ACADEMIC WRITING GUIDE

How to Transcribe and Anonymize Research Interviews for a Dissertation (With Worked Examples)

By Acadelyra Editorial TeamPublished

An interview recording is not yet an analysis-ready research transcript. It may contain missed words, unclear speakers, interruptions, references to identifiable people and places, and transcription choices that change how a participant's account is understood. Turning that recording into a trustworthy dissertation dataset requires more than pressing an automated-transcription button.

This guide explains how to choose a transcription style, produce and check a transcript, label speech consistently, protect participant identities, and describe the process accurately in your methodology chapter. You'll see worked before-and-after examples, a sample transcript extract, and a practical quality-control log. Every dialogue excerpt and identifier in the examples is invented for teaching; none is a real participant's information.

What is interview transcription in research?

Interview transcription converts spoken interaction into a written or otherwise analysable record. The transcript represents the recording according to a stated convention: which words and features are retained, which are omitted, how speakers are identified, and how uncertain audio is marked. It is a representation of the interaction, not a perfect replacement for the recording.

For a study exploring students' experiences of academic feedback, the transcript might preserve the interviewer's question, a participant's pauses before answering, the actual words used, and a note that part of the audio was unclear. Researchers must decide which details are meaningful for their questions and analytic approach rather than assuming every transcript needs the same level of detail.

Transcription, anonymisation and analysis are different tasks. Transcription makes speech available in text; de-identification or anonymisation addresses the risk that someone could be recognised; analysis develops interpretations from the resulting evidence. Our guide to thematic analysis from interview transcripts covers the subsequent interpretive stage, not the full preparation process addressed here.

Decide what kind of transcript your study needs

Different transcription conventions preserve different information. Choose yours based on the research question and the analytic method; describe it consistently rather than calling every transcript “verbatim.”

Interview transcription approaches and cautions
ApproachTypical treatmentAppropriate use and caution
Detailed or conversation-focused transcriptionMay retain pauses, interruptions, overlap, laughter and features of delivery using a defined notation systemUseful when interaction or speech patterns are part of the research question; conventions must be explained and applied reliably
Verbatim transcriptionUsually records the participant's actual words, including repetitions or fillers to the degree specifiedUseful for many interview studies; “verbatim” is not a universal, self-explanatory standard
Cleaned or edited verbatimRemoves some non-meaningful repetitions or fillers while retaining intended meaningCan improve readability, but editing choices must be transparent and should not alter meaning
Selective transcriptionTranscribes only designated sections of a recordingRequires a justified selection rule and careful acknowledgement of what was not transcribed

Suppose a participant says, “I—I got feedback, um, late, so I couldn't really, you know, use it.” A detailed transcript might keep the repetitions and fillers. A cleaned version could read, “I got feedback late, so I couldn't really use it.” But an edit that turns the statement into “Feedback was never useful” changes the claim; it is not simple cleanup.

A participant's dialect, non-standard grammar or way of speaking must not be silently rewritten into a different argument. Where transcription decisions affect interpretation, retain the relevant audio features or explain your practice. If working across languages, distinguish original-language transcription from translation and document who translated, how meaning was checked, and which wording is quoted.

Step 1: Establish permissions and a safe workflow

Before transferring a recording to any device, contractor, transcription platform or cloud service, verify what the participant consent, ethics approval, institutional policies and any data-sharing agreements permit. A tool's convenience does not grant permission to upload confidential research audio.

A workable sequence is to create a restricted-access recording, assign a study identifier (such as P07), keep any identity-to-code key separately, and make an analysis copy with access limited to authorised people. Agree in advance on storage location, encrypted transfer where appropriate, retention, backup and deletion requirements.

If external transcribers are permitted, clarify confidentiality arrangements, processing location, onward sharing, deletion requirements and the possibility that the service retains or uses recordings. For research involving sensitive interviews, seek specific university or ethics-team guidance before using external or AI services. Never paste raw identifiable transcripts into a public or unapproved AI tool.

