Quick answer
Verbatim transcription records everything: every "um," stutter, false start, repeated word and audible sound like laughter or a cough. Clean verbatim keeps exactly what the speaker meant, in their own words, but strips out the noise so it reads smoothly. A third style, edited (sometimes called intelligent) transcription, goes further and fixes grammar and structure.
Pick strict verbatim when how something was said is evidence or data (legal, linguistics, discourse analysis). Pick clean verbatim for almost everything else: research interviews, podcasts, meetings, captions and content repurposing.
| Dimension | Strict verbatim | Clean verbatim | Edited / intelligent |
|---|---|---|---|
| Filler words (um, uh, like) | Kept | Removed | Removed |
| Stutters and false starts | Kept | Removed | Removed |
| Repetitions | Kept | Removed unless for emphasis | Removed |
| Grammar errors | Kept as spoken | Mostly kept as spoken | Corrected |
| Non-speech sounds | Tagged, e.g. [laughs] | Only if meaningful | Rarely |
| Pauses | Often marked | Not marked | Not marked |
| Slang and dialect | Kept | Kept | May be standardized |
| Readability | Low | High | Highest |
| Typical uses | Legal, linguistics, discourse analysis | Research, podcasts, meetings, captions | Articles, reports, published Q&As |
| Effort to produce | Highest | Medium | Medium to high |
What each style looks like
The fastest way to understand the difference is to see the same 15 seconds of speech in all three styles.
Strict verbatim
Um, so I, I think the... the main problem was, like, we didn't, uh, we didn't really have a budget for it? [laughs] And then, you know, (pause) the vendor kind of, they, they pulled out.
Clean verbatim
So I think the main problem was we didn't really have a budget for it. And then the vendor pulled out.
Edited
The main problem was the lack of a budget, and then the vendor withdrew.
All three are "accurate." They are accurate to different things. Strict verbatim is accurate to the sound. Clean verbatim is accurate to the words the speaker meant to say. The edited version is accurate to the meaning, but it is no longer quite the speaker's voice, which is why you would not put it in quotation marks in a news story without care.
Strict verbatim, in detail
Strict verbatim (also called true verbatim or full verbatim) aims to make the transcript a faithful record of the audio. A good strict verbatim transcript includes:
- Filler words and hesitations: um, uh, er, hmm, like, you know, I mean.
- False starts and self-corrections: "We went to the, I mean, we drove to the office."
- Stutters and repeated words: "I, I, I don't know."
- Non-speech sounds: [laughs], [coughs], [sighs], [phone rings], [crosstalk].
- Pauses: often marked as (pause) or with a duration, such as (3 seconds).
- Interruptions and overlapping speech: marked with a dash or a crosstalk tag.
- Unclear audio: [inaudible 00:14:32] or [unclear], usually with a timestamp so someone can re-listen.
Who needs strict verbatim
- Legal and law enforcement. In depositions, interviews and recorded evidence, a hesitation or a self-correction can change how a statement is read. Courts and agencies often specify verbatim, and court reporting has its own formatting rules that vary by jurisdiction.
- Linguistics and conversation analysis. Researchers studying turn-taking, hesitation or discourse need every sound. Some use detailed systems such as Jefferson notation that mark overlaps, intonation and pause length.
- Some qualitative methods. Narrative and discourse analysis often treat the way something was said as data.
- Market research where emotion matters. A long pause before "yes, I'd buy it" tells you something a clean transcript hides.
The cost of strict verbatim is readability. A page of it is hard to skim, and quotes pulled from it often need tidying before anyone outside the project sees them.
Clean verbatim, in detail
Clean verbatim (also called non-verbatim or intelligent verbatim by some services, which adds confusion) keeps the speaker's words and sentence structure but removes the parts that carry no meaning. Typical clean verbatim rules:
- Remove fillers like um, uh and filler uses of "like" and "you know."
- Remove stutters and false starts, keeping the final, intended phrasing.
- Remove repeated words unless the repetition is for emphasis ("it was very, very expensive" stays).
- Keep the speaker's vocabulary, slang and grammar. Do not "correct" someone's dialect.
- Tag non-speech sounds only when they matter to meaning, such as [laughs] after a sarcastic remark.
- Keep [inaudible] markers with timestamps.
Clean verbatim is the default at most transcription services and is what most AI transcription tools produce out of the box. It is the right choice when people need to read the transcript rather than analyze the audio through it.
Who needs clean verbatim
- Qualitative research using thematic analysis. You care about what participants said, not how many times they said "um." See our guide on how to analyze interview transcripts.
- Podcasts and video. Show notes, blog posts and quotes all start from clean verbatim. Our podcast repurposing guide walks through that workflow.
- Meetings and interviews for hiring or journalism. Readable, quotable and still faithful.
- Captions. Caption guidelines generally favor accuracy to the dialogue, but captioners routinely drop meaningless fillers to keep reading speed manageable. More on this below.
Edited (intelligent) transcription
Edited transcription treats the audio as a draft. The transcriber or editor fixes grammar, breaks up run-on sentences, removes tangents and may reorder for clarity. It is useful for:
- Published Q&A interviews and executive profiles.
