Resources

By Clara Team

Is it safe to use ChatGPT for therapy notes?

Short answer: pasting notes or transcripts that could identify a client into personal ChatGPT (Free, Plus or Pro) is not a safe practice. They are health data, which the GDPR protects as a special category, and those accounts come with no contract governing how OpenAI processes that data on your behalf, what you write may be used to train its models unless you turn that off, and the data is processed outside the European Economic Area. OpenAI's business plans solve part of this, but not all of it. If you want AI to help with your notes, use a tool with a processing agreement, data in the EU and no training on your data, and always review what it writes.

This guide is for orientation and is not legal advice. The OpenAI and AEPD quotes come from their official pages as read on 19 November 2026; the AEPD's are our translations from Spanish, and each link goes to the original.

Why therapy notes are not just any text

What a person says in session is health data, one of the GDPR's special categories. Article 9 prohibits processing it except in specific cases, and the one that lets a psychologist work is health care provided by, or under the responsibility of, a professional bound by professional secrecy (Articles 9(2)(h) and 9(3)). The AEPD's guide for health professionals, last revised in October 2024 and now marked "under review", sums it up: this data "deserves reinforced protection and may only be processed, that is, stored, transmitted or disclosed, among other forms of processing, under certain conditions and with specific safeguards".

Professional secrecy comes on top of that. The Spanish psychologists' code of ethics says everything you learn in your work "is subject to a duty and a right of professional secrecy" (Article 40) and that electronic records are kept "under conditions of security and secrecy that prevent outsiders from accessing them" (Article 46). Spain's data protection act, Organic Law 3/2018, extends the duty of confidentiality to everyone involved in the processing, providers included (Article 5).

And the responsibility is yours. In the AEPD's words, "the health professional is the controller when providing health services individually". If you work for a centre, the centre is usually the controller.

What happens to what you type into personal ChatGPT

This is what OpenAI says today about its personal accounts (Free, Plus and Pro):

  • It may train on your conversations. "When you use our services for individuals (Free, Plus, Pro), we may use your content to train our models" (how your data is used to improve model performance). It is on by default and can be turned off under Settings › Data controls, with "Improve the model for everyone".
  • With one exception. Even with that off, if you rate a response with a thumbs up or down, OpenAI "may use the entire conversation associated with that feedback to improve our models".
  • Temporary chats are kept too. They are not used for training, but OpenAI "may keep a copy of a temporary chat for up to 30 days for safety purposes" (temporary chat).
  • Deleting does not always delete. Its EU privacy policy says data is removed within 30 days, unless it has already been de-identified and disassociated from your account to improve the models.
  • The data leaves the European Economic Area. The same policy says "OpenAI processes your Personal Data on servers located outside of the EEA, Switzerland and the UK".
  • There is no processor agreement. OpenAI's services agreement, which its data processing addendum is part of, covers its services for businesses and developers and "does not apply to OpenAI services used by consumers or individuals" (services agreement).

Turning off training helps, but it does not fix the rest. The GDPR requires anyone processing data on your behalf to do so under a contract that sets out what they may do with it (Article 28), and any transfer outside the EU to meet its conditions (Article 44). A personal account gives you neither.

What changes with OpenAI's business plans

OpenAI has several plans for organisations: ChatGPT Business, Enterprise, Edu, ChatGPT for Healthcare and its API. For these, its commitments are different (enterprise privacy):

  • No training by default: "By default, we do not use your business data for training our models."
  • A contract: "we are able to execute a Data Processing Addendum (DPA) with customers for their use of ChatGPT Business, ChatGPT Enterprise, and the API". For customers in the European Economic Area, it is signed with OpenAI Ireland Ltd (OpenAI's DPA).
  • Control over retention: on Business, workspace admins decide how long conversations are kept.

Three questions remain open for a psychologist in Spain:

  1. Where the data is stored. OpenAI offers storage in Europe to "eligible ChatGPT Enterprise, ChatGPT Edu, ChatGPT for Healthcare, and API platform customers" (business data). ChatGPT Business is not on that list. And even with data stored in Europe, OpenAI notes that "non-GPU processing may still occur globally" (data residency).
  2. Whether the contract allows for health data. The DPA's schedule says: "No sensitive data is intended to be transferred unless the user includes it unexpectedly in unstructured data." Therapy notes are sensitive data by definition.
  3. Whether the health product is available here. ChatGPT for Healthcare is an enterprise version for health organisations, built around US law (HIPAA) and bought through OpenAI's sales team; OpenAI does not say whether it is offered in Spain. ChatGPT for Clinicians is only for verified clinicians in the United States, and its list of professions does not include psychologists.

A business plan is a serious step, but it does not do your homework for you: read the contract, know where the data goes, weigh the risk to your clients and tell them.

What if I remove the client's name?

It helps, but it is rarely enough. The GDPR treats pseudonymised data that can still be attributed to a person as personal data (Recital 26). A therapy note tends to carry age, job, family set-up, a specific event or a neighbourhood, and that combination can be enough to recognise someone, especially in a small town.