Planning these responsibilities belongs in the study's research data management plan. The University of Bath also offers practical guidance on transcription arrangements, including external services.

Step 2: Prepare a transcript template and naming convention

Choose a consistent structure before transcribing your first recording. A useful template records the interview ID, date or generalised collection period where permissible, transcriber or reviewer role, language, transcript version, convention and any quality notes. Avoid putting a participant's real name in the filename.

Illustrative header (not real data):

Illustrative interview transcript header — invented data
FieldExample
Study interview IDP07
FilestudyA_P07_transcript_v02.docx
Interview formatIndividual semi-structured interview
Recording duration24 minutes, 10 seconds
Transcription conventionVerbatim, with meaningful pauses and laughter marked
Reviewed against audioYes — by authorised researcher
Identity keyStored separately; not included in transcript

Use stable speaker labels such as INT: for interviewer and P07: for participant. Where multiple participants speak, ensure every voice is labelled accurately. Decide how you will represent silence, overlaps, inaudible speech, emphasis or translator notes. Do not present a notation as standard unless you are actually applying a named convention.

Step 3: Produce the first transcript without inventing words

Listen in manageable segments. Record what you can hear and mark what you cannot establish. For instance, use [inaudible, 04:12] rather than guessing a word because it seems to fit the story. Repeatedly listening may resolve uncertainty, but if it does not, preserve the uncertainty.

Automated transcription can help create a first draft only when its use is authorised. It may mishear accents, names, technical terms, negations, overlapping speech and quiet sections. It may also generate fluent-looking wording that a participant did not say. Treat automated text as unverified until it has been checked against the original recording.

For ambiguous audio, note the location and revisit the source. Where the recording remains unclear, acknowledge that limitation rather than manufacturing clarity. Keep a record of any substantial editorial intervention.

Step 4: Check the transcript systematically against the recording

A responsible quality check goes beyond using spelling correction. Listen while reading, confirm the speaker labels and check that negation, quantities and relevant expressions have not been changed. Mistaking “I didn't understand” for “I understood” reverses the meaning of the account.

A practical two-pass method is:

  1. Accuracy pass: Compare text with audio in sequence, correcting omitted turns, misstated words and speaker attribution. Flag uncertain passages and check time stamps.
  2. Consistency and privacy pass: Apply the chosen notation and pseudonymisation rules consistently, identify direct and indirect identifiers, and confirm any editing did not change the intended meaning.

Another authorised researcher may review a selected portion when justified by the design and resources. Do not claim every transcript was independently double-checked if that did not happen. Record the actual checks performed.

Worked example 1: Repair an inaccurate draft

Hypothetical audio:

INT: What happened when the feedback came?

P07: I didn't open it until Monday. I was worried about the grade.

Unverified automated draft:

P07: I did open it on Monday. I wasn't worried about the grade.

Corrected transcript after checking the invented audio scenario:

INT: What happened when the feedback came?

P07: I didn't open it until Monday. I was worried about the grade.

The unverified draft introduces two meaning-changing errors: it converts a delay into an action and reverses the participant's emotional response. Neither spelling software nor fluent phrasing would necessarily flag the errors. The correction depends on listening to the source, not guessing from context.

Step 5: Distinguish pseudonymisation from anonymisation

Removing names helps, but a transcript can remain identifiable. A participant might mention being the only specialist in a small department, a precise date of a rare event or a distinctive combination of family and workplace circumstances. Other people may recognise them even without a name. The UK Data Service calls attention to these contextual and indirect identification risks.

Pseudonymisation replaces direct identifiers with codes or pseudonyms while additional information, such as a separately held key, can still connect the data to people. Anonymisation, when genuinely achieved, means people are not reasonably identifiable considering relevant information and context. Do not describe a coded transcript as fully anonymous solely because it uses P07.

Consider replacing a specific institution with [university], a named employer with [employer] or a rare location with [region] when those details are not necessary to the analysis. Preserve analytically meaningful context wherever safe. Some research requires controlled access rather than publishing a heavily redacted transcript. Anonymisation is not always achievable, and no single checklist guarantees it.

