Best AI Tools for Therapists in 2026

How AI scribes work — and what to expect

An AI scribe for therapists works in three stages. First, it captures session audio — either through your phone microphone for in-person sessions, or through your computer's audio for telehealth calls. Second, it transcribes that audio into text and runs it through a clinical language model trained on therapy session content. Third, it outputs a structured clinical note in whatever format you selected — SOAP, DAP, BIRP, or a custom template.

The full process takes 30 to 90 seconds after the session ends. In practice, by the time you have seen your client out and returned to your desk, the note is ready for review and editing.

The editing step matters. Every AI-generated note needs clinical eyes on it before signing. In testing across multiple platforms, the proportion of output requiring substantive editing drops from roughly 20% in the first week to under 10% by week three — the tools learn from your corrections and the format consistency of your session style. But "AI-generated, clinician-reviewed and signed" is the correct mental model. These are drafting tools, not autonomous documentation systems.

The net time saving: 45 to 75 minutes per day for a therapist seeing 6–8 sessions. That accounts for review time, EHR entry, and occasional re-generates. For a full-time private practice, it is the equivalent of reclaiming one full working day per week.

What HIPAA compliance actually means for AI tools

Every AI scribe that processes clinical audio operates at the intersection of two legal frameworks: HIPAA's Privacy Rule (which governs who can access Protected Health Information) and HIPAA's Security Rule (which governs how PHI must be protected in digital systems). Both apply.

Business Associate Agreement. Any vendor that handles PHI on your behalf must sign a BAA. This is not optional — it is a legal requirement. Before using any AI documentation tool with real session audio or notes, verify that the vendor offers a BAA and sign it. All of the tools reviewed here provide BAAs on paid plans. Heidi Health, Berries, and Scribeberry provide BAAs on their free tiers too.

SOC 2 certification. HIPAA compliance can be self-declared. SOC 2 Type II cannot — it requires an independent third-party audit. The tools in this review with SOC 2 Type II certification: Mentalyc, Blueprint, Heidi Health, Freed AI, Scribeberry. AutoNotes and Supanote AI are HIPAA compliant but have not confirmed SOC 2. For practices with institutional compliance requirements, SOC 2 matters.

Data retention and deletion. HIPAA requires you to know how long PHI is retained and who can access it. All reviewed tools process session audio and delete it after note generation. The generated note text is retained per your account settings, which is appropriate since it is a clinical record.

Model training. None of the reviewed tools train their AI models on user session data or PHI. This is important because some general-purpose AI tools do use submitted content for model improvement by default. Purpose-built clinical tools have explicit contractual prohibitions on this.

The practical checklist before going live with any AI scribe:

1. Confirm a BAA is available and sign it before processing any session audio
2. Add the AI tool to your Privacy Notice as a business associate
3. Get explicit client consent for AI-assisted documentation — verbal consent documented in the chart, or a written consent addendum
4. Know the vendor's data deletion policy and verify it matches your retention obligations

How to choose the right AI scribe

Step 1 — Define what you actually need

Before comparing pricing or features, answer these three questions:

Do you need an all-in-one platform or a documentation add-on?

• All-in-one (EHR + scheduling + billing + AI notes): Blueprint is the strongest option — free EHR with pay-per-session AI. Upheal bundles telehealth, scheduling, and AI notes. DeepCura adds automated billing and a 24/7 AI receptionist for group practices.
• Documentation add-on to an existing EHR: Mentalyc, Heidi Health, AutoNotes, Freed AI, Berries, Supanote AI, and Scribeberry all layer on top of whatever system you already use.

Do you need native EHR integration or is copy-paste acceptable?

• Native EHR push (no copy-paste): Mentalyc (SimplePractice, TherapyNotes, Jane App), Supanote AI (SimplePractice, TherapyNotes, Valant, Tebra, ICANotes), Scribeberry (Epic, Jane, Accuro, Oscar Pro, TELUS), DeepCura (Epic, athenahealth, eClinicalWorks, and 5 more).
• Copy-paste or browser extension: AutoNotes, Heidi Health, Berries, Freed AI — notes are generated and exported manually.

What note formats do your documentation requirements call for?

• Therapy-specific formats (DAP, BIRP, GIRP, EMDR): Mentalyc has the widest native format library. Supanote AI and Berries also support therapy-specific formats natively.
• SOAP with custom templates: Freed AI and Scribeberry both do SOAP well with template customisation.
• Flexible/customisable: Heidi Health's template builder is highly flexible with no fixed format constraint.

Step 2 — Match pricing model to session volume

Step 3 — Factor in compliance requirements

If your practice operates under institutional compliance requirements — hospital affiliation, CCBHC certification, Medicaid oversight — Eleos Health is the only tool in this review built for that context.

The tools at a glance

What the research says about documentation burden in therapy

Therapists in independent practice spend an average of 15–25% of their working hours on clinical documentation — progress notes, treatment plans, intake assessments, and discharge summaries. For a full-time therapist seeing 35 sessions per week, that is seven to nine hours per week on paperwork.

The downstream effects are well-documented in clinical literature: therapist burnout, reduced session availability, delayed note completion (which creates both compliance risk and clinical risk when notes are written days after the session from memory), and reduced time for consultation and continuing education.

AI scribes do not eliminate documentation — notes still require clinical review and signature, and the therapist is still legally responsible for the accuracy of the clinical record. What they eliminate is the mechanical drafting task: translating session content into structured clinical language in the required format. That specific sub-task is what consumes the majority of documentation time, and it is the part AI is genuinely good at.

Used correctly — record, generate, review, edit, sign — these tools represent a real improvement in the documentation side of clinical practice. They are not, and are not trying to be, a replacement for clinical judgment or the therapeutic relationship.