Transcription for Qualitative Research: Conventions and Workflow

Transcription for qualitative research means turning interviews or focus groups into text that you can analyse, cite and share safely. Whether you type it yourself or use qualitative transcription services, in practice that comes down to four decisions: which transcription style to use (verbatim, intelligent verbatim or a notation system), how to anonymise participants, how to check accuracy, and which format your analysis software imports. This guide covers how to transcribe qualitative interviews step by step and ends with a workflow for preparing an interview transcript for coding.

If you need the basics of recording and transcribing interviews for journalism or research, start with our interview transcription guide. This article goes deeper into the research-specific parts of how to transcribe an interview for qualitative research: conventions, ethics, anonymisation and coding software.

Verbatim vs intelligent verbatim: choose a style first

The style you choose should follow from your method, and it should be written down before anyone starts typing. Changing conventions halfway through a project means rechecking every earlier transcript.

The verbatim vs intelligent verbatim choice is mostly about what counts as data. If hesitation, self-correction or the way someone softens a statement matters to your analysis, intelligent verbatim removes evidence you need. If you care about themes across 30 interviews, full verbatim adds reading time without much analytic value. Either way, intelligent verbatim should never change the participant's words or grammar beyond removing fillers; it is not a paraphrase.

Transcription conventions for qualitative research

Conventions are the rules that make transcripts consistent across a project and across transcribers. There are two broad options.

Simple conventions

Most qualitative studies use a short, project-specific list. A typical set:

• Speaker labels: I: for interviewer, P07: for participant 7.
• Pauses: (pause) for a noticeable pause, or (3s) if length matters.
• Overlapping speech: mark with [overlap] or start the overlapping turn with a slash.
• Non-verbal sounds: [laughs], [sighs], [long silence].
• Unclear speech: [inaudible 00:14:32] with a timestamp, or [unclear: word?] for a best guess.
• Emphasis: CAPITALS or italics, used sparingly.
• Redactions: [name of town], [employer].

Put this list in your project documentation and in each transcript header.

Jefferson notation basics

Conversation analysts use the Jefferson system, named after Gail Jefferson. Loughborough University's conversation analysis notation tutorial summarises the core symbols:

• (.) a just-noticeable micropause; (0.5) a timed pause in seconds.
• [ aligned on adjacent lines for the start of overlapping talk, sometimes ] for its end.
• = latching: no gap between turns.
• wo:rd stretched sound; °word° quiet speech; capitals for loud speech.
• >word< faster talk; slower talk.
• ((sniffs)) transcriber's description of a sound or action.
• ( ) unclear talk; (word) a probable but uncertain hearing.

For the full system, methods texts cite Jefferson's "Glossary of transcript symbols with an introduction" (2004) and Atkinson and Heritage's Structures of Social Interaction (1984). Jefferson transcription is slow and specialised; no automatic tool produces it reliably, so plan for detailed human work if your method needs it.

Ethics approval, consent and GDPR

Before any recording reaches a transcription service, check three documents:

1. Your ethics approval. It may name how audio is transcribed and by whom. Using a third-party service, human or AI, may need to be stated or approved.
2. The participant consent form. Participants should know that recordings will be transcribed, who will have access, how long audio is kept and whether anonymised transcripts may be archived or shared.
3. Your data management plan. It should state where audio and transcripts are stored and when they are deleted.

In the EU and UK, interview recordings are personal data under GDPR, and they often contain special category data such as health, religion or political views. When you use an external service, it acts as a processor, so you need to know where it processes data, how long it keeps it, whether it uses recordings for model training, and whether it signs a data processing agreement. Our article on GDPR and transcription goes through the checks in detail.

Audiotexta, for example, supports scheduled deletion and does not use user content for model training; the Team plan includes a DPA. The privacy page lists the details, which you can quote in an ethics application. Whatever tool you use, check it against your institution's rules; some universities require approved suppliers only.

Anonymisation and pseudonymisation

Anonymise during or straight after transcription, not at the end of the project. The UK Data Service guidance on anonymising qualitative data recommends planning anonymisation early, applying it consistently, and combining it with consent and access controls rather than relying on redaction alone.

Practical rules:

• Direct identifiers (names, addresses, emails, phone numbers) are replaced with pseudonyms or bracketed descriptions: "Maria" becomes "Ana" or [colleague's name].
• Indirect identifiers (exact age, rare job titles, small towns, unusual events) can identify someone in combination. Generalise them: "a nurse in her forties in a small town in the north".
• Be consistent. The same person or place gets the same replacement in every transcript.
• Keep a key. Record every replacement in a log stored separately from the transcripts, with restricted access.
• Avoid blanking out text. Replacements keep the transcript readable; black bars do not.
• Do not over-anonymise. Removing too much context can make data useless for analysis.

Remember that pseudonymised data is still personal data under GDPR while a key exists. Only data that can no longer be linked back to a person counts as anonymous.

Accuracy checking: AI draft plus a human pass

Manual transcription is slow. A commonly quoted range is four to six hours of work per hour of audio for a simple verbatim transcript, and much more for detailed notation or noisy focus groups. Automatic transcription changes the workflow: the machine produces a draft in minutes, and the researcher's time goes into checking.

A workable checking routine:

1. Generate the draft with speaker labels and timestamps. Whisper-class models are good on clear audio but still mishear names, jargon, accents and crosstalk.
2. Listen to the whole recording while reading. An editor that highlights words during playback and lets you click a word to jump to it makes this much faster.
3. Fix speaker turns first, then words, then add your conventions (pauses, overlaps, non-verbal notes).
4. Mark uncertainty rather than guessing: [unclear 00:21:05].
5. Spot-check a second time on a sample, or ask a colleague to check one transcript per batch.

The checking pass usually takes at least as long as the recording itself, and often longer, so if you are planning how to transcribe interviews for a dissertation, budget for it in your timeline. It is also valuable familiarisation with your data, which many qualitative methods treat as the first stage of analysis. Our transcription accuracy tips explain how recording conditions affect the draft.

A transcript header template

Start every transcript with the same header so files stay identifiable without revealing who the participant is.

How to transcribe an interview for coding: format and export

The interview transcript format that works best for coding is simple: one paragraph per speaker turn, a speaker label at the start, and a timestamp at the start of each turn or at regular intervals.

• NVivo imports transcripts in DOCX or TXT on Windows, with timestamps at the start of paragraphs or in a table column, and accepts formats such as hh:mm:ss. NVivo for Mac imports transcripts as plain text, so check your version's help page.
• ATLAS.ti imports DOCX, RTF and TXT transcripts and, in current versions, SRT and VTT files, and can synchronise the transcript with the audio for playback.
• MAXQDA imports transcripts with timestamps from DOCX or RTF files and also supports SRT and VTT, with a list of timestamp formats from common transcription tools.

With Audiotexta, you can export TXT, SRT or VTT on the free plan and DOCX on Pro. SRT or VTT is often the quickest route into ATLAS.ti or MAXQDA because the timestamps come with the file. For NVivo, export DOCX or TXT with timestamps and check that they match a format NVivo recognises before importing a whole batch. Start in the web app with one interview and test the import before transcribing the rest.

Once the transcript is imported, linked to the audio and anonymised, you can begin coding with confidence that every quote can be traced back to a moment in the recording.