Worked example 2: Protect identity while keeping meaning

Illustrative raw transcript — invented and not based on a person:

P07: I'm the only night-shift laboratory technician at Northbridge College. When Dr. Marina Ellis emailed our four-person team on 14 February, everyone knew the complaint was mine.

Proposed de-identified version:

P07: I'm the only [specialist role] on a particular shift at [institution]. When a [senior colleague] emailed our small team, everyone knew the complaint was mine.

Residual-risk note: The unusual role and small team might still make the speaker recognisable to insiders. Before sharing or quoting, reconsider whether “only” and the shift detail must remain, whether further generalisation would change the substantive meaning, and whether restricted rather than public access is appropriate.

The goal is not to wipe out all context indiscriminately. The fact that the participant felt identifiable to colleagues may be central to the research question. A privacy review should balance that analytic value against disclosure risk and institutional requirements.

Keep any secure replacement log or code key separate from files meant for wider circulation, with access restricted according to the approved data plan. Check quotes again before including them in a thesis, presentation or repository deposit.

Step 6: Keep a transparent transcript-change log

A modest audit log helps show that decisions were reasoned rather than silently made to strengthen a preferred interpretation.

Hypothetical transcript quality-control log
ItemObservationActionReason
P07, 03:14Automated draft omitted “didn't”Corrected against audioMeaning-changing recognition error
P07, 08:36Named institution and lecturerReplaced with descriptive bracketsDirect identifiers not needed for the research question
P08, 11:02Unclear phrase despite reviewMarked [inaudible, 11:02]Avoid inventing speech
P09, 17:45Overlapping voicesAdded overlap marker consistent with the project conventionPreserve interaction detail relevant to interpretation

This is a hypothetical log. Your real log should record only events that actually occurred. Avoid storing a disclosive log inside an otherwise de-identified dataset.

Worked example 3: A dissertation-ready transcript extract

The following fictional extract demonstrates a readable and documented convention. It should not be treated as actual interview evidence.

Study ID: P12. Convention: Verbatim with meaningful pauses and laughter marked; direct identifiers replaced where appropriate.

[06:18] INT: Can you describe when you first tried to use the feedback?

[06:23] P12: I opened the comments after my next class. [pause] Some were clear, but one said to “develop the argument” and I didn't know what that meant.

[06:36] INT: What did you do next?

[06:39] P12: I asked [tutor] for an example. They explained how to link the evidence back to the question, and that was easier to follow.

[06:52] INT: Was there anything that made asking for help difficult?

[06:56] P12: A bit. I wasn't sure if it was a silly question. [laughs]

What this extract demonstrates: Consistent speaker turns, optional time stamps, restrained notation and a replacement label that protects a third person's name while retaining the role relevant to the account. The excerpt does not establish that tutoring improves grades or that all students respond similarly. Those claims would require appropriate evidence and analysis.

If you include an extract in an appendix, confirm that doing so is permitted by the consent agreement, institutional policy and ethics approval. A quote can carry identification risk even if the complete transcript is not shared.

Step 7: Describe transcription and confidentiality in your methodology chapter

A reader should know whether recordings were transcribed manually, automatically or through an approved service, the degree of detail retained, how accuracy was checked, how identifiers were handled, and how access was controlled. Keep proposals in future tense; report completed activities only when they happened.

Worked methodology paragraph — a hypothetical completed study:

“With the participants' consent and the required institutional permissions, interviews were audio-recorded and transcribed using a verbatim convention that retained participant wording and marked analytically relevant pauses. The researcher checked each draft transcript against the corresponding recording, corrected recognition and speaker-attribution errors, and flagged passages that remained inaudible. Direct identifiers were replaced with descriptive labels, while contextual details that could enable indirect identification were reviewed before quotations were selected. Recordings and the separately stored identity key were accessible only to authorised researchers under the approved retention plan. The de-identified working transcripts were then prepared for qualitative analysis.”