- Turning a talk or webinar into a written article.
- Internal reports where nobody needs the exact words.
The risk is drift. Every edit is a judgment call about what the speaker meant. If you publish an edited transcript as a quote, many editorial teams send it back to the speaker for approval.
Notation conventions you will see
There is no single universal standard, but most style guides use some version of these:
| Notation | Meaning | Used in |
|---|---|---|
| [inaudible 00:12:04] | A word or phrase could not be heard, with timestamp | All styles |
| [unclear: budget?] | Best guess at a hard-to-hear word | All styles |
| [crosstalk] | Two or more people speaking at once | All styles |
| [laughs], [coughs] | Non-speech sound | Verbatim; clean only if meaningful |
| (pause) or (3s) | A noticeable pause, sometimes with length | Verbatim |
| Word- | A word cut off mid-way | Verbatim |
| -- | Interruption or abandoned sentence | Verbatim and clean |
| SPEAKER 1:, Interviewer: | Speaker labels | All styles |
| [redacted] or [NAME] | Removed for privacy | Research and legal |
Whatever you choose, write your conventions down in a short style sheet and share it with everyone producing or reviewing transcripts. Inconsistent tagging across a set of interviews is a bigger problem than choosing the "wrong" style.
How the choice affects captions
Captions sit in an interesting middle ground. Accessibility guidance asks for captions that are accurate to the dialogue and include meaningful sound information, such as [door slams] or [music]. At the same time, captions need to be readable at speed, and a stream of "um, uh, I, I" on screen slows viewers down.
In practice, most professional captioning keeps the speaker's words and meaning (clean verbatim, plus important sound cues), while transcripts used as evidence or data stay strict. If you are producing captions for compliance, see our video accessibility and WCAG guide, and for the related terminology our explainer on captions vs subtitles.
How to choose: a short decision list
- Will anyone analyze how something was said? Hesitation, emotion, interruptions, turn-taking. If yes, strict verbatim.
- Could the transcript become evidence? If yes, strict verbatim, and follow any rules your court, agency or ethics board specifies.
- Will people mostly read, search or quote it? Clean verbatim.
- Will it be published as polished prose? Start from clean verbatim, then edit, and keep the clean version as your source of truth.
- Unsure? Transcribe in strict verbatim. You can always clean a verbatim transcript; you cannot recover fillers and pauses that were never written down.
Common myths
- "Verbatim means more accurate." Both styles should be accurate. They are accurate to different targets: sound versus meaning.
- "Clean verbatim lets you fix grammar." It shouldn't. Correcting grammar is editing, and it changes the speaker's voice.
- "AI transcripts are verbatim." Most speech recognition engines lean clean: they often drop fillers and do not tag sounds consistently. If you need strict verbatim from an AI transcript, budget for a human pass.
- "You only need one style per project." Many teams keep a verbatim master for analysis and generate a clean version for sharing.
Working with transcripts from video
If your source material is on YouTube, such as recorded talks, interviews, lectures or panels, you rarely need to start from zero. YouTubeTranscript.dev pulls a video's captions, or AI-transcribes it when captions don't exist, and exports TXT, SRT or VTT. Treat that output as a clean verbatim first draft: review it against the audio, add speaker labels and tags where your style requires them, and use the remove timestamps tool when you want plain readable text. For bulk interview or research collections, the API returns the same transcripts as JSON.
Frequently asked questions
What is the difference between verbatim and clean verbatim?+
Verbatim (also called strict or true verbatim) captures every sound the speaker makes, including filler words, stutters, false starts, repetitions and non-speech sounds like laughter. Clean verbatim keeps the speaker's words and meaning but removes fillers, stutters and false starts so the text is easier to read.
Is clean verbatim the same as an edited transcript?+
No. Clean verbatim removes noise but does not rewrite. An edited or intelligent transcript goes further: it fixes grammar, restructures run-on sentences and may condense rambling passages. Clean verbatim should still sound like the speaker; an edited transcript may read more like written prose.
Which transcription style is used for court and legal work?+
Legal work such as depositions, hearings, police interviews and evidence recordings generally uses strict verbatim, because hesitations, self-corrections and pauses can matter to how testimony is interpreted. Court reporters follow the formatting rules of their jurisdiction, so check the local standard.
Should qualitative research interviews be transcribed verbatim?+
It depends on the method. Discourse, conversation and narrative analysis usually need full verbatim because the way something is said is part of the data. Thematic analysis and most UX research work well with clean verbatim. Decide before you transcribe and note the choice in your methods section.
Does clean verbatim cost less than verbatim?+
Usually. Strict verbatim takes longer to produce because the transcriber has to capture and tag every sound, so many services charge more for it. Clean verbatim is the default at most transcription services.
Do AI transcription tools produce verbatim or clean verbatim?+
Most speech recognition engines output something close to clean verbatim by default: they tend to drop many filler words and stutters, and they do not tag non-speech sounds consistently. If you need strict verbatim, plan on a human review pass or a tool setting that explicitly keeps disfluencies.