Free AI for psychologists: what you can try, what its limits are, and when you need more control

Artificial intelligence has become a part of the daily routine in many psychology practices. And as with any new tool, the first thing most of us want to do is try it out for free to see if it's a good fit.

There's good news: the free versions of major general-purpose AIs (ChatGPT, Claude, Gemini, Perplexity) allow for many useful tasks, including in a clinical setting. And there's also bad news: there are things you should not do with those same tools, especially when patient data is involved.

In short: free AI is useful for organizational tasks, drafting, training, or informational material, but it should not be used with identifiable patient data or for making clinical decisions. For that, you need an AI designed for clinical use, with encryption and privacy guarantees. This post explores safe uses, real limitations, and where the line is drawn between "this is okay" and "this is better avoided."

Free AI tools you can try

The most common options, all with free versions:

• ChatGPT (OpenAI). General-purpose assistant with good capabilities for writing, organization, and brainstorming.
• Claude (Anthropic). General-purpose assistant focused on long texts, nuance, and reasoning. A good option for writing.
• Gemini (Google). Assistant integrated with the Google ecosystem. Useful if you work extensively with Docs or Gmail.
• Perplexity. AI-powered search engine. Instead of chatting, it provides answers with cited sources. Useful for quick research.
• NotebookLM (Google). Allows you to upload your own documents and consult them as if they were a knowledge base. Designed for study and organization, not for sensitive data.

All of them have paid versions with more features and, generally, better privacy policies. The free version is sufficient to start exploring.

What you CAN safely use free AI for

There are uses where free AI provides real-time benefits without compromising sensitive information. They all share one characteristic: they do not require entering identifiable patient data.

• Writing content for your website, blog, or social media. Drafts of posts, service descriptions, biographies, educational content.
• Generic psychoeducational material. Summaries of a therapeutic approach, explanations of a concept, psychoeducational handouts for patients (always reviewed by you).
• Organizing your own schedule and tasks. Lists, priorities, weekly planning, summaries of your day.
• Study and training. Summarizing articles, comparing approaches, generating review questions, translating bibliography.
• Internal communication for the center. Drafts of emails between colleagues, coordination templates, meeting minutes.
• Marketing and online presence. Ideas for reels, captions, newsletters, email subject lines.

A good rule of thumb: if what you're typing into the chat could be read by anyone without compromising a specific patient, then the use is reasonable.

What you should NOT use free AI for

Here's the red line. There are uses that seem tempting but are not appropriate with free general-purpose tools:

• Uploading session notes, transcripts, or recordings with identifiable patient data. Even if you delete the name, other data (age, profession, context) can allow for re-identification.
• Requesting diagnoses or clinical decisions. Generalist AIs are not designed for clinical diagnosis, do not have access to the patient's complete medical picture, and can fabricate information that appears reliable.
• Processing questionnaires completed by real patients. Like notes, they contain protected personal data.
• Sharing signed documents. Informed consents, clinical reports, invoices with personal data.
• Generating final clinical reports to be given to the patient without very strict professional control over the content.

The reason is twofold: privacy (free versions often use your conversations to train future models, unless you activate specific options) and clinical reliability (a generalist AI does not replace professional judgment or know your patient).

The Real Limitations of Free Tools

There are three technical limitations to be aware of before incorporating free AI into your workflow:

1. Privacy and training. Free versions from most providers reserve the right to use your conversations to improve their models. While some allow you to disable this option, assuming otherwise is the prudent approach.

2. No traceability or access control. If you work in a center with multiple therapists, a generalist AI does not have roles, permissions, or a record of who has accessed what. This is incompatible with GDPR for health data.

3. No integration with your history. Generalist AI doesn't know your patients' context. Every time you use it, you start from scratch. This means that real time savings in clinical tasks are limited: you have to input the context every time.

When you need AI designed for your practice

Free tools are a good starting point for exploration. However, beyond a certain volume of use, limitations become apparent: you can't input real data, you have to copy and paste context every time, and the responsibility for privacy rests entirely with you.

AI integrated into a clinical management platform addresses these three points. In the case of MIA, Eholo's AI, operates within a closed, encrypted environment, does not train models with clinical data, and works directly with the patient's real medical history. This means it can:

• Transcribe sessions and generate clinical history drafts in seconds.
• Search for invoices or summarize payment status without you having to open multiple modules.
• Identify debtors and assist with collections communication.
• Support marketing and management tasks, such as we saw in this post about AI applied to practice.

If you want to learn more before trying it, you can download the basic AI guide for psychologists, with specific criteria for incorporating it into your practice.

Frequently Asked Questions

Is it legal to use free AI in a psychology practice? Yes, as long as no identifiable patient data is entered. GDPR prohibits transferring personal health data to a third party without a legal basis and adequate technical safeguards, which free generalist AI versions do not offer.

Is free ChatGPT safe for writing session notes? Not with identifiable data. Yes for generic drafts, templates, or psychoeducation, without real patient information. For actual notes, an integrated tool within the clinical record is needed, with encryption and without using data for training.

Can AI diagnose a patient? No. Generalist AIs can offer hypotheses based on text, but they don't have access to the full picture, don't know the patient's life context, and can generate seemingly confident answers without real basis. Diagnosis is the professional responsibility of the psychologist.

What's the difference between ChatGPT and AI integrated into management software? ChatGPT is a general-purpose AI. AI integrated into management software, like MIA in Eholo, operates within a closed environment, works with real patient history with privacy guarantees, and is specifically designed for consultation tasks: session transcription, history drafts, invoice search, or payment management.

How to start using AI in practice without compromising data? Start with uses that don't involve patient data: content writing, task organization, training. When you want to integrate it into the clinical workflow, do so with a tool designed for practice and a clear protocol for what is entered and what is not.

The next step

Free AI is a good starting point to understand its benefits and limitations. When it's time to incorporate it into real clinical work, it's advisable to switch to a tool designed for practice.

See how MIA helps save time in Eholo or download the basic AI guide for psychologists