AI note-taking for couples and family therapy: how to choose

An AI note-taking tool for couples and family therapy is software that transcribes a session where several people speak, attributes each statement to the right person, then helps the therapist write the session note. You do not choose it like a tool for individual work: voice attribution, the agreement of every person present and how the record is organised matter as much as the quality of the text.

What a multi-person session changes for the tool

With a couple, one reproach attributed to the wrong partner is enough to distort the whole sequence. Four difficulties come on top of individual work.

• Several voices, often at once. Vendors say so themselves: attribution gets harder as the number of voices grows. Yet a systemic note rests on sequences (“when she says A, he answers B”).
• Each person's consent. Two people consent, not “the couple”.
• Access to the record. A couples note describes both. If one asks for a copy, for example after a separation, the other's data is at stake.
• Secrets in individual sessions. A disclosure made in an individual session must not end up in the joint record.

How to keep a couples record is covered in our couples therapy session notes template, and for groups in our group therapy session note template. Here, one question only: what these situations demand from the tool.

What the rules say, and what they do not

No text we read for this article deals with AI note-taking in couples therapy. The useful rules are general.

• GDPR, Article 9. Processing health data is prohibited except in listed cases, including the person's explicit consent (9.2 a) and health or social care under a contract with a health professional (9.2 h). Full text on EUR-Lex; French text on the CNIL website.
• GDPR, Article 13. The controller must give the person, at the time the data is obtained, the purposes, the legal basis, the recipients and the retention period, among other things. In couples work, this applies to each person.
• GDPR, Article 15. Each person has a right of access to their data. Paragraph 4 sets a limit: “The right to obtain a copy referred to in paragraph 3 shall not adversely affect the rights and freedoms of others.”
• French code of ethics for psychologists (2021). Article 7: psychologists are bound by professional secrecy within the limits of the Penal Code. Article 9: they seek the free and informed consent of those who consult them. The code has no article specific to couples or families. Text of the code (in French).
• French Public Health Code, Article L1111-7. It gives access to health information held by health professionals and institutions, except information stating it was obtained from third parties not involved in the care, or concerning such a third party. The text does not settle whether it applies to psychologists and psychotherapists in private practice. Text on Légifrance.
• CADA opinion of 21 February 2013. Asked about family therapy in a public hospital (CH Esquirol, Limoges), France's access-to-documents commission found the notes concerned the health of each participant, not third parties, and could be released without redaction. An opinion for a public institution, not a general rule for private practice. Opinion on the CADA website.

Our practical reading: assume a joint note may one day be read by each partner. The GDPR applies the same way in Belgium; in Switzerland, we recommend checking Swiss data protection rules before choosing a tool.

Selection grid: criteria specific to couples and families

General criteria (hosting, price, language quality) are covered in our guide to software for therapists and our comparison of AI note-taking software. For couples and families, add these.

• Separate voices: does the transcript distinguish voices, or is it one continuous text? Continuous text is not enough for couples.
• Names or numbers: “Speaker 1, 2, 3” or first names? Numbers are fine for the transcript, not for the final note, and you have to redo the mapping every session.
• Fixing: can you reassign one line or a whole voice, and does the note follow?
• Overlap and number of voices: what happens to a sentence spoken by two people at once? Can the tool handle a family of five, and does the vendor state a limit?
• Therapist clearly identified: a hypothesis you voice must not become a client statement.
• Record structure: can you keep a couples record and an individual record for each partner, and choose for each session where it goes?
• Balanced summary: does the case summary cover both partners, or only the one who opened the file?
• One person refuses: can you switch to dictation after the session without changing tools?
• Export per record: if access is requested, can you cleanly export the notes of one specific record?
• Audio and hosting: is audio kept after transcription? Where is the data stored? In France, an HDS-certified host is the expected benchmark for health data.

The test to run during the trial

Do not test a tool on a real couple. Run a 20-minute role play with two colleagues on an invented situation.

• Three voices: you and two fictional partners, one of whom speaks little and quietly.
• Three traps: a sentence where both speak at once, a very short reply (“No.”) and a line where one partner quotes the other (“He told me I was overreacting”).
• Count attribution errors on ten lines chosen in advance. An error on a reproach counts double.
• Fix one error, time it, and check that the note changes too.
• Read the note: are both points of view present and balanced?
• Simulate an individual session, file it in a separate record, then export each record and look at what comes out.
• Look on the vendor's website for hosting, what happens to audio and whether data trains a model. If it is not written down, ask in writing.

Comparison: what the vendors' websites say

Read on each vendor's official website on 5 November 2026. Anything missing is “not stated on its website”. These pages change, so check before you choose.

• Heidi | Voices: according to its website, handles “multiple speakers”; attributing statements to each person is not stated on its website. | Hosting: data in the EU/EEA according to its GDPR page; country and HDS certification not stated. | Audio: not kept after transcription, according to its safety page. | Strengths: more than 110 languages, according to its website.
• Upheal | Voices: according to its help centre, a couple is set up as one client and you confirm or reassign speakers after the session; family therapy is in beta, up to 10 clients, speaker labels sometimes need fixing, attribution is best when each person joins from their own device. | Hosting: AWS, GDPR and SOC 2 Type II according to its compliance page; hosting country not stated. | Audio: retention not stated on that page. | Strengths: the most detailed couples and family features among the tools read here.
• Tandem Health | Voices: voice separation not stated on its website. | Hosting: according to its security page, data processed and stored exclusively in European data centres on HDS servers; ISO 27001 and ISO 13485 certified. | Audio: deleted after transcription, according to its website. | Strengths: a clearly described health-data hosting set-up.
• Plaud | Voices: according to its help centre, “Speaker 1 / Speaker 2 / Speaker 3” labels that are not verified names, less reliable when voices overlap. | Hosting: for Europe, AWS in Frankfurt according to its trust centre; HDS not stated. | Audio: not kept in the cloud unless the user turns on Cloud Sync. | Strengths: small standalone device; general-purpose tool.
• Delta | Voices: distinguishes voices in couples, family and group sessions. | Hosting: in France, with an HDS-certified host, AI processing in France. | Audio: no audio file kept. | Strengths: session report and patient record updated in one place.