This passage is a model, not a statement about your study. Use it only after replacing it with your actual procedures. If checking was limited to selected segments, say so. If you used an external transcription provider, state the arrangements that are relevant and allowed to be disclosed. Do not label transcripts anonymous unless the risk assessment supports that conclusion.

The transcription paragraph should align with your full data collection procedure and ethical considerations, rather than contradicting them.

Step 8: Prepare transcripts for analysis or appendices

Before analysis, check that transcript IDs match the dataset register, file versions are clear, speaker labels are consistent and any redactions are documented. Keep the original source securely if permitted and distinguish it from the edited working copy. Confirm that an analysis program, supervisor or collaborator is authorised to receive the data.

A full transcript does not automatically belong in an appendix. Many universities prefer a sample extract or require restricted submission, especially for sensitive interviews. Our guide on dissertation appendices explains how to label supplementary material and cross-reference it from the main text. Follow your institution's instructions about what can be submitted or shared.

Common mistakes to avoid

Treating automated transcription as final data. Plausible-looking text can misstate negation, names or speakers. Check it against the source.

Changing participant meaning to improve grammar. Readability edits should not create an interpretation the speaker did not express.

Calling pseudonymised material fully anonymous. Codes alone do not eliminate re-identification through context.

Uploading identifiable audio to an unapproved service. Verify the actual consent and institutional permissions first.

Sharing a transcript but forgetting its metadata. Filenames, document properties, notes or associated logs may contain identifiers.

Over-redacting material without considering research value. Remove or generalise the minimum necessary under the approved risk approach; restricted access may be preferable for some datasets.

Claiming a more rigorous workflow than actually occurred. Report what was checked, by whom and at what stage, without inventing independent verification.

Final checklist: is your interview transcript ready?

  • The transcription convention fits the research question and analysis.
  • Consent, ethics and tool/service permissions have been confirmed.
  • Recording and working files are stored and accessed appropriately.
  • Every speaker has a stable, consistent identifier.
  • Meaning-changing words, pauses and unclear passages were checked as needed.
  • Unintelligible sections are marked rather than guessed.
  • Direct identifiers were handled consistently.
  • Contextual and third-party identification risks were considered.
  • Any remaining identification risk and sharing restrictions are documented.
  • Transcript versions, decisions and relevant quality checks are recorded.
  • Dissertation quotations and appendices were reviewed for disclosure risk.
  • The methodology chapter describes the procedures actually followed.

Frequently asked questions

Must dissertation interviews be transcribed word for word?

Not always. The required detail depends on the research question, analytic approach, programme rules and ethical constraints. Explain what you mean by “verbatim” or “cleaned” and avoid edits that distort meaning.

Can I use AI to transcribe research interviews?

Only when the approved research protocol, participant information and institutional data-protection requirements permit the specific service and data transfer. Even then, human checking remains necessary. An AI output should not be assumed accurate or confidential merely because the tool is widely used.

Should I remove filler words and pauses?

That depends on whether those features matter to your analysis. Removing every hesitation may erase meaningful uncertainty; preserving every sound may be unnecessary for another design. Set a consistent convention and explain it.

Can I include an entire interview transcript in my appendix?

Only if your institution requires or permits it and participant consent, confidentiality obligations and disclosure-risk assessment support that use. Often a carefully reviewed excerpt is more appropriate than publishing an entire account.

What is the difference between pseudonymisation and anonymisation?

Pseudonymisation replaces identifiers with codes while re-identification remains possible using additional information. Anonymisation requires assessing whether a person remains reasonably identifiable through the text or related information. A code label by itself does not make a transcript anonymous.

Final takeaway

A defensible dissertation transcript is accurate enough for its intended analysis, transparent about its conventions and limitations, and handled in a way that respects participants and third parties. Produce the text carefully, compare it with the recording, document meaningful corrections, and address identification risks before circulating extracts. Good transcription protects both the quality of the evidence and the people who supplied it.

If you would like help reviewing the clarity, structure or methodological reporting of your own research work, explore Acadelyra's academic-support services. Keep control of your data, obtain necessary approvals and follow your university's academic-integrity requirements.

Authoritative resources and further reading