- clinical AI
- informed consent
Recording a therapy session is not recording a medical visit: consent, third parties and what remains
A Mexican medical-software vendor’s FAQ asks whether it is legal to record the visit so the AI can write the note, and answers: «Con el consentimiento del paciente, y tratando los datos conforme a la ley de protección de datos personales» (with the patient’s consent, and handling the data under the data-protection law). In a medical visit, almost everything recorded is about the body of the person who signed. In a therapy session, much of it is about people who signed nothing: a partner, a mother, a boss, someone who caused harm. Wherever you practise, three things follow. Your client’s consent covers your client, not the people they talk about. It can be withdrawn mid-session, and you need a way to honour that in the room. And deleting the audio does not delete the session: the transcript is the same session in text. Mexico is the worked example here, because its data-protection statute makes consent carry the legal weight and states a deletion duty in plain words. Verify the rules where you practise; this is documentation and practice management, not legal advice.
What each layer asks, and whom it covers
That recording needs permission is not in dispute; how to write it into your consent form is in the guide to informed consent in psychotherapy. What is rarely compared is whom each source covers:
| Source | What it requires | Whom it covers |
|---|---|---|
| APA Ethics Code, Standard 4.03 | permission before recording | “individuals to whom they provide services” |
| The Sociedad Mexicana de Psicología’s ethics code (SMP), one of several Mexican professional codes, art. 121 | «permiso» for «grabación electrónica de sesiones o entrevistas» | «pacientes, clientes, sujetos de investigación, estudiantes y supervisados» |
| Mexico’s data-protection law, arts. 7 and 8 | consent «expreso y por escrito» (express and written), revocable «en cualquier momento» (at any time) | the «persona titular»: the person the data belong to |
The SMP code asks for consent «por escrito» (in writing) to record only in research (article 126); for clinical sessions it speaks of «permiso», and the writing requirement comes from the data law. And the last column is the one nobody reads. Both ethics codes speak of the people you serve; the statute speaks of the person the data belong to. In a medical visit those coincide. In a session they do not.
One signature, many voices
Under the Mexican statute, personal data are «cualquier información concerniente a una persona identificada o identificable» (any information about an identifiable person; article 2, fraction V), and article 7 makes processing subject «al consentimiento de la persona titular»: the consent of the person the data belong to. When your client says “my sister has HIV and hasn’t told her husband,” that health datum, sensitive by definition, belongs to the sister. Your client’s signature is not hers.
None of the seven consent exemptions in article 9 covers recording. Fraction I (where a legal provision so provides) could cover what Mexico’s clinical-record standard, NOM-004, requires you to note, such as family history (clause 6.1.1), but no rule requires you to record; fraction VI covers health care only «mientras la persona titular no esté en condiciones de otorgar el consentimiento» (while the subject is unable to consent). NOM-004 authorises nothing about third parties either; it protects what is said: clause 5.5.1 makes confidential the «datos proporcionados al personal de salud, por el paciente o por terceros» (data given by the patient or by third parties). We found no regulator guidance on third-party data that arrives through the client’s own account, and for anything beyond what the standard requires you to note, we found no exemption that covers it.
The contrast with the UK is instructive. The UK GDPR lifts the ban on health data where processing is necessary for “the provision of health or social care or treatment” under a professional duty of secrecy (article 9(2)(h) and 9(3)); explicit consent is one route among several. Mexico’s health exemption is far narrower, so consent does most of the work, and the third party never gave any.
What the Mexican statute does offer is proportionality. Mexico’s article 12 limits processing to what is «necesario, adecuado y relevante» (necessary, adequate and relevant), and the ethics codes apply the same principle to what you communicate: APA Standard 4.04(a) asks psychologists to include in reports “only information germane to the purpose,” and the SMP code’s article 133 says the same. A note applies that principle by construction: you write “reports distress over a family health secret” and lose nothing clinically. A recording never does: it keeps the name, the diagnosis and the husband’s name. A note is a filter. A recording is not.
What the recorder does to the session
If recording changes what your client says, the recorder is a clinical variable, not paperwork. Does it? The evidence is thinner than either the record-everything or the never-record camp assumes.
In a CBT centre for anxiety disorders that records routinely, Shepherd, Salkovskis and Morris surveyed 72 patients at the start of therapy, 31 at the end and 15 therapists: most were positive, “feeling self-conscious” was among the disadvantages, and therapists worried more than patients that listening would be distressing. With five clients and 25 therapists, Brown, Moller and Ramsey-Wade found that all clients said the device was soon forgotten, that both groups saw the benefits as being “purely for the therapist,” and that clients perhaps did not always understand how recordings were used.
