Using AI Responsibly in Your Practice: Notes, Scribes, and Ethical Guardrails
Most therapists in 2026 have at least considered it. The notes pile up. Documentation eats evenings. And then a colleague mentions that her AI scribe writes a competent progress note in under a minute, and you find yourself wondering whether you have been doing things the hard way for no reason.
AI-assisted documentation is no longer fringe. Tools like Mentalyc, Upheal, Blueprint, Eleos, and a growing field of competitors have moved into thousands of practices. Furthermore, most Electronic Health Records (EHRs) are now integrating AI as a standard feature, which often helps mitigate potential pitfalls associated with standalone tools. The early adopters are not wrong about the time savings. The harder question is not whether to use AI in your practice. It is how to use it without compromising the clinical, ethical, and relational standards that make your work yours.
This is a practical guide, not a position paper.
What These Tools Actually Do
Most AI documentation platforms fall into one of three buckets.
- Scribes listen to the session (with consent) and generate a draft progress note in your preferred format - SOAP, DAP, BIRP.
- Summarizers take your dictated or typed notes and clean them up.
- Clinical assistants go further, suggesting treatment plan goals, interventions, or risk language based on session content.
The risk profile rises as you move down that list. A tool that drafts a note you edit is different, ethically and clinically, from a tool that proposes what your treatment plan should say.
The HIPAA Questions That Actually Matter
Before you sign up for anything, three questions answer most of what you need to know.
Will the vendor sign a Business Associate Agreement (BAA)? If the answer is no, the conversation is over. A BAA is not optional for any tool that touches Protected Health Information (PHI).
Where is session data stored, who can access it, and how long is it retained? "HIPAA compliant" is not a single setting. It is a set of practices. Ask whether audio is recorded or processed in real time and then discarded, whether data is used to train the company's models, and what the retention default is. Set retention to the minimum your credential and state require.
What happens if you leave the platform? You need a clear, documented way to export or delete records. A vendor that cannot answer this cleanly is not ready for a clinical work application.
State licensing boards are increasingly issuing their own guidance on AI use and several have moved faster than federal regulators. Check yours. The answer in California is not necessarily the answer in Texas.
Informed Consent Is Not Optional
Clients have a right to know when AI is part of their care - even if it is "only" listening to write a note. Slipping AI into a session without disclosure is a consent failure, regardless of how convenient the tool is.
Your consent process does not need to be elaborate but it should be explicit. Cover what the tool does, what it captures, what it does not, who has access, how long data is kept, and the client's right to decline without it affecting their care. Many therapists now include AI language in updated intake paperwork and verbally name it in the first session. Both is better than either.
A useful test: if your client found out later about the AI tool you have been using, would they feel informed or surprised? If the honest answer is surprised, your consent process needs work.
Where AI Helps and Where It Overreaches
AI does well as a drafting tool. A scribe-generated progress note you read, correct, and sign is not unlike a transcribed dictation. You remain the clinician of record. The note is yours.
AI gets murky as a clinical tool. Treatment plan suggestions, risk assessments, "session insights," and recommended interventions are increasingly available and they can be useful prompts. However, they are not a substitute for your formulation. The moment a tool's output shapes your clinical decisions without your full critical engagement, you have stepped over the line.
AI is most dangerous as a replacement tool. Watching for missed material, summarizing what a client "really meant," or generating language for sensitive conversations risks outsourcing the very things that make therapy therapeutic. Use the tool to free your attention for the work. Do not let it do the work.
A Simple Evaluation Framework
When considering any AI tool for your practice, run it through four questions:
Does it save me time on tasks that do not require clinical judgment? If yes, that is a strong use case. If it is touching clinical reasoning, slow down.
Can I explain to a client, in plain language, exactly what this tool does? If you cannot, you cannot get informed consent.
Would I be comfortable defending this tool - and my use of it - to my licensing board? If there is hesitation, find out why before adopting.
Is the tool making me a better clinician or a faster one? Both are legitimate goals, but they are not the same goal. Know which one you are choosing.
Adopt Carefully, Not Reluctantly
AI is not going away, and neither is the documentation burden that drove most of us to consider it in the first place. The therapists who use these tools well tend to share a posture: curious, cautious, current on guidance, and willing to revisit decisions as both the tools and the standards evolve.
The goal is not to be early. It is to be deliberate.
TherapyCloud supports therapists navigating the realities of modern practice - including the rapidly shifting conversation around technology, ethics, and clinical craft. If you are weighing tools, reviewing consent language, or just looking for a community of colleagues thinking carefully about the same questions, you do not have to figure it out in isolation.
Sources: HIPAA Privacy and Security Rules (U.S. Department of Health & Human Services); American Psychological Association guidance on the use of AI in psychological practice; APA Ethics Code; state licensing board advisories on AI use in behavioral health.