The AEPD takes the same line. Its guide "Cuidado con lo que le confIAs" suggests describing "a fictional case" and avoiding "details that could identify you". If you want help thinking through a case, an invented example is safer than a trimmed real note.

What the AEPD says about AI and client data

Spain's data protection authority has no specific guide on AI in health care, but its general recommendations from 2026 are clear:

  • In January 2026 it advised against sharing "delicate or sensitive information" with AI, with "medical details" as the first example, and, at work, against including information that reveals confidential data "about staff or clients".
  • In October 2026, in its ten points on AI assistants and agents, it was more direct: "Don't install agents on your own on your work computer or connect your work email or documents to a personal AI account. The data of clients, patients, students or citizens remains your organisation's responsibility, and you are responsible for how you use it." And: "Being able to use a tool in your personal life does not mean it is suitable for professional use."

On any provider that processes your clients' data, its guide for health professionals says the relationship "must be set out in a written contract" and that you should check, among other things, that it meets the rules on international data transfers.

What to ask any AI tool

Before using an AI tool with clinical material, whether ChatGPT or a tool for psychologists, ask in writing:

  • Contract. Will you sign a processor agreement with me (GDPR Article 28)? Does it cover health data?
  • Location. Where is the data stored and where is it processed, including transcription and the AI models? If anything leaves the EU, under what safeguards?
  • Training. Do you train models on my data or my clients'? By default, or only if I opt in?
  • Retention. How long do you keep audio, transcripts and notes? Can I delete them?
  • Access. Who on your team can see clinical content? Is that logged?
  • Security. How is the data encrypted, in transit and at rest?
  • Breaches. How quickly will you tell me? You have 72 hours to notify the AEPD once you know. The guide on what to do if your software has a breach explains how.
  • Leaving. What happens to the data when I leave?
  • Clients. How do I inform them, and how do I collect their consent to the use of AI?
  • Review. Does anything the AI writes reach the clinical record without my review?

If an answer is vague, treat it as a no. The GDPR guide for psychologists and the guide on AI notes: limits and safety go into these points.

How Clara handles it

Clara is an AI tool built for psychologists. After each session it prepares a draft clinical note in the language of your orientation, which you review, and it keeps the clinical history, calendar and bookings, reminders, consent and invoicing in the same place. This is how it answers the questions above:

  • Contract: we sign a Data Processing Agreement with every professional, which you can read before signing up.
  • Location: notes, transcripts and audio stay in the European Union: the database and audio in Spain (Microsoft Azure), transcription in Sweden and Ireland, and the AI models in Sweden (Azure). Emails you send from Clara, such as consent requests and invoices, go through a US provider under the EU-US Data Privacy Framework and standard contractual clauses. If you turn on WhatsApp reminders, Meta may also process those messages in the United States, with the same safeguards.
  • Training: clinical data is never used to train models.
  • Retention: session audio is deleted after 14 days; notes and transcripts when you delete them.
  • Access: Clara's team never sees notes or transcripts.
  • Security: notes, transcripts and client details are encrypted one by one with AES-256-GCM before they reach the database, and everything travels encrypted with TLS 1.2 or higher.
  • Breaches: we tell you without undue delay and within 48 hours at most.
  • Clients: the consent form they sign in Clara asks separately whether they agree to the session being recorded and to the use of AI.
  • Review: the note is a draft until you review it and mark it as reviewed.

Clara is built for the GDPR, but that does not replace your own assessment: you remain the controller. The details are on the security and data protection page.

Frequently asked questions

Can I use free or Plus ChatGPT for my clients' notes?

Not safely, with anything that could identify a client. On personal accounts (Free, Plus and Pro), OpenAI may use what you write to train its models unless you turn it off, signs no processor agreement with you, and processes the data outside the European Economic Area. Spain's data protection authority, the AEPD, advises against sharing medical details or clients' data with AI tools.

What if I buy ChatGPT Business or Enterprise?

Important things change: by default, OpenAI does not train on that data, and it signs a data processing addendum with the customer. But storing data in Europe is only offered on Enterprise, Edu, ChatGPT for Healthcare and the API, not on Business, and OpenAI's own addendum says no sensitive data is intended to be transferred. Before using it with health data, check the contract, where the data is stored and processed, and the risk to your clients.

Is removing the client's name enough?

Usually not. A therapy note tends to carry details, such as age, job, family or specific events, that can identify a person. As long as someone can be identified, the data is still personal data under the GDPR. Removing the name lowers the risk but does not make the note anonymous.

What does the AEPD say about using AI with client data?

In its 2026 recommendations it advises against sharing medical details or clients' or patients' data with AI tools and against connecting work documents to a personal AI account, and it points out that being able to use a tool in your personal life does not make it suitable for professional use.

What should I ask an AI tool before using it with clients?

Whether it signs a processor agreement with you that covers health data, where it stores and processes the data, whether it trains models on it, how long it keeps it, who on its team can see it, how quickly it tells you about a breach, and what happens to the data when you leave.

Try Clara free →

Clara in your practice

Your clinical judgment stays in charge.

Clara prepares a session draft for you to review, adjust, and decide what belongs in the record.

Try Clara free