Both studies are small and self-reported. And neither can measure what a client left unsaid because a recorder was running. What the evidence does support is more useful: the effect, if any, is not uniform. You will not see it in an average but in one person, on the day they were about to tell you something. The signals are behaviours you already track: asking whether it is still recording, lowering their voice, “don’t record this,” shifting into the hypothetical just before something important. They are material, not an obstacle, and they belong in the note like any other process observation.
When the recording is the intervention
In some treatments the recording is part of the therapy. Prolonged Exposure for PTSD is the clearest case: the US Department of Veterans Affairs’ PE Coach app is meant, in the words of its official flyer, to “Record sessions with your therapist and listen to them during the week as part of weekly homework.” In the centre Shepherd and colleagues studied, 90% of patients reported listening to recordings between sessions to some extent, and patients typically planned to keep them after therapy ended.
Then the copy lives on your client’s phone, under another regime: Mexico’s statute exempts anyone who collects and stores data «para uso exclusivamente personal, y sin fines de divulgación o utilización comercial» (for exclusively personal use, without disclosure or commercial use; article 1, fraction II). Your obligations do not travel with the file. What you can do is talk it through first (where it will be kept, who else uses that phone) and note that you did. The same carve-out is why the reverse case, a client recording you, is a different question with different rules.
Withdrawal mid-session
The last paragraph of Mexico’s article 7 has two sentences, and the second is the one that matters with the recorder running:
«El consentimiento podrá ser revocado en cualquier momento sin que se le atribuyan efectos retroactivos. Para revocar el consentimiento, el responsable deberá, en el aviso de privacidad, establecer los mecanismos y procedimientos para ello.»
(Consent may be revoked at any time, without retroactive effect; the privacy notice must set out how.) You define the mechanism. Write into your notice and consent form that, during a session, saying so aloud is enough: an email procedure is no way to stop a recorder running in front of you.
“Without retroactive effect” protects what you already did, not what you would do next, and transcribing or keeping is processing: the statute’s definition (article 2, fraction XIX) includes «conservación», «almacenamiento» and «posesión» (retention, storage, possession). The law does not say what to do with the minutes already recorded. If the recording is raw material rather than part of the record, the prudent reading, ours and not an official one, is not to transcribe them and to delete them, unless your client expressly asks you to keep them:
“I’m switching it off now. I won’t transcribe what we’ve recorded so far, and I’ll delete it today unless you’d rather I didn’t. And if you like, we can talk about what made you ask.”
The pause can also be yours, when you see someone else’s information coming:
“I’m going to pause the recording for a moment. What you’re about to tell me is about someone else, and I don’t need it on tape for us to work on it.”
Of the withdrawal itself, the note records the fact, not a transcript: “at the client’s request, recording stopped at 25 minutes; the session continued without it.” Stopping the recorder does not stop your duty to document: whatever was clinically relevant in what came next, risk included, goes in the note as in any unrecorded session.
Deleting the audio does not delete the session
Ask tools what happens to the recording and they tend to say the same thing. The Mexican vendor above: «el audio se borra a los 60 días» (the audio is deleted after 60 days). A Spanish-language tool for psychologists: «El audio no se queda» (the audio does not stay), and on the same page, «En unos minutos tienes un borrador con lo que se habló» (in a few minutes you have a draft of what was said); both pages as read on 8 October 2026. Ours: «El audio se transcribe y se elimina de inmediato: nunca se almacena» (the audio is transcribed and deleted immediately; it is never stored).
Even if all three are honoured to the letter, none says how long what survives the audio is kept: the transcript or draft, the session in text, with the same names and the same sister’s secret. In gesell.ai the audio is not kept on our servers (an outside provider transcribes it); the transcript is, as part of the record: it goes into the note’s original-text field, which locks when you sign, and is kept with the record for at least five years. The useful question for any vendor, us included, is not “do you delete the audio?” but “what is left of the session, where, and for how long?” The rest of that checklist is in is AI safe for patient data?