Information and consent form (one per participant)

No text we read requires a signed form. We recommend one for couples and families: it keeps a trace of each person's agreement. One sheet per person, before the first transcribed session; adapt the brackets.

Information about AI-assisted note-taking in couples or family sessions

Practitioner: [Name, profession, practice address, phone, email]. Controller of your data.

What happens: during our joint sessions, software ([tool name]) transcribes what is said and distinguishes the voices of the people present. The transcript is used only to prepare the session note, which I review and correct myself.

Why: to keep an accurate record of our work, including what each of you expressed.

Legal basis: your care and your explicit agreement, given on this sheet.

Where the data goes: [host, country, certification]. Audio [is not kept after transcription / is kept for X days]. The data [is not / is] used to train an AI model.

Who has access: only me. [If relevant: the software vendor, as processor.]

How the record is organised: [joint sessions are noted in a joint record / each person has their own record]. Any individual sessions are noted in a separate record.

Retention: [period] after the last session.

Your rights: you may ask for access to your data, correction or erasure within the limits set by law. An access request is handled with regard to the rights of the other people present in session.

Your choice: you may refuse, now or at any time, including during a session. If any one person present refuses, the session is not transcribed and I take my notes another way. Refusing changes nothing about your care.

• I have read this information and been able to ask questions.
• I agree that joint sessions are transcribed by the software named above: YES / NO
• I understand I can withdraw my agreement at any time.

Participant's full name: ........ Date: ........ Signature: ........

For a minor: name and signature of the legal guardian(s): ........ Child's view sought orally on: ........

Fictional example: a couples session note from a transcript with separated voices

Invented situation. Session 4 with Claire M. and Julien M., together for nine years, seeking help for repeated arguments since the birth of their second child. Both signed the consent sheet at the first session.

Transcript excerpt, after voice attribution

• Therapist: What happened on Sunday, each in your own words?
• Claire M.: I asked him to do the bath, and he said “in five minutes”. An hour later, nothing.
• Julien M.: I was finishing a file for Monday. I'd said so that morning.
• Claire M.: You always say that. Monday always comes first.
• Julien M.: And that's when I go quiet, because otherwise it escalates.

Attribution check: the tool had assigned “I'd said so that morning” to Claire M. Line reassigned to Julien M. before the note was generated.

Session note

Setting: session 4, in person, 60 minutes. Present: Claire M., Julien M. Transcription accepted by both, agreement restated at the start.

What each expressed: Claire M. describes repeated requests left unanswered and a feeling of coming after Julien's work. Julien M. says he warned in the morning about work due and goes quiet to avoid escalation.

Observed sequence: Claire's request, Julien's delay, Claire's generalised reproach (“always”), Julien's silent withdrawal, then a new request from Claire. Sequence already noted in sessions 2 and 3.

Interventions: describing the sequence in both partners' words; circular question (“What does Julien do when Claire raises her voice? What does Claire do when Julien goes quiet?”).

Working hypothesis (to check): Julien's withdrawal, meant to calm things, is felt by Claire as absence, which restarts the request.

Task: each notes once this week the moment they feel the sequence start, without discussing it with the other before the session.

Next: go over each person's observations; discuss sharing weekend tasks.

More ready-to-copy outlines are in our clinical document templates.

Tricky situations: what we recommend

• A partner withdraws agreement mid-session: stop transcription at once and say so out loud. Ask whether they want the part already transcribed left out of the note.
• A request for a copy after separation: first check which record holds the notes. For a joint note, seek advice (professional body, union, indemnity insurer) before releasing it, keeping GDPR Article 15(4) in mind.
• A child in a family session: the code (Article 11) asks psychologists to seek the authorisation of legal guardians; also explain transcription to the child in simple words.

Common mistakes

• Getting agreement from the partner who booked and assuming the other goes along.
• Filing an individual session in the couples record out of convenience.
• Letting the tool turn a hypothesis into a fact (“Julien avoids conflict” instead of “hypothesis to check”).

Why Delta fits couples and family work

Delta distinguishes voices in couples, family and group sessions, then writes the session report and updates the patient record, where the whole course of care sits. If a participant declines transcription, you dictate your observations after the session. Data is hosted in France with an HDS-certified host, names and identifying details are pseudonymised before the AI, no audio file is kept and data is never used to train a model. See what Delta is or see Delta for therapists.

Sources

• GDPR, Regulation (EU) 2016/679, on EUR-Lex
• GDPR, Articles 13 and 15, French text on the CNIL website (accessed 5 November 2026)
• French code of ethics for psychologists, 2021 version (Articles 7, 9, 11, 13)
• French Public Health Code, Article L1111-7, on Légifrance
• CADA opinion 20130447 of 21 February 2013, family therapy