Verbatim vs clean verbatim transcription

What full verbatim transcription includes

Everything audible, in the order it happened. A full verbatim transcript (some services say "true verbatim" or "strict verbatim") keeps filler words like "um", "uh", "like" and "you know", every false start and abandoned sentence, repeated words, and the short interjections a listener makes while someone else talks ("right", "mm-hmm", "yeah"). Slang and grammatical slips stay exactly as spoken. Nonverbal events that a reader would otherwise miss go in square brackets: [laughs], [long pause], [crosstalk], [inaudible].

The point of all this is that the record itself is the product. A researcher doing discourse or conversation analysis reads hesitation as data: a three-second pause before "yes" is a different answer from an instant one. A lawyer wants the false start on the record because the abandoned sentence may matter. A market researcher watching how people react to a concept needs the nervous laughter, not just the words around it.

The cost is bulk and labor. People speak at roughly 130 to 160 words a minute, so an hour of conversation holds about 8,000 words before you add a single bracket. Typing full verbatim by hand roughly doubles the work compared with a clean transcript, because every stumble you would normally skip must be caught and placed; how long transcription takes walks through that arithmetic. Full verbatim is also harder to read, which is exactly why nobody should choose it by default.

What clean verbatim removes, and the line it never crosses

Clean verbatim removes sounds, never substance. Out go pure hesitations ("um", "uh"), stutters and immediately repeated words ("we, we tried"), false starts that the speaker abandoned and restarted, and listener noises that answer nothing. Obvious slips of the tongue can be silently corrected ("the 15th, I mean the 16th" becomes "the 16th"). Grammar gets smoothed only lightly: a run-on sentence can gain a full stop, but the speaker's word choice, dialect and phrasing stay theirs.

The line is meaning. Some words that look like filler carry real content. "Sort of", "I guess", "kind of" and "maybe" are hedges: a source who says "I guess the numbers were fine" has not said the numbers were fine, and a transcript that deletes the hedge has changed the testimony. The same goes for a "yeah" that actually answers a question rather than filling space. A good clean verbatim rule is: remove a word only if the sentence means exactly the same without it, and when in doubt, keep it.

If you will publish quotes from the transcript, journalism's standard applies whatever you call the style: tidy the stumbles if you like, but never alter what the person said, and check any quote you print against the recording, not against the transcript.

Want to try it now? Upload a file to FastScribe. One file a week is free, no signup.

Which style fits which job

Choose full verbatim when the delivery is evidence: thesis and qualitative research interviews headed for coding or discourse analysis, focus groups where reactions matter as much as words, legal and disciplinary records, oral history projects that archive speech as it was, and any recording where you expect someone to later argue about what exactly was said. If your recordings are thesis or dissertation interviews, ask your supervisor or methods text which convention your analysis needs before you transcribe anything, because adding brackets after the fact means listening to everything again.

Choose clean verbatim for nearly everything else: interviews you will quote in an article, podcast transcripts and show notes, meeting and lecture records, video captions, and any transcript whose job is to be read, searched or skimmed. For captions there is an extra push toward clean: caption lines run about 42 characters, and every "um" you keep steals space from words that matter on screen. The complete guide to interview transcription covers where this decision sits in the wider interview workflow.

If a project needs both, transcribe full verbatim first and derive the clean version from it by deletion. Going the other direction is not editing, it is re-transcribing.

The same passage, both ways

Here is one exchange rendered in each style. Full verbatim first. Interviewer: "So how did the, uh, the first launch go?" Maya: "So, um, we, we shipped the, you know, the first design in March, and it, uh [laughs], honestly it just, people didn't, didn't get it. Like, the numbers were, I guess the signups were sort of fine? But retention, retention was, yeah. [pause] We pulled it after three weeks." Interviewer: "Mm-hmm."

Now clean verbatim. Interviewer: "How did the first launch go?" Maya: "We shipped the first design in March, and honestly people didn't get it. I guess the signups were sort of fine, but retention was bad. We pulled it after three weeks."

Notice what the clean version kept: "I guess" and "sort of" survive because Maya is hedging about the signups, and deleting the hedges would turn a doubt into a claim. Notice also what it had to add: "retention was, yeah" contains a judgment Maya never finished saying, so the clean version writes "bad" on her behalf. That is a real editorial decision, defensible for a readable record, wrong for research data. Full verbatim never has to make it, which is precisely why researchers pay its cost.

What an automated transcript gives you, and how to reach either style

An automated transcript is neither style. Speech recognition models drop some hesitations on their own and keep others, collapse some false starts and transcribe others faithfully, and they do not mark laughter, pauses or crosstalk in brackets. So treat the automated draft as raw material: it lands closer to clean verbatim than to full, but it will not meet either standard until you do a pass yourself. How to edit and clean up a transcript covers that pass; the style decision from this page is its first step.

This is also where the honest limit sits. If your project requires guaranteed full verbatim, every utterance captured and notated to a convention, no automated tool can promise that, FastScribe included, because the model itself smooths speech unpredictably. That job belongs to a human transcriber or to your own ears and keyboard; DIY vs automated vs human transcription compares those routes. Services that offer both styles usually price full verbatim above clean verbatim, because it is genuinely more work.

For clean verbatim, though, an automated draft plus a deletion pass is the fast route. Keep your own copy of the recording whichever way you go: you will need it to resolve hedges and unfinished sentences during the style pass, and to check any quote you publish. If who said what matters too, speaker labels are their own decision with their own rules, and they belong in both styles; the styles only govern which words and sounds appear after the name.

Whichever style you choose, write it down. A one-line note at the top of the document, for example "clean verbatim, hedges kept, slips corrected, names verified", tells a collaborator, or you in six months, what standard the text follows and what was deliberately left out.

Where FastScribe fits

For clean verbatim, an automated draft plus a deletion pass is the fast route, and that is the job FastScribe fits. It transcribes on our own hardware, no third-party AI service receives your audio, and the audio is deleted the moment the transcript is ready, so keep your own copy of the recording for the style pass. Your first file needs no account, up to 10 minutes and 50 MB, one per rolling week; a free account gives 5 files a day at up to 30 minutes and 100 MB each, with transcripts kept for 7 days; Pro is $12 a month for files up to 5 hours and 2 GB and adds DOCX export, with TXT, SRT and VTT on every plan. If who said what matters, speaker labels are automatic on the no-account file, a free account labels one file a day with the switch on the uploader, and Pro labels every file; renaming is free on any plan and the names carry into every export. What no automated tool can give you, this one included, is guaranteed full verbatim: the model smooths speech unpredictably, so a transcript notated to a convention belongs to a human transcriber or to your own ears and keyboard.