Record or raw material: decide what each file is
NOM-004 never names audio, but does not exclude it: clause 4.4 says the record «consta de documentos escritos, gráficos, imagenológicos, electrónicos, magnéticos» (consists of written, graphic, imaging, electronic, magnetic documents) «y de cualquier otra índole» (and of any other kind). File the recording or transcript in the record, and clause 5.4 applies: retention «por un periodo mínimo de 5 años, contados a partir de la fecha del último acto médico» (at least five years from the last clinical act). Treat it as raw material for writing the note, as the guide to better session records suggests, and article 10 of the data law applies, blunter than the minimisation clause usually cited. Once data «hayan dejado de ser necesarios para el cumplimiento de las finalidades previstas en el aviso de privacidad» (are no longer needed for the purposes stated in the privacy notice), they «deberán ser suprimidos» (must be deleted). If the stated purpose was drafting the note, the recording stopped being necessary once the note was signed.
| Question | Signed note | Transcript and audio |
|---|---|---|
| Required by a rule? | Yes: progress note (NOM-004, 6.2, adapted by 5.16) | No |
| Filters third parties? | Yes, if written well | No |
| How long is it kept? | At least 5 years (5.4) | As raw material, deleted once its purpose is met (art. 10) |
| Can the client request it? | Yes (art. 22) | Yes, while it exists, except what would harm third parties’ rights (art. 33, fr. III) |
In the US, the HIPAA right of access excludes psychotherapy notes kept apart from the record (45 CFR 164.524(a)(1)(i)); Mexico has no such category, as the guide to client record requests explains. What separates what can be requested from what cannot is possession. Article 22 grants access to data «que obren en posesión del responsable» (held by the controller), and article 33, fraction II, allows refusal «cuando los datos personales no se encuentren en posesión del responsable» (when the controller does not hold them). What you keep is within reach of that right, subject to article 33, fraction III, for what concerns third parties; what you deleted is not.
For your client, the transcript holds few surprises: the words are mostly their own. The risk is whatever takes it out of the consulting room. Where it is in force, article 271 of Mexico’s national civil and family procedure code obliges third parties to produce «sin demora» (without delay) the «archivos electrónicos que tengan en su poder» (electronic files in their possession), while the same article bars testimony from anyone bound to professional secrecy. Your testimony is inadmissible; your files have no such rule: you can object, but the judge rules «sin ulterior recurso» (with no further appeal), and a recording turns what was said into a file. In a family case, the person your client talked about may be the other party. The code enters into force state by state, by local declaration, no later than 1 April 2027, so check whether it governs your case yet. Why testimony and records travel on separate tracks is the subject of the guide to confidentiality and legal requests.
A consent that describes a session, not a visit
A consent that only asks “may I record?” describes a medical visit. One that describes a session answers five things: who benefits (usually you, and it is worth saying), who processes the audio, what remains afterwards and for how long, how the recording is stopped, and what you do when someone else comes up. What to say about an AI vendor is in must you tell clients you use AI?; the recording-specific part fits in a paragraph:
“I’d like to record the audio of some sessions. It mainly helps me write accurate notes; it gives you no direct benefit, and you can say no without anything changing in your treatment. If you agree, a tool turns the audio into text; the audio is deleted [when] and the transcript is kept [where, and for how long]. To stop the recording, just tell me, out loud, at any point. And when you talk about other people I can pause it: I don’t need their names on tape to work with you.”
Recording is a clinical decision made session by session, not a setting you switch on once.
References
- Cámara de Diputados — Ley Federal de Protección de Datos Personales en Posesión de los Particulares (DOF 20 March 2025, last amended 14 November 2025)
- Diario Oficial de la Federación — NOM-004-SSA3-2012, Del expediente clínico
- Cámara de Diputados — Código Nacional de Procedimientos Civiles y Familiares (current text)
- Sociedad Mexicana de Psicología — Código Ético del Psicólogo (4th ed., Trillas, 2007; reprinted 2009)
- American Psychological Association — Ethical Principles of Psychologists and Code of Conduct (2017)
- UK GDPR, Article 9 — Processing of special categories of personal data (legislation.gov.uk)
- Shepherd, Salkovskis and Morris — Recording therapy sessions: an evaluation of patient and therapist reported behaviours, attitudes and preferences (Behav Cogn Psychother, 2009)
- Brown, Moller and Ramsey-Wade — Recording therapy sessions: What do clients and therapists really think? (Counselling and Psychotherapy Research, 2013)
Ready to spend less time on documentation?
gesell.ai turns your session notes into professional clinical documentation — SOAP, DAP, BIRP, or GIRP — and generates treatment plans with SMART goals.
Try gesell.aiHave a question about your practice?
How to document a case, which format suits you, or how Gesell fits your practice: write to us and we will answer.
About the author
Gesell Team
Clinical and product content written by the gesell.ai team together with certified clinical psychologists.