Pepys

Guide

How to transcribe a therapy session

A working guide for therapists and clinical supervisors – getting consent, a private two-speaker transcript, and a de-identified record you can actually use for notes, supervision, or training.

The short answer

To transcribe a therapy session, first get every participant's recorded consent, which professional ethics codes require. Upload the recording to a transcription tool that doesn't train on your files and auto-deletes them, and get a two-speaker, timestamped draft in minutes. Then de-identify the transcript, stripping names and other identifiers, before it goes into supervision notes, research, or the record.

Do you need consent to record a therapy session?

Yes, and it isn't optional. Professional ethics codes require it directly: the APA Ethics Code says psychologists obtain permission from all persons before recording their voices or images (Standard 4.03), and the ACA Code of Ethics requires counselors to get client permission before recording sessions (Section B.6.c). Recording consent is separate from consenting to treatment.

State wiretap law sits on top of the ethics rules. Most US states let one party consent to recording a conversation, but about a dozen require every party's consent (Reporters Committee, 2012) – among them California, Florida, Illinois, Pennsylvania, and Washington. A therapy room usually has two parties. In an all-party state, you need the client's clear yes on the record, not just your own.

Capture the consent in the recording itself, the way you would timestamp any agreement. Informed consent for the therapy is already required separately (APA Standard 3.10); permission to record is an additional yes on top of it. For a minor or a client with a guardian, you need the legal representative's permission too. We can't give legal advice – when your state or your board's rules are unclear, ask before the session starts.

How to transcribe a therapy session with an AI first pass

Typing a session transcript by hand is the slow path. Manual transcription can take up to six hours of work for one hour of audio (Haberl et al., 2023, citing Bell et al. 2018) – most of a clinical day spent on a single 50-minute session. An AI first pass turns that into minutes of processing plus a focused review, so you're editing, not retyping.

This isn't a bot that joins the session. You're not putting a live meeting-notes assistant in the room between you and a client; you record the session yourself, then upload your own file when you're ready. That keeps the tool out of the therapeutic space and leaves you in control of what gets processed. The base workflow – record, AI first pass, then manual cleanup – is the same one covered in how to transcribe an interview. A session just adds stricter consent and privacy steps around it.

Where the machine still needs a human: names, medication and diagnostic terms, and the moments where therapist and client talk over each other. Those are exactly the passages that carry clinical weight, so spend your review time there. If a stretch is genuinely inaudible, bracket it with its timestamp rather than guessing at what a client said.

How do you keep a therapy transcript private?

Treat the recording as sensitive from the moment you hit stop. Two controls matter most: use a tool that doesn't train AI on your files and deletes them on a known schedule, and de-identify the transcript before it travels anywhere. Removing a name isn't enough – a 5-digit ZIP, gender, and date of birth uniquely identify roughly 87% of Americans (Sweeney, 2000).

Work from an established checklist rather than eyeballing it. HIPAA's Safe Harbor method lists 18 categories of identifiers to remove (45 CFR 164.514), from names through to any other unique code, and it's a solid baseline for a session transcript. The full de-identification pass is its own job. You strip the direct identifiers, then generalize the quasi-identifiers that still single someone out – covered step by step in how to anonymize a transcript. Sweeney's figure was later recomputed at about 63% on 2000 census data (Golle, 2006), but either way, combined details re-identify people.

Where the audio and transcript live matters more here than almost anywhere. Pepys never trains on your audio or text, and source media auto-deletes 30 days after upload by default, with an unclaimed anonymous job purged in about 12 hours. That said, identifiable session audio is protected health information if you're a covered entity, and we can't give you compliance advice. You're responsible for knowing what your obligations require before you upload.

How do you get an accurate two-speaker transcript?

A therapy session is a two-speaker recording, and keeping the two straight is half the value. A therapist-and-client two-speaker labeling job separates who said what, so a supervisor, or you a week later, can follow the exchange without re-listening. Record so the voices stay separable: use two mics, or a room where you aren't talking over each other. The labels come out cleaner.

Decide your transcription style before you edit, because in clinical and research use it shapes the analysis, not just readability. Transcription is an act of representation that affects how participants are understood and the conclusions drawn (Oliver, Serovich & Mason, 2005). Strict verbatim keeps every pause, 'um,' and false start, which you want when how something was said is the clinical point. Clean verbatim drops filler but keeps the client's actual words.

If the recording is data for psychotherapy process research, don't invent your own formatting. Standardized transcription guidelines exist precisely because collaborative research is hindered by the lack of them (Mergenthaler & Stinson, 1992). When session tapes feed a coding workflow, the transcript is analysis data, and qualitative research transcription covers how it fits the coding step. Match the standard your method or your supervisor expects.

Progress notes, supervision, and how long to keep the recording

Know which record you're creating. HIPAA draws a hard line: psychotherapy notes are the notes a provider records analyzing a counseling session, kept separate from the rest of the medical record (45 CFR 164.501). A transcript you build for your own process notes or supervision can sit in that separate category – but only if you actually keep it separate.

That separation carries a real benefit. A covered entity generally must obtain a separate patient authorization to use or disclose psychotherapy notes (45 CFR 164.508), with narrow exceptions. So a transcript used for your own supervision or training is handled differently from one going into the shared medical record. Decide the transcript's purpose up front, because it changes who can see it and under what agreement.

Retention is set by law and your board, not by habit. The APA Record Keeping Guidelines set a default that state law can override (APA, 2007). Hold records until 7 years after the last date of service for adults, or 3 years past a minor reaching majority, whichever is later. A session recording and its transcript are part of that record, so store the master somewhere access-controlled and delete working copies when they've done their job.

The steps, in order

  1. 01

    Get recorded consent to record

    Ask every participant for permission to record, on top of consent to treatment, and capture the yes in the recording. In an all-party-consent state, you need the client's clear agreement, not just your own.

