AI Skills

Organize Interview Recordings to Find Any Quote

Arpit TripathiArpit TripathiLinkedIn·November 8, 2026·11 min read

Name files before the first interview, log each quote with a timestamp, store recordings by your ethics rules, and verify quotes against the audio.

A reliable way to organize interview recordings is to settle three habits before the first interview: one file-name pattern that holds no real names, a quote log that points to a timestamp, and a storage and deletion plan that follows your consent and ethics rules. A fourth habit comes last and is not optional: check every quote against the audio before it is published.

Picture a quote that is still on disk but cannot be found. It sits in a file called "Interview final 2", in a folder of forty similar names, inside a transcript nobody has checked, with no note of who agreed to be quoted. The steps below need only a spreadsheet and a folder, and they get faster when a tool does part of the capturing.

Set the Rules Before the First Recording

Consent and ethics rules decide what you may record, keep and share, so settle them before any naming or tooling choice. CESSDA's Data Management Expert Guide notes that in some countries informed consent is mandated by law. Its country table says the UK has no legislative requirement for consent to be sought from participants, although many funders and research ethics committees require it. Requirements differ by country, funder and ethics board, so check the ones that apply to you.

The guide also recommends granular consent: its sample form asks separately about archiving the non-anonymised audio recording and the anonymised transcript. It adds that verbal consent discussions can be audio-recorded if participants agree.

Recording law is a separate question. The Reporters Committee for Freedom of the Press publishes a guide summarising each US state's laws on recording phone calls and in-person conversations. It warns that many of these laws carry criminal penalties and some allow civil lawsuits, and says the guide is no substitute for a lawyer's advice. Outside the United States, ask your newsroom, institution or ethics board what applies.

Institutions can restrict tools as well as practice. The University of Liverpool's research data guidance says researchers whose work requires transcription tools must use one of its University-managed services (Microsoft Teams, Microsoft Office or Microsoft Co-Pilot). Whether a phone app or a vendor service is acceptable is a compliance question for your institution, not a matter of taste.

Insight

Record the consent status of every interview in the quote log from day one. A quote with no recorded consent is a quote to resolve before anyone uses it.

Name Every File Before the First Interview

Use one pattern, decided before fieldwork starts: date, interviewee code, short topic, file type and version, joined by hyphens or underscores. Liverpool's guidance says file names should be short but meaningful. CESSDA's guide says to create meaningful but brief names, to write dates as YYYY-MM-DD, and, for qualitative data, to give every file from one collection event the same event identifier at the start of the name so the files can be linked. Its interview example begins with a date.

A pattern that works

  • Date first, as YYYY-MM-DD, so files sort in the order the interviews happened.
  • Interviewee code next, such as P07 or J-014, never a real name.
  • Two or three words of topic, such as housing-budget.
  • A file-type word, such as audio, transcript or notes, so a recording and its transcript share every other field.
  • A version number on anything you edit, such as v01 and v02.

Put together, the pattern gives "2026-11-08_P07_housing-budget_audio.m4a" and "2026-11-08_P07_housing-budget_transcript-v02.docx".

What stays out of the file name

  • Real names. File names show up in folder listings, attachments and sync services, so use a code.
  • Spaces and special characters. CESSDA's guide says to avoid spaces, dots and characters such as & or ? or !.
  • Vague words. The UK Data Service says to avoid terms like final, latest or new.
  • Facts that live only in the name. CESSDA's guide says background information must never be stored in the file name only.

A code needs a key, and the key needs its own home. Erasmus University Rotterdam's guidance on anonymising interview transcripts says to keep a pseudonymisation key (an anonymisation log) of all replacements, stored securely and separately from the anonymised files. The same logic suits interviewee codes: the sheet that links P07 to a real person belongs in a restricted location, not beside the audio.

Folders need the same discipline. The UK Data Service says a logical structure reduces confusion between raw and processed data, and that structures should be simple, predictable and documented, with a short README at the top level. CESSDA suggests keeping folders three or four levels deep. A workable start is a 00-readme file plus four folders: 01-raw-audio, 02-transcripts, 03-anonymised and 04-quote-log.

Keep a Quote Log That Points to the Audio

A quote log is a spreadsheet with one row per usable quote, and its job is to take you from a sentence in a draft back to a moment in the recording. Folders find files. The log narrows the hunt to a sentence. Give each row these columns:

  • File: the exact audio file name.
  • Timestamp: start and end of the quote, read from the audio player.
  • Speaker code: P07, not a name. If a transcript labels speakers generically, swap in codes at once.
  • Topic: one to three keywords from a short fixed list, so searches stay consistent.
  • Verbatim or paraphrase: mark which it is. Only verbatim words belong inside quotation marks in a draft.
  • Consent status: what the person agreed to (on the record, background only, anonymised, not for quotation) and whether it was written or spoken on tape.
  • Verified: date and initials of whoever checked the quote against the audio.
  • Where used: the story, chapter or paper that carries the quote.

One sample row reads: 2026-11-08_P07_housing-budget_audio.m4a, 00:41:12 to 00:41:38, P07, rent-rises, verbatim, on the record (spoken consent on tape), not yet verified, unused.

Fill the log while reading the transcript for the first time, when speaker, topic and consent details are fresh.

Anonymise, Move, Then Decide What to Delete

Anonymise early, move recordings off collection tools into approved storage, and delete only when your ethics approval and data plan say you may. The University of Liverpool's guidance says to anonymise transcripts as soon as possible after interviews, and to move recordings you need to keep into university-managed storage such as M:Drive, OneDrive or ADS.

Anonymise early, and keep the original apart

Erasmus University Rotterdam's guidance says anonymisation is best planned early, and that, generally speaking, direct identifiers and strong indirect identifiers need to be removed or replaced with pseudonyms. For interview transcripts it advises pseudonyms or generic descriptors rather than blanking out text, a log of every replacement, and the original unedited files stored separately from the anonymised copies. It also warns that excessive removal can make data unusable, unreliable or misleading.

Reporters who quote people by name do not anonymise, so their equivalent habit is the consent column plus a restricted folder for anything given in confidence, with newsroom protocols deciding the details.

Move recordings off collection tools

Liverpool's page says that when survey, interview or other tools are used to collect data, the data should move to a more suitable storage service that is secure and backed up as soon as possible. Its list of guidance for handling recordings says: "Do not store recordings on collection tools." By the same logic, a phone recorder or a meeting platform is a place to capture rather than to keep. Copy the file to approved storage, confirm the copy plays from start to end, and only then clear the device if your rules allow. The page also says Teams recordings are automatically deleted after 60 days. That is guidance for Liverpool researchers, and institutions differ, so look up the number that applies to you.

Retention: delete only when the plan allows

Liverpool's guidance says that when transcripts are enough and the recordings are not needed, the recordings should be deleted as soon as possible after the transcriptions have been made and quality checked. That is one institution's advice. Your ethics approval, data management plan, funder and consent forms set the actual rule, and some projects archive audio where participants have consented. Ask before deleting anything.

Data protection regulators add a second layer. The UK Information Commissioner's Office says the storage limitation principle requires personal data to be kept in identifiable form for no longer than is necessary, that the UK GDPR does not set fixed retention periods, and that data should not be kept indefinitely just in case. It lists the need to defend possible legal claims among the factors to weigh, and allows longer keeping if the only purpose is archiving in the public interest, scientific or historical research, or statistics, with appropriate safeguards. The ICO's page also carries a notice that this guidance is under review because of the Data (Use and Access) Act, so check its current wording. Other countries have different laws.

Check Every Quote Against the Audio Before It Is Published

Play the recording at the logged timestamp, compare it with the quote word for word, and only then mark the row verified. A transcript is a working copy of the audio. The Society of Professional Journalists' code of ethics, in its Seek Truth and Report It section, tells journalists to verify information before releasing it and to use original sources whenever possible. For what a person said, the original source is the recording.

Automatic transcripts make the check more important, not less. A study presented at the ACM FAccT conference in June 2024, with a Cornell researcher as lead author, ran more than 13,000 short audio segments from AphasiaBank, a repository of recordings from people with aphasia and a control group of people without aphasia, through OpenAI's Whisper speech-to-text API in April and May 2023. Roughly 1% of audio transcriptions contained entire hallucinated phrases or sentences that exist nowhere in the audio, and 38% of the hallucinations included explicit harms such as perpetuating violence.

That is one tool, one dataset and one point in time. The authors' own retest in December 2023 found that many of the hallucinated examples had been resolved, so the figures are a snapshot, not a measure of the tool you use today. A fluent, confident sentence in a transcript can still contain words nobody said, so any transcript, typed or automatic, needs a listen before a quote leaves your desk.

A five-step check

  • Open the log row and play from ten seconds before the timestamp.
  • Compare word for word. Names, numbers, negatives such as "not" and "never", and qualifiers such as "sometimes" are where meaning flips.
  • Listen to the question before and the sentence after, to confirm the quote means what the draft says.
  • If you trim a quote, show the trim with an ellipsis or brackets, and log the edit.
  • Mark the row verified with a date, and recheck any quote you edit later.
Pro Tip

Verify before you delete. Once the audio is gone, a transcript can never be checked against it again, so make the Verified column the last gate before any deletion your plan allows.

Four Ways to Find a Quote Later

No single method does everything. Folders plus a log need no new tool, while search tools can shorten the hunt through a long recording, provided your institution or newsroom accepts them.

MethodStrengthWeak spotBest for
Folders and file names onlyFree, works offline, no new tool to approveFinds a file, not a sentence inside itA handful of short interviews
Spreadsheet quote logHolds timestamp, speaker, consent and verification in one placeManual to fill, and it only finds quotes someone loggedAny project that will print or cite quotes
Transcription service with searchKeyword search across transcripts, with answers linking to timestamps in the source, as Rev describes itMay need institutional approval, plus a check of the vendor's own security termsLarge archives where the service is permitted
Phone recorder with a searchable memory (MemX Record)Transcript and notes made from a recording in the phone app, then asked for in plain words with the source recording shownTranscription and search need a connection; storage approval stays with youInterviews you are allowed to record on a phone

Rev's archive page describes one central repository organised by story, searchable by keyword. It gives no file-naming guidance and does not discuss security, so a vendor tool sits on top of the habits above rather than replacing them.

Where MemX Fits When You Record Interviews

MemX helps with the finding step, and this section covers only recordings made in its recorder. It does not replace the consent check, the storage plan or the listen-before-publishing step.

MemX Record is a recorder inside the phone app. Tap Record, and MemX writes a live transcript in the language spoken, then produces timestamped notes and a summary, searchable once the recording has been processed. Each transcript is indexed as a searchable memory, so a question such as "what exactly did the mayor say about the budget?" returns an answer and shows the recording it came from. Open that recording and play the moment before using any words from it.

From a recording you can share the transcript, the notes or the original audio file, and any original file or its extracted text can be saved to your device from its share sheet. A copy you can export is what makes the storage steps above possible.

Insight

Limits to know: transcription, search and answers need a connection, although uploads queue offline. MemX is a phone app, so capture and search happen on the phone. The Free plan lists 15 minutes of recording a month, shorter than one hour-long interview, and paid plans list more, so check the pricing page before planning a long project. MemX does not provide legal protection or guarantee source anonymity. It is private by architecture, with a key per account, encryption at rest and every decryption audit-logged, and it does not train any AI model on your content. None of that makes it an approved store for participant data at your institution.

Two more limits. The MemX Record page does not mention speaker labels, so keep speaker codes in your own quote log. The MemX Upload page lists audio files such as MP3, M4A and WAV, up to 20 MB each, but transcripts and notes are described for MemX Record sessions, so test an imported file with non-sensitive audio first. What to keep or delete stays with your ethics approval and data plan.

Frequently Asked Questions
01How do you name interview recording files?

Use one pattern fixed before the first interview: date as YYYY-MM-DD, an interviewee code instead of a name, a short topic, a file-type word and a version, joined by hyphens or underscores. Avoid spaces and words like final. Keep background details in a separate log, not only in the file name.

02How long should you keep interview recordings?

It depends on your ethics approval, data plan, consent forms and local law. One university's guidance says to delete recordings once transcripts are quality-checked if they are not needed, and that its Teams recordings are deleted after 60 days. The UK regulator says personal data should be kept no longer than necessary.

03Should interview transcripts be anonymised?

Where your ethics approval or consent requires it, yes, and early. University of Liverpool guidance says to anonymise transcripts as soon as possible after interviews. Replace identifiers with pseudonyms or generic descriptors, and store the key that links them separately from the transcripts.

04Can you trust AI transcripts for exact quotes?

Use them to find quotes, not to prove them. A 2024 study that ran OpenAI's Whisper API in 2023 found roughly 1% of audio transcriptions contained entire hallucinated phrases. Results for other tools and later versions may differ, but check any quote against the recording before it is published.

05Can MemX find a quote from an interview recording?

Record the interview with MemX Record and the transcript is indexed as a searchable memory. Ask for a quote and MemX shows the recording it came from. Transcription and search need a connection. MemX gives no legal protection and does not guarantee source anonymity, so verify every quote against the audio.

Read Next

Or try MemX to access 40+ AI models in one place, including Claude Sonnet 4.6 and GPT-5.4, and get your questions answered today.

Was this article helpful?

Found this useful? Share it with someone who needs it.

Free · iOS, Android & WhatsApp

Stop losing what you save.
Let MemX remember it for you.

Every screenshot, photo, PDF and voice note: captured, encrypted, and instantly searchable. Ask in plain English, get the answer in seconds.

  • Reads text inside images and handwriting
  • Private and encrypted by default
  • Free to start, no credit card

Takes under a minute to set up. Your data stays yours.

Arpit Tripathi
Written by
Arpit TripathiLinkedIn

Founder of MemX. Ex-Google Staff Tech Lead Manager, ex-AWS Senior SDE (Elastic Block Store). Writes about practical AI on the MemX blog.

Keep reading

More guides for AI-powered students.