MentAITherapy session transcription: how to do it safely
Turning a recorded session into text has stopped being a technical problem. The problem is where the recording ends up and what remains of it once the note is ready.
A conversation about transcription usually starts with the tool, when it should start with the question of why you need it. Only the answer to that question tells you how much of the recording you may keep, and for how long.
Three different reasons, three different answers
Therapists reach for transcription for three reasons that are easy to confuse and that lead to completely different decisions:
- A note for the records. You want an entry in the client's file without typing it up in the evening. The recording is only an intermediate product here; once the note exists, there is no reason for it to remain.
- Material for your own work. You want to go back to a passage and see what happened in the conversation. Here it is tempting to keep the transcript longer, and this is exactly where it gets risky.
- An archive. You want the recordings "just in case". This is the weakest of the three reasons: a collection of session recordings is a collection of health data that has to be guarded for years and is hardly ever opened.
The principle of data minimisation works very practically here: the shorter a recording exists, the smaller the problem you have if a laptop is stolen, a file is shared by mistake or a client asks for their data to be erased.
Consent from every participant, collected before recording
The content of a session is health data, a category under special protection. The legal basis for recording is explicit consent under Article 9(2)(a) of the GDPR, given before recording starts, not ticked off afterwards.
In couples sessions and in groups each participant gives consent separately. One person without consent means no recording; there is no "the majority agreed" version here.
In MentAI the module asks you to record the consent of every participant before starting and accepts three ways of collecting it: online, in writing, or verbally with a note. What stays in the system is the method and the date, not the wording of the consent or the participants' names. You can print a consent template, together with a clause for group sessions, from the panel.
We take apart the legal side of the line between AI help and professional confidentiality separately in AI session notes and professional confidentiality.
The question that settles the choice of tool
What remains after transcription, once the note is ready?
This one question says more about a provider than their whole features page. It is worth asking it directly and requesting an answer in writing. If the answer is evasive, in practice it usually means: everything stays.
Our answer is this: the recording, the raw transcript and the processing job itself are deleted as soon as the transcript is returned (the draft note is created only on request, from the text in the therapist's window). The only lasting trace is the note approved by the professional and an access log that contains no content of the conversation. A draft you do not approve is kept nowhere.
If you want to keep a copy for yourself, the module lets you download the recording and the transcript to your own drive. At that point, however, the copy becomes your practice's documentation, with your retention period and your responsibility, and it is good to know that before you click.
Recordings you already have
Many therapists have an archive on a voice recorder or a phone, and that archive is usually the real reason for looking into transcription. The module accepts existing files in mp3, m4a, wav, ogg, flac and webm, up to 150 MB and 90 minutes per recording.
One thing needs to be clear here: responsibility for the legality of the recording and for the consent in force at the time it was made stays with the practice. No system can check retrospectively whether the client knew and agreed back then. If you do not have that certainty, an old recording is better deleted than transcribed.
Couples and group sessions
With several people in the room, plain text without speaker separation is of little use. The engine separates the utterances of individual people, so the transcript shows who said which line. It does not recognise identities: it labels speakers in order, and matching them to specific people stays with you.
You declare how many people take part in the conversation besides you, and the separation depends on that. With larger groups the separation can be less accurate, especially when participants talk at the same time.
How much it costs
Session transcription costs PLN 69 per month per practice (about €16), which includes up to 1,500 minutes of recordings a month. It is one of two optional paid add-ons to the subscription (the other is video sessions, also PLN 69 a month); everything else, from online booking to invoicing, is included in the price.
The technical limits are stated openly: 90 minutes per recording and 20 recordings a day.
During the trial the module works as a taster, and we say so plainly so that nobody signs up with a false expectation: 50 minutes in total, not per month, and only recordings made live in the panel. Uploading an archive from a voice recorder and the full 1,500 minutes work from a paid subscription.
What transcription will not do
A transcript of a conversation is not supervision and does not replace it. Nor is it a finished note: it is a draft that a human approves, and only that decision creates a record.
Speech recognition can be unreliable with overlapping voices, quiet recordings or specialist vocabulary. You read a transcript with that in mind, not like a court stenographer's record.
If what you are looking for is not so much a transcript as a look at your own thinking about the session, the panel has a separate tool for that: instead of summarising the conversation, it asks questions about your own reasoning in the work. It is not supervision either, and does not replace it.
Frequently asked questions (FAQ)
Can I record a therapy session in order to transcribe it?
With the explicit consent of every participant in the conversation, collected before recording starts. The content of a session is health data, which enjoys special protection; the legal basis is explicit consent under Article 9(2)(a) of the GDPR. In couples and group sessions each participant gives consent separately.
What happens to the recording after transcription?
In MentAI the recording, the raw transcript and the processing job are deleted as soon as the transcript is returned (the draft note is created only on request, from the text in the therapist's window). What remains is the note approved by the professional and an access log without content. Other providers may answer differently, and it is worth finding out before you decide.
Can I transcribe recordings I already have on a voice recorder?
Yes, in mp3, m4a, wav, ogg, flac and webm, up to 150 MB and 90 minutes per recording. The legality of the recording and the consent from the time it was made stay with the practice. Uploading archives works from a paid subscription.
How much does session transcription cost in MentAI?
PLN 69 per month per practice (about €16), up to 1,500 minutes of recordings. During the trial: 50 minutes in total and live recordings only.
This article is for information only and does not constitute legal advice. It is worth consulting a lawyer about the design of consent forms and records for your own practice. Current as of 31 August 2026.
See the panel in action: a demo with no account
The product interface is in Polish. Prefer to start for real straight away? Sign-up is open: 30 days of the full version, no card, at mentai.pl. During that period the transcription module has a limit of 50 minutes and accepts live recordings only.
Related posts: AI session notes and professional confidentiality · An AI assistant for therapists: what AI should not do · Client data security in your practice.
New posts every week (in Polish). Follow MentAI on Facebook or Instagram: one thing a week that takes paper off a practice's desk.