The AI Session Notes Boom: What Therapists Must Know in 2026
Two-thirds of therapists remain wary of AI documentation tools — while the other third has already adopted them. Here's the HIPAA compliance checklist, the consent problem that ended one therapy relationship, and what the top tools actually cost.
Why Two-Thirds of Therapists Are Still on the Sidelines
The American Psychological Association's 2025 Practitioner Pulse Survey — the profession's most recent large-scale measurement of AI adoption, published December 2025 and drawing on 1,742 clinician responses — found that 29 percent of practitioners now use AI tools at least monthly, more than double the 11 percent who reported doing so in 2024 (American Psychological Association, 2025). Fifty-six percent have used AI in their practice at least once.
That growth is real and it is accelerating. But the same survey found that 67 percent of therapists remain concerned about potential data breaches. That concern is not irrational. Ten platforms now compete seriously in the therapy AI documentation space — Twofold Health, Upheal, Mentalyc, Supanote, Freed AI, Soapnote.ai, DeepScribe, Therabyte AI, Abridge, and ChartNote among them — each making HIPAA claims in their marketing and each promising to reduce the documentation overhead that extends a 30-client week into a 45-hour one. The tools are arriving faster than the clinical guidelines for evaluating them.
Forty-two percent of survey respondents said AI can help reduce practitioners' administrative burden — up from 33 percent in 2024. The case for these tools is genuine. So is the case for proceeding carefully. This post covers both: what HIPAA actually requires of every practice that touches these tools, what the consent question means for the therapeutic relationship, and a plain-language breakdown of what the leading tools cost and what they demand of you clinically.
How These Tools Actually Work
Three input methods define the market:
Ambient recording keeps a microphone open throughout the session. After the client leaves, the recording is transcribed by the AI and used to generate a structured note draft. Most tools delete the audio immediately after processing; the note is what persists in the system. This is the mode with the highest documentation fidelity and the most complex consent implications.
Post-session dictation lets you summarize what happened after the client has gone, speaking your own version of events, then transforms that summary into a formatted clinical note. Lower fidelity to session specifics, but no live recording and no ambiguity about whether the client was aware they were being captured.
Audio upload accepts a recording you made through your own means and returns a note draft. Less immediate, but useful for practices that already record sessions for supervision or other clinical purposes.
The output in all three cases is almost always a structured progress note: SOAP (Subjective, Objective, Assessment, Plan), DAP (Data, Assessment, Plan), or BIRP (Behavior, Intervention, Response, Plan), depending on your license board requirements and EHR preferences.
The EHR integration question matters more than most review articles acknowledge. Standalone tools — Freed, Mentalyc, Upheal — generate note drafts that you review and then paste or push into your practice management system. EHR-native tools like SimplePractice Note Taker sit inside the platform you're already using and write directly into the patient chart. SimplePractice powers its Note Taker through Claude, Anthropic's AI system, and as of 2026 has extended that coverage beyond telehealth to in-person sessions, dictation, and uploaded audio.
Neither architecture is universally superior. Native tools are frictionless inside a single platform. Standalone tools work across any EHR and typically offer more output customization — but nothing ports if you ever switch platforms. For a deeper look at how therapists are deploying AI across the full admin stack, the tradeoffs between documentation tools and broader practice management AI are worth weighing before committing to either.
The HIPAA Requirement Most Practices Miss
The phrase "HIPAA compliant" appears in the marketing materials of every tool on this list. It is not a certification, and it is not a guarantee. There is no government agency that certifies software as HIPAA compliant. HIPAA places requirements on covered entities — that is, you and your practice — and specifically on how you handle protected health information (PHI), including what contractual safeguards you must obtain from the vendors you involve in that handling.
The mechanism is the Business Associate Agreement (BAA). Under HIPAA, any vendor that processes, stores, or transmits PHI on your behalf is a "business associate," and you are required to have a signed BAA with them before that processing begins. Using a tool without one is a HIPAA violation regardless of what that tool's marketing says.
The APA's Ethical Guidance for AI in Professional Practice of Health Service Psychology, released in June 2025 and updated in December 2025, is explicit on when PHI begins: "AI-generated clinical summaries are considered protected health information from the moment of creation." Many therapists assume the note becomes PHI only after it enters the chart. Under the APA's guidance — which tracks the plain