Qualitative Research Transcription: From Interview Audio to Coding-Ready Text

Qualitative research transcription is the conversion of recorded interviews or focus groups into written text that you can code, quote and archive. The workable method in 2026 is an AI draft followed by a researcher's check against the audio, in a convention you choose before the first interview: intelligent verbatim, strict verbatim or Jefferson notation. The transcript is your dataset, so every decision about it (what to keep, how to label speakers, how to anonymise) belongs in your methods section.

This guide walks through those decisions in order, from recording to a file that imports cleanly into NVivo or MAXQDA.

What is transcription in qualitative research?

In most projects the transcript, not the recording, is what gets analysed. Codes are attached to its lines, quotes in the paper are copied from it, and an archive or a second coder sees only the text. That makes transcription an interpretive step rather than typing: you decide which features of talk survive onto the page, and anything you drop can no longer be coded.

Three questions shape the whole job:

• What level of detail does the analysis need? Thematic analysis works on content; conversation analysis works on timing, overlap and intonation.

• Who produces the first draft? You, a research assistant, a human transcription service or speech recognition.

• Where does the audio go? Your ethics approval and consent form have to cover every system that processes the recording.

Types of verbatim transcription in qualitative research

Researchers use "verbatim" loosely, so name the exact convention in your protocol and apply it to every file. The three common levels:

Jefferson's symbols are set out in her Glossary of transcript symbols (2004), the reference most conversation analysts cite. If your research question is about what people say rather than how the talk is organised, it is more detail than you need, and it multiplies transcription time.

A practical middle ground for many interview studies: intelligent verbatim as the base, plus a short list of features you always keep, such as laughter, long pauses and audible emotion, written as bracketed notes like [laughs] or [long pause].

How to transcribe qualitative interviews: step by step

This workflow holds up in a methods section and keeps the researcher close to the data:

1. Record for the transcript. A quiet room, the recorder between interviewer and participant, and separate microphones for remote calls if you can get them. Audio quality sets the ceiling on accuracy; our guide on why AI transcription accuracy varies covers the factors that matter most.

2. Pilot the pipeline. Run your pilot interview through the whole process and note how long the correction pass takes. That number, not a vendor's benchmark, tells you how many hours to budget for the full sample.

3. Generate a draft with speaker labels. Speaker diarization splits the text by voice. A one-to-one interview is the easiest case for it.

4. Replace generic labels with participant IDs. "INT" and "P07" rather than "Speaker 1" and "Speaker 2", in every file, so the coding software and later readers see the same identifiers.

5. Check the draft against the audio. Listen to the whole recording, not only the passages that look wrong. Pay extra attention to names, numbers, negations ("I could" versus "I couldn't") and every passage you might quote. This pass doubles as your first close reading, so keep a memo file open.

6. Mark what you cannot hear. Write [inaudible 00:14:22] or [unclear: possibly "contract"] instead of guessing. A silently guessed word is an invented datum.

7. Anonymise, then save a coding copy. Keep the verified original under restricted access and work from the pseudonymised version (see the anonymisation section below).

8. Export and import. DOCX or TXT for coding software, with timestamps at speaker turns if you want to jump back to the audio.

For the general mechanics of interview transcripts outside academia, such as journalism or hiring, see how to transcribe an interview.

Transcription conventions for qualitative research

Write a one-page conventions sheet before the first file and give it to anyone who transcribes or checks. These conventions save the most time at the coding stage:

Qualitative interview transcript example

A short excerpt in intelligent verbatim with the conventions above (invented for illustration):

[00:12:45] INT: How did the move to remote work change your week?
[00:12:49] P07: At first it was fine. Then [manager 1] started booking calls at eight, and I just [pause] stopped switching off.
[00:13:02] INT: Stopped switching off how?
[00:13:04] P07: [laughs] Laptop on the kitchen table, basically. My partner hated it. [inaudible 00:13:09] the whole spring.

Each turn starts with a timestamp and an ID, uncertainty is marked, and the only identifying detail has become a tag.

How to anonymise qualitative transcripts

The UK Data Service, which archives qualitative data for reuse, gives the most practical public guidance in its page on anonymisation for text data. Its main points, applied to interview transcripts:

• Use pseudonyms or descriptive tags such as [colleague 1] instead of deleting details, so the narrative still makes sense.

• Apply replacements consistently across all transcripts and related materials. The same person gets the same tag in every file.

• Keep an anonymisation log of what you changed, stored separately from the anonymised files.

• Do not over-anonymise. Stripping every contextual detail can make the data useless; controlled access is often a better tool than heavier editing.

Indirect identifiers are the hard part. A job title, a small town and a rare illness together can point to one person even with no name in the text. Read each transcript once purely for those combinations.

For EU and UK studies, keep the legal distinction in mind. Under GDPR Recital 26, pseudonymised data that can be re-identified with additional information is still personal data. A transcript with tags and a separate key is pseudonymised, not anonymous, and your data management plan should treat it that way. Recording consent rules differ by country; our overview of recording consent in the US and EU covers the basics.

Importing transcripts into NVivo, MAXQDA and ATLAS.ti

Coding tools read structure from plain patterns, so a consistent transcript imports with almost no cleanup.

• NVivo can import a transcript and link it to the media file. Its transcript import help (NVivo 14) accepts timestamps as hh:mm:ss or mm:ss and reads a speaker name when it is followed by a colon and a space, which is why stray colons in the text cause trouble. Accepted file types differ by version: the Mac help for NVivo 14 asks for TXT, CSV or TSV for transcript rows, while the Windows help also lists Word and RTF files, so check the page for your edition.

• MAXQDA imports DOCX and RTF transcripts and recognises a long list of timestamp formats, including [hh:mm:ss] and hh:mm:ss, according to its transcript import page. It also imports VTT files, the caption format Zoom and Teams produce.

• ATLAS.ti and similar tools take DOCX or TXT as text documents; timestamps then act as reference points you read rather than media links.

Test with one file first. If timestamps fail to sync, the cause is usually a format the tool does not recognise or timestamps out of order.

Transcribing focus groups

Focus groups are harder than interviews on every count: six to ten voices, overlapping talk and participants far from the microphone. Speaker separation gets less reliable as the number of similar voices grows, so plan for more correction time.

• Place one recorder in the centre of the table and, if possible, a second one at the other end as backup.

• Have the moderator use names in the first round ("Thanks, P03") so voices can be matched to IDs later.

• Ask a note-taker to log the first words of each turn with the clock time during heated stretches.

• Accept that some overlaps will stay as [crosstalk]. Mark them honestly rather than assigning words to the wrong speaker.

For group analysis, a misattributed turn matters more than a misheard filler, so spend the check on who said what.

Qualitative research transcription software vs services

There are three realistic routes, and many projects mix them:

Whichever tool you pick, check these points against your ethics approval before uploading a single recording:

• Where the audio is processed and stored, and whether that country is acceptable under your protocol.

• Whether files are used to train models.

• How long recordings and transcripts are kept, and whether you can delete them yourself.

• Whether a data processing agreement is available for your institution.

• Which export formats you get (DOCX and TXT for coding, timestamps at turns).

BlaBlaScribe is one option for the AI route: it transcribes uploaded recordings in 120+ languages with speaker labels, runs transcription on its own servers in Germany, does not use files to train AI, and exports DOCX, TXT, CSV and subtitle formats. It is not a meeting bot and does not join your calls, it will not produce Jefferson notation, and pseudonymising the text remains your job. The free plan gives 30 minutes once; files longer than 10 minutes show only the first 10 minutes until you buy a minute pack or plan, so a full pilot interview needs one of those. To understand what accuracy figures actually measure, see word error rate explained.

Transcribing interviews in other languages

Transcribe in the language of the interview and code in it for as long as your team can. Translating before coding replaces the participant's phrasing with the translator's, and that phrasing is often what an in vivo code hangs on.

1. Transcribe in the original language, for example with a dedicated Spanish transcription or German transcription setting.

2. Code in the original wherever a team member reads it fluently.

3. Translate selected quotes for publication late, and have a fluent speaker check them.

4. Report in the methods section who translated and at which stage.

Machine translation of the full transcript is useful for giving monolingual supervisors access, as long as everyone agrees that the original remains the dataset.

How to report transcription in your methods section

Reviewers want to know who produced the transcripts, at what level of detail, and how they were checked. A sentence like this covers it: "Interviews were audio-recorded with consent, transcribed in intelligent verbatim using automated speech recognition, and checked in full against the recordings by the first author, who also pseudonymised names and places." Add the tool name if your field expects it, and state how unclear passages were marked.

Frequently asked questions

Is AI transcription acceptable for qualitative research?

Yes, provided your ethics approval and consent form cover the tool and a researcher checks every transcript against the audio. Describe both the tool and the checking step in your methods section.

What is verbatim transcription in qualitative research?

It is a word-for-word written record of what participants said. Strict verbatim keeps fillers, repetitions and pauses, while intelligent verbatim removes fillers and false starts but keeps every content word and the speaker's own grammar.

Should I transcribe my own qualitative interviews?

Transcribing yourself gives the closest familiarity with the data but takes far longer. Checking an AI draft line by line against the audio gives much of the same immersion in less time, which is why many researchers now work that way.

How do I transcribe a focus group with many speakers?

Record with a central microphone, have the moderator use participant IDs early on, and expect to correct speaker attribution by hand. Mark overlapping talk as [crosstalk] rather than guessing who spoke.

What format does NVivo need for transcripts?

NVivo reads timestamps written as hh:mm:ss or mm:ss and treats text followed by a colon and a space as a speaker name. Accepted file types vary by version and platform, so test one transcript in your edition before importing the full set.

Is a pseudonymised transcript anonymous under GDPR?

No. If the tags can be linked back to people using a key you hold, the transcript is pseudonymised personal data under GDPR and needs the same protection as other personal data.