  2. 02

    Choose a private tool and upload

    Record the session yourself, then upload the file to a transcription tool that doesn't train on your data and deletes it on a known schedule. Avoid live bots that join the session.

  3. 03

    Get a two-speaker, timestamped draft

    Run the AI first pass to turn the hour of audio into a labeled therapist-and-client transcript in minutes, instead of up to six hours of manual typing.

  4. 04

    Review the clinical passages

    Read the draft against the audio and fix names, medication and diagnostic terms, and crosstalk. Mark anything unclear as inaudible with its timestamp rather than guessing.

  5. 05

    De-identify before it travels

    Strip direct identifiers and generalize the quasi-identifiers that still single someone out, working on a copy, before the transcript goes into supervision, research, or the record.

  6. 06

    Store and retain by the rules

    Keep the master in access-controlled storage, retain it for the period your board and state law require, and delete working copies once they've served their purpose.

Tips from people who do this a lot

  • Consent to record is separate from consent to treat. A client can agree to therapy and still decline the recording, so get the recording yes on its own and capture it in the audio.

  • You can't scrub a spoken name out of the audio before uploading, so the real controls on the raw recording are consent and a vendor that doesn't train and auto-deletes. Save de-identification for the transcript.

  • Removing the name isn't de-identification. A ZIP, gender, and birth date alone identify most of the US population, so hunt the quasi-identifiers hardest: an employer, a rare diagnosis, a specific date.

  • Keep the psychotherapy-notes transcript physically separate from the medical record. That separation is what gives it stronger disclosure protection under HIPAA.

  • Keep one un-redacted master in access-controlled storage and de-identify a copy, so you never lose the true attribution you may need for the record or your board.

Try it now

Drop in your recording or paste a link and get a clean, speaker-labeled transcript in minutes. Your first 60 minutes are free.

or paste a link
InstagramTikTokYouTubeFacebookSpotifyApple Podcasts

60 min free · no card required · we never train on your audio

PodcasterJournalistContent creatorResearcherStudent
Trusted by 100,000+ creators, podcasters, journalists & researchers

How to transcribe a therapy session – questions, answered

Do you need consent to record a therapy session?

Yes. Professional ethics codes require it: the APA Ethics Code (Standard 4.03) and the ACA Code of Ethics (Section B.6.c) both require permission before recording a client. State wiretap law adds to that – about a dozen US states require every party's consent. Get the client's clear yes on the record.

Is it safe to upload therapy audio to a transcription tool?

It depends on the tool and your obligations. Use one that doesn't train AI on your files and deletes them on a known schedule. Identifiable session audio is protected health information if you're a covered entity, so get consent, de-identify the transcript, and know your own compliance requirements before uploading.

How do I de-identify a session transcript?

Strip direct identifiers – names, addresses, employers, ID numbers – then generalize the quasi-identifiers that still single someone out. HIPAA's Safe Harbor method lists 18 identifier categories as a baseline. A ZIP, gender, and birth date alone identify roughly 87% of Americans, so removing the name is never enough.

Should a therapy transcript be verbatim or cleaned up?

It depends on use. Strict verbatim keeps every pause and 'um,' which matters when how something was said is the clinical point; clean verbatim drops filler but keeps the client's words. Transcription method shapes analysis, not just readability, so pick one style and apply it consistently.

How long should I keep a session recording?

By law and your board's rules, not by habit. The APA Record Keeping Guidelines suggest holding records until 7 years after the last service for adults, or 3 years past a minor reaching majority, whichever is later. This is subject to superseding law. Store the master securely and delete working copies.

References

  1. 1.APA Ethics Code, Standards 4.03 (Recording) and 3.10 (Informed Consent)American Psychological Association
  2. 2.ACA Code of Ethics (2014), Section B.6.c (Permission to Record)American Counseling Association
  3. 3.Reporter's Recording Guide – state-by-state recording-consent lawsReporters Committee for Freedom of the Press
  4. 4.Haberl et al. (2023), Take the aTrain – transcription time cost, citing Bell et al. (2018)arXiv / University of Graz
  5. 5.Sweeney (2000), Simple Demographics Often Identify People Uniquely (Data Privacy Working Paper 3)Carnegie Mellon University Data Privacy Lab
  6. 6.Golle (2006), Revisiting the Uniqueness of Simple Demographics in the US Population (WPES'06)ACM Workshop on Privacy in the Electronic Society (peer-reviewed)
  7. 7.45 CFR § 164.514(b)(2) – HIPAA Safe Harbor de-identification (18 identifiers)Cornell Law School Legal Information Institute (official CFR text)
  8. 8.Oliver, Serovich & Mason (2005), Constraints and Opportunities with Interview TranscriptionSocial Forces (Oxford University Press), via PubMed Central
  9. 9.Mergenthaler & Stinson (1992), Psychotherapy Transcription Standards, Psychotherapy Research 2(2)Psychotherapy Research (Taylor & Francis)
  10. 10.45 CFR § 164.501 – HIPAA definition of psychotherapy notesCornell Law School Legal Information Institute (official CFR text)
  11. 11.45 CFR § 164.508(a)(2) – HIPAA authorization to disclose psychotherapy notesCornell Law School Legal Information Institute (official CFR text)
  12. 12.APA Record Keeping Guidelines (2007), Guideline 7 – Retention of RecordsAmerican Psychological Association (American Psychologist, 2007)

Keep reading

Don't just take our word for it.

Ask ChatGPT, Claude, or Perplexity what Pepys is and who it's for. One click, and your favorite AI does the homework.

Get your transcript – free to start

Pay as you go – credits never expire, nothing to cancel. Or start free with 60 minutes, no card.