.

• Fillers tend to disappear. In practice, Whisper-class models often drop "um", "uh" and repeated words, producing text that reads more like what the speaker meant than what they said. Researchers behind the CrisperWhisper paper (Interspeech 2024) note that standard Whisper eliminates many filler words and repetitions, and fine-tuned a variant specifically to keep them. OpenAI's own speech-to-text guide lists "preserving filler words" as something you can steer with a prompt, which tells you it is not the default.
• The result is close to clean verbatim, but not exactly. Models may also smooth over a false start or a half-finished word, and occasionally keep a filler they happen to recognise. The output is inconsistent rather than following a style guide.
• Non-verbal sounds and pauses are usually missing. Most tools do not mark [laughs] or timed pauses, and none we know of produce Jefferson-style notation reliably.
• Punctuation is a guess. Sentence breaks and commas are inferred, which can subtly change meaning in long, run-on answers.

So an AI draft is a good starting point for clean verbatim and needs a human pass for true verbatim. Audio quality matters too: our transcription accuracy tips explain how microphones, crosstalk and accents affect the draft.

How to choose a transcription style: checklist

Work through these questions before you transcribe or order a transcript:

• Will the transcript be used as evidence or in a legal process? Use true verbatim and follow the instructions of whoever requires it.
• Does your analysis look at how people speak (hesitation, repair, turn-taking)? Use true verbatim, possibly with notation.
• Do you mainly need what people said for themes, quotes or decisions? Clean verbatim is usually enough.
• Will the text be published or read by people who were not there? Consider an edited version, and keep the clean transcript as the source.
• Is it for subtitles? Clean verbatim, then shorten lines to fit reading speed.
• Have you written the rules down? List how to treat fillers, slang, non-verbal sounds, unclear passages and speaker labels, and apply them to every file.
• Who checks the transcript against the audio? Whatever the style, someone should listen at least once.

Turning an AI draft into the style you need

A practical workflow is to start with an automatic transcript and adjust it. In Audiotexta, for example, you can upload a file or record from the microphone, get a draft with speaker labels and timestamps, then correct it in an editor that highlights each word during playback and lets you click any word to jump to that moment in the audio. That makes it quicker to insert missing fillers and pauses for true verbatim, or to remove leftover stumbles for clean verbatim. You can try it in the web app on the free plan, which covers files up to 30 minutes.

When the transcript matches your chosen style, export it in the format your next step needs: DOCX or PDF for reading copies (Pro plan), TXT for analysis software, or SRT and VTT for subtitles. Keep a note of the style and rules you used with every file, so anyone reading it later knows exactly what "verbatim" meant in your project.