Transcription Guidelines: Free Style Guide Template
Dzianis Kukareka, Founder · Updated

Transcription guidelines are the written rules that make every transcript in a project look and read the same: which verbatim level to use, how to label speakers, how often to add timestamps, how to mark audio you can't make out, and how to write numbers, names and sensitive details. You can copy the full style guide template below and fill in your choices in ten minutes. The rest of this page explains each decision, with the same passage shown in full verbatim and clean verbatim so you can see the difference.
There is no single correct standard. Researchers who studied transcription for qualitative projects concluded that no one format fits every method or setting, and that what matters is choosing rules up front and applying them consistently (McLellan, MacQueen and Neidig, Field Methods, 2003). Commercial services work the same way: each publishes its own guide, and its transcribers follow it to the letter.
What transcription guidelines should cover
| Decision | Typical options | Why it matters |
|---|---|---|
| Verbatim level | Full verbatim, clean verbatim, edited | Sets how much of the speech you keep; changes reading time and analysis |
| Speaker labels | Real names, roles, codes (P01) | Readers must know who said what; codes protect identity |
| Timestamps | Every speaker turn, every 30 seconds, every 2 minutes, only at tags | Lets anyone jump back to the audio to check a quote |
| Unclear audio | [inaudible], [unclear], [crosstalk], best guess with a question mark | A marked gap is honest; a silent guess is an error nobody can find |
| Non-speech events | [laughs], [pause], [phone rings], or nothing | Some analysis needs them, most business transcripts don't |
| Numbers and spelling | Words for zero to nine, numerals for 10+, US or UK spelling | Prevents five styles in one document |
| Sensitive details | Keep, replace with [name], replace with a pseudonym | Privacy, ethics approval and data protection rules |
| File format | DOCX, TXT, SRT/VTT, with or without a header | Decides whether the transcript works in your next tool |
Verbatim levels: clean verbatim vs full verbatim, with examples
The verbatim level is the biggest choice, so make it first. Here is the same ten seconds of an interview in three styles.
Full verbatim keeps everything: fillers, false starts, repetitions, stutters and audible events.
P03: So, um, I- I think the, the main problem was, uh, nobody actually owned the, you know, the handover. [laughs] Like, it just sort of... happened. Or didn't.
Clean verbatim (also called intelligent or true verbatim) removes fillers and stumbles but keeps the speaker's words and grammar.
P03: I think the main problem was nobody actually owned the handover. [laughs] It just happened. Or didn't.
Edited transcript also fixes grammar and tightens wording for publication. Use it only when the reader will never compare the text to the audio.
P03: The main problem was that nobody owned the handover. It either happened or it didn't.
A rule of thumb:
Full verbatim for legal work, conversation analysis, linguistics, and any research where hesitation and overlap are part of the data.
Clean verbatim for most interviews, qualitative research using thematic analysis, journalism, meetings and podcasts.
Edited for published Q&As, articles and show notes.
Clean verbatim has one trap: a filler is only removable when it carries no meaning. "I paid something like 200 dollars" needs its "like". GoTranscript's guide puts it the same way: remove fillers such as "um", "you know" and "I mean" when they add nothing, keep them when removing them changes the meaning (GoTranscript transcription guidelines). For a deeper definition, see our glossary entry on verbatim transcription.
Free transcription style guide template
Copy this into a document, delete the options you don't use, and send it to everyone who transcribes or edits for the project. Every line is a decision you would otherwise make again in every file.
TRANSCRIPTION STYLE GUIDE
Project: Customer onboarding interviews, wave 2
Owner: Research lead (questions go to this person, not to the group chat)
Version and date: v1, November 2026
1. Verbatim level
Clean verbatim. Remove um, uh, er, false starts and repeated words. Keep fillers that change meaning ("like 200 dollars"). Keep the speaker's grammar; do not correct it.
2. Header (top of every file)
File name, interview date, interviewer code, participant code, length of recording, transcriber, date transcribed.
3. Speaker labels
Bold label, colon, one space. Interviewer = INT. Participants = P01, P02 and so on, in the order they first speak. Never use real names, even if they are said aloud. Start a new paragraph at every change of speaker. If you are not sure who is speaking, write the most likely label followed by a question mark: P02?:
4. Timestamps
Format [HH:MM:SS], counted from the start of the file. Add one at every change of speaker and at every tag in section 5.
5. Unclear audio and events
[inaudible HH:MM:SS] for speech you cannot hear.
[unclear: best guess HH:MM:SS] when you can hear a word but are not sure of it.
[crosstalk HH:MM:SS] when people talk over each other and cannot be separated.
[pause] for silences longer than 5 seconds.
[laughs], [sighs], [phone rings] only when they affect meaning.
Do not invent other tags. Do not guess silently.
6. Numbers, dates and money
Spell out zero to nine, use numerals for 10 and above. Always numerals for money, percentages, times and measurements: $5, 30%, 9:30 AM, 4 km. Dates as 7 November 2026.
7. Spelling and names
US spelling. Product and company names as the company writes them (iPhone, YouTube). Check every proper noun against the glossary below; add new ones to it.
8. Sensitive information
Replace personal names with [name], or with the participant's code if they are a participant. Replace employers, towns, schools and other identifying details with a generic description in square brackets: [employer], [small town in the north]. Keep a separate, access-controlled log of every replacement.
9. Profanity and slang
Transcribe as spoken. Do not censor.
10. Non-English speech
Transcribe in the original language if you can, followed by [translation: ...]. Otherwise mark [foreign language HH:MM:SS].
11. Delivery
DOCX named PROJECT_PARTICIPANT_DATE.docx, plus SRT if the clip will be published with captions. Upload to the project folder, never by email.
12. Glossary
Terms, names and acronyms specific to this project, with correct spelling.
The template defaults to clean verbatim and codes instead of names because that suits most interview work. For full verbatim, change section 1 to "keep everything as spoken, including fillers, false starts, repetitions and stutters; mark audible breaths and overlaps", and add timestamps every 30 seconds so analysts can find their place.
How to label speakers in a transcript
Three habits prevent most speaker problems:
Pick one label style and never mix it. "Interviewer", "INT" and "Q" in one file look careless and break search.
Use codes when anyone outside the team will read the text. A name in a transcript is personal data under privacy law and most ethics approvals.
Mark uncertainty instead of guessing. A question mark after the label tells the reviewer exactly where to listen again.
If you start from an AI transcript, speaker labels come pre-filled by speaker diarization, which works best with a few distinct voices taking turns and gets worse as voices multiply. Rename the labels to your codes first, then check attribution where it matters; our focus group guide covers the hard cases.
Timestamps: how often and in what format
Timestamps exist so someone can check the text against the audio. Pick the frequency by how often that will happen:
| Use | Frequency | Format |
|---|---|---|
| Quotes for articles or reports | Every speaker turn | [00:14:32] |
| Qualitative coding | Every turn, or every 30 to 60 seconds in long answers | [00:14:32] |
| Meeting records | Every 2 to 5 minutes, or per agenda item | [00:14] |
| Legal or evidence work | As the instructing party requires | Per their format |
| Captions | Every line | SRT or VTT timing, not inline stamps |
Always count from the start of the file, not from the start of each section, and use the same format everywhere. Word-level timestamps from an AI tool make this almost free: you can click any word and hear it. More on that in word-level vs segment-level timestamps.
How to mark inaudible or unclear audio
The rule every professional guide shares: never guess silently. A wrong word that looks right is the most expensive error in a transcript, because nobody knows to check it. Use a small, fixed set of tags with a timestamp, so a reviewer can jump straight to the spot. GoTranscript, for example, separates [inaudible] (cannot be heard) from [unintelligible] (heard but not understood) and tells transcribers never to invent their own markings. Whatever tags you choose, list them in the guide and allow nothing else.
Transcription guidelines for AI-generated transcripts
Most transcripts now start as an AI draft and end with a human edit. The guidelines don't change, but where you spend effort does. AI rarely makes the formatting inconsistent; it makes plausible mistakes on names, jargon, numbers and crosstalk. So your edit pass should:
Rename speaker labels to your codes and check that the first few turns are attributed correctly.
Search for every proper noun and term in your glossary and fix the spelling.
Listen to every number, date and amount; these are cheap to check and costly to get wrong.
Replace guesses on unclear stretches with your tags.
Apply the verbatim level, the redaction rules and the file naming last.
Audio quality decides how much of this you have to do. Why AI transcription accuracy varies explains what to fix before you record. If you want a draft to apply this guide to, BlaBlaScribe's transcription tool gives you speaker labels and word-level timestamps in an editor, in 120+ languages, with accuracy up to 99.8% on clear audio; export DOCX or TXT for the edit pass. New accounts get 30 free minutes once, and the free plan exports SRT and TXT only. It won't apply your verbatim level or redaction rules for you; that is still the editor's job.
Transcription guidelines for qualitative research
Research adds three requirements to the template above. Decide them with your supervisor or ethics committee before the first interview, not after:
Anonymization rules that match your consent form and ethics approval, plus a separate key that links codes to real people.
What counts as data. If pauses, laughter or overlaps matter to your analysis, say so in the guide; clean verbatim deletes them.
A verification step. Someone checks each transcript against the audio, or a sample of them, and records that it was done.
Our guide to transcription for qualitative research covers storage and consent, and how to transcribe an interview covers the recording side.
Frequently asked questions
What are transcription guidelines?
They are a written set of rules for how transcripts in a project should be produced: verbatim level, speaker labels, timestamps, tags for unclear audio, numbers, spelling and redaction. Their job is to make every transcript consistent no matter who produced it.
What is the difference between clean verbatim and full verbatim?
Full verbatim keeps every filler, false start, repetition and audible event. Clean verbatim removes fillers and stumbles that carry no meaning but keeps the speaker's own words and grammar.
How do you write inaudible in a transcript?
Use a bracketed tag with a timestamp, such as [inaudible 00:12:45], so a reviewer can find the spot. If you can hear a word but aren't sure of it, mark your best guess as unclear rather than writing it as fact.
Should numbers be written as words or numerals in a transcript?
A common rule is words for zero to nine and numerals for 10 and above, with numerals always used for money, percentages, times and measurements. Whatever you pick, put it in the guide so every file matches.
Do I need a style guide if I use AI transcription?
Yes. AI gives you a draft, but the guide defines the edit: how speakers are labeled, which fillers stay, how unclear passages are tagged and what gets redacted.
Which verbatim level should I use for research interviews?
Clean verbatim suits most thematic analysis. Use full verbatim if hesitations, overlaps or the way something is said are part of what you analyze, as in conversation or discourse analysis.
Turn your recordings into text you can use
Upload audio or video, paste a YouTube link, or record in the browser — and get a speaker-labeled transcript in minutes, in 120+ languages.
Try BlaBlaScribe for freeNo credit card required · Stored in the EU · Never used for AI training





