How do you prepare interview transcripts for qualitative analysis?

Prepare each transcript in four steps: make it an editable file, check it against the audio, label speakers and timestamps consistently, and anonymise it before coding. Keep the recording and the first draft untouched, and do every change on a copy. Coding software can only work with text you can edit and trust, so the preparation decides how much of your coding you can rely on.

Make the transcript an editable file before you import it

Qualitative analysis software reads the transcript you give it, so choose a format it can edit. NVivo 15 for Windows says that documents imported from Microsoft Word or text files can be edited inside NVivo, for example by applying paragraph styles and formatting text, and its help page lists interview transcripts among those documents ¹.

PDFs are the wrong working format. NVivo 14 for Windows says that PDF files are created by importing PDF files, and that you cannot edit PDFs in NVivo ². Keep your working transcripts as Word or plain text files, and export a PDF only for reading. Check the help for the version you use, because the formats differ between releases.

Check every transcript against the audio before you code it

Transcription is part of the analysis, not a step before it. Davidson's 2009 review of the literature opens by calling transcription a practice central to qualitative research, although the literature often treats it as taken for granted ³. If the transcript is wrong, every code built on it is wrong too.

Wollin-Giering and colleagues recommend that you review the text of any transcript to ensure accuracy, whichever tool made it ⁴. Play the audio at normal speed while you read, and fix names, numbers and negatives first. Mark unclear words in square brackets rather than guessing.

Speaker labels deserve the same care. OpenAI's speech-to-text guide says its speaker-identification model should be used only when you need to identify who speaks during different parts of a recording ⁵. If your tool does not label speakers, add the labels yourself during the check, using one label per person for the whole project.

A labelled turn looks like this, with invented names and times:

[00:14:32] Interviewer: How did the first year change the way you plan your work?

[00:14:40] P04: I stopped planning by the week and started planning by the client.

Read the glossary entry on timestamps for the format, and the entry on speaker diarization for why machine labels need checking.

Choose a transcription style that the analysis can use

The style of the transcript sets what you can later say about it. Oliver, Serovich and Mason describe denaturalized transcription, in which "idiosyncratic elements of speech (e.g., stutters, pauses, nonverbals, involuntary vocalizations) are removed" ⁶. That style suits content analysis, where the words matter more than the delivery. Thematic coding usually works from clean verbatim text, and our guide to verbatim and clean verbatim transcription explains each option.

Whatever you choose, record the choice in your methods notes. A reader cannot judge a coded theme without knowing what the transcript kept and what it removed.

Anonymise while you transcribe, and keep the original

The CESSDA Data Management Expert Guide advises that you plan anonymisation at the time of transcription or initial write-up ⁷. Replace names with pseudonyms, use the same pseudonym for the same person in every transcript, and keep the key that links pseudonyms to real names in a separate, secured place. Renaming someone in one file after coding has begun is how a project ends up with two different people called the same name.

Work on copies. The UK Data Service's recommendations for a 2023 anonymisation project of 38 interview transcripts say that a copy of the original file is to be made, that all changes go to the new file, and that the original is kept until the whole project is anonymised ⁸. Those are one project's recommendations, not a general rule, but the procedure works for a single interview too.

The order that avoids rework

1. Choose the format. Export each transcript as a DOCX or TXT file, and keep a PDF only for reading.
2. Check the draft. Compare it with the audio, fix errors and mark unclear words.
3. Label and timestamp. Use the same speaker labels in every file.
4. Anonymise a copy. Replace names with pseudonyms, and store the key separately.
5. Import and code. Only then import the files into your analysis software.

For the wider method, see our guide to interview transcription. If you are still deciding whether a tool is appropriate for research interviews at all, our question on using AI transcription for research interviews covers consent and data rules. For a recording that is not an interview, such as a public show, the question on how to transcribe a podcast sets out the steps.