Transcription Guidelines: Rules for Accurate, Consistent Transcripts

Priya and Tom split six interview recordings for a local news series. When the transcripts came back, Priya had written (laughs), Tom had written [laughter], and only one of them had kept the "um"s. Tom's timestamps sat on every line; Priya's appeared only at the unclear bits. (They're a composite, not real people.)

Neither of them was wrong. They'd simply never agreed on transcription guidelines before they started, and now an editor had to reconcile two styles before a single quote could be checked.

If you've searched for transcription rules, you've probably found a pile of company style guides that each sound official. They are, for that company. This guide lines up four of the published ones, shows where they agree and where they flatly contradict each other, and ends with a one-page house guideline you can paste into your team's notes.

Short on time? Skip to the table, then the template. And if you're still typing transcripts by hand, ParrotNotes gives you a first draft from your phone so these rules are all you have left to apply.

Two kinds of transcription guidelines

People who search for transcription guidelines usually want one of two things.

Job guidelines. You transcribe for a company such as Rev or GoTranscript, or you're applying. The company's style guide is the rulebook, graders score you against it, and nothing in this article overrides it. Read theirs first; this post shows you where theirs is unusual.

House guidelines. You run interviews as a reporter, podcaster or researcher, and you need your transcripts to look the same whoever types them. Nobody grades you, so the job is to make choices once and write them down.

Both groups need the same core rules. They differ on who gets to make the choices.

The core transcription guidelines every published guide agrees on

Across Rev, GoTranscript, a university captioning office and a university research guide, these rules hold everywhere:

1. Transcribe what was said, not what was meant. Rev's guide puts it bluntly: "Do not type what you think the speaker meant to say." No paraphrase, no tidying of meaning.
2. Don't correct grammar. A speaker who says "she's wrote" said "she's wrote." Clean verbatim removes fillers; it doesn't fix sentences.
3. Attribute every line. Each turn gets a speaker label, even with a single speaker (GoTranscript requires a label in that case too).
4. Tag what you can't hear, and never guess. Every guide has an inaudible tag. None lets you fill a gap with a plausible word.
5. Research names and terms before tagging them. Rev says the inaudible tag "should never be used in place of research."
6. Choose verbatim or clean verbatim before you start. Both styles are legitimate; mixing them in one file isn't. Our verbatim vs clean verbatim guide helps you pick.
7. Keep profanity as spoken. Rev and GoTranscript both say transcribe it; JMU says never censor harsh language.
8. Describe sounds, don't interpret them. Humble's guide warns against "[nervous laughing]": write [laughing] and let the reader judge.
9. Be consistent. The same sound, label and number gets the same treatment on page 1 and page 40.

For reporters there's one more rule above all of these. The Associated Press says in its News Values and Principles: "We do not alter quotations, even to correct grammatical errors or word usage." Any transcript you quote from has to be checked against the audio, whatever style it's in.

Where the transcription guidelines disagree: four guides compared

Here's the part nobody's style guide tells you: the published guidelines contradict each other on basic conventions. The table below compares four of the guides that rank for this search, as each read on October 24, 2026: Rev's Transcription Style Guide v5 (for Rev freelancers), GoTranscript's transcription guidelines (for its transcribers), James Madison University's transcription guidelines (for captions) and Áine Humble's Guide to Transcribing (Mount Saint Vincent University, for research interviews).

Three disagreements matter most in practice:

• Brackets or parentheses. Rev uses (laughs); GoTranscript, JMU and Humble use square brackets. Pick one. Square brackets are the safer house choice because parentheses also appear in normal speech punctuation.
• [sic] or no [sic]. Humble's research guide uses it to show an error was the speaker's; GoTranscript bans it. In a transcript (not a published quote), leaving the error untagged is cleaner, as long as your guideline says errors are kept as spoken.
• Contractions in clean verbatim. Rev allows expanding "gonna" and GoTranscript requires it; our verbatim style sheet keeps it as spoken. For a reporter checking quotes, keeping the speaker's words is the safer default.

A one-page house transcription guideline you can copy

This template makes every choice from the table for you, tuned for reporters and qualitative researchers. Change any line you disagree with; the point is that the choice is written down before the first file.

HOUSE TRANSCRIPTION GUIDELINES  (version: ____  owner: ____)

1. Style: clean verbatim unless the brief says verbatim.
   Clean = remove um/uh, stutters and abandoned starts.
   Never: correct grammar, reword, or expand contractions.
2. Header on every file: project, interviewee, interviewer,
   date, place, recording file name, transcriber, style.
3. Speaker labels: full name on first turn, surname after.
   Unknown: Speaker 1, Speaker 2. Unsure: ?Speaker 2.
   Never describe people ("tall guy").
4. Timestamps: [hh:mm:ss] at every speaker change.
5. Unclear audio: [inaudible hh:mm:ss]. Best guess, if any:
   [unclear: word? hh:mm:ss]. Never guess silently.
6. Overlap: [crosstalk hh:mm:ss].
7. Sounds: square brackets, lowercase, present tense,
   two words max: [laughs], [phone rings]. Only when the
   sound matters to meaning. Never interpret ([nervous]).
8. Pauses: [pause] only for 5+ seconds that matter.
9. Numbers: zero to nine in words, 10+ in numerals.
   Always numerals: money, percentages, ages, times, dates.
10. False starts kept in verbatim: "--". Interruptions: "-".
11. No [sic]. Errors stay as spoken.
12. Names and terms: research first; log spellings in the
    glossary below. Unverified: [spelling?].
13. Quotes: nothing goes inside quotation marks in print
    until checked against the audio at its timestamp.
14. Research only: replace identifying names with
    pseudonyms or (general term) and log every change.

GLOSSARY: names, places, organisations, jargon

Researchers will want to expand line 14; our qualitative interview transcription guide covers anonymizing in detail. For the full file layout around these rules, see the interview transcription example and template.

Worked example: one passage under the house rules

Here's an invented exchange, first as a raw draft might come back, then formatted under the house guideline above (clean verbatim).

Raw draft:

so um the council the council voted what was it seven to two
on the the bridge thing and uh honestly [laughter] nobody expected
that. marta kowalsky she was the one who pushed it

Formatted:

[00:14:32] Interviewer: How did the vote go?

[00:14:35] Dana Reyes: The council voted, what was it, seven to
two on the bridge thing, and honestly [laughs] nobody expected
that. Marta Kowalsky [spelling?], she was the one who pushed it.

What changed, rule by rule: the fillers and the repeated "the council" and "the" came out (rule 1); "what was it" and "she was" stayed, because they're words the speaker chose, not stumbles (rule 1); "seven to two" stayed in words because both numbers are under 10 (rule 9); "bridge thing" stayed, because clean verbatim doesn't upgrade a speaker's words; [laughter] became [laughs] to match the house tag (rule 7); the name got capitals and a [spelling?] flag until someone checks it against the council's website (rule 12). The vote count would then be checked against the audio before it's quoted (rule 13).

Applying transcription guidelines to an AI-drafted transcript

Most transcripts now start as an AI draft, and that changes where your guideline earns its keep. The draft gets the words down; your rules decide labels, tags, timestamps and every judgment call.

Take Leila, a freelance reporter (an illustration, not a real person). She records a 50-minute interview with a school superintendent on her phone. She gets a draft transcript minutes after hanging up, and spends her time on the steps a machine can't own: checking the three quotes she'll use, flagging two unclear passages with timestamps, and confirming the spelling of a district program against its website. Her house guideline is pinned in her notes, so the cleanup takes the same shape on every story.

That's the workflow ParrotNotes is built for. You record in person or import a file, and you get a transcript with an AI summary. On Pro, speaker identification infers speakers from name cues in the conversation and full-text search finds the quote you half remember. Then you apply your guideline and export the transcript as text or markdown. For the correction pass itself, follow the steps in how to transcribe an interview properly, and use how to quote an interview before anything goes to print.

Want a draft to apply these rules to? Download ParrotNotes free and record your next interview, with your source's OK. Got an existing file? Try the free audio to text tool.

If you transcribe for Rev or GoTranscript

If the guidelines you need are a company's, the rule is simple: theirs, exactly, every time. Both companies grade the work.

• Rev scores each transcript from 1 to 5 on two categories, accuracy and formatting, and lists custom descriptive speaker labels as a mistake that earns a 1 in formatting. A file ordered as verbatim but delivered in the default clean style is graded 1/1.
• GoTranscript has editors rate transcriptions, and its guide says new members whose average is 3.6 or lower after three files may be removed.

Picture Marcus, a part-time transcriber applying to both (a composite). His habit from university work is (inaudible) in parentheses and [sic] after slips. On GoTranscript that's two errors on page one: the tag needs square brackets, bold and a timestamp, and [sic] isn't allowed. On Rev the parentheses would still be wrong for the inaudible tag. The fix isn't to learn "the" rules. It's to keep each company's guide open beside the editor and drop his own habits for the length of the job.

Write your transcription guidelines down once

Transcription guidelines aren't a universal standard. They're a set of choices, and the published guides make different ones. What every good guideline shares is the core: transcribe what was said, label every speaker, tag what you can't hear, and never guess or fix someone's grammar.

Your next steps:

1. Working for a transcription company? Use its guide, exactly.
2. Running your own interviews? Copy the house guideline above, change the lines you disagree with, and give it a version number.
3. Share it with everyone who touches your transcripts, including freelancers.
4. Check every quote against the audio before it's printed.

Then let a first draft do the typing. Download ParrotNotes free: 100 minutes of recording a month with an AI summary on every recording, so your time goes into the rules, not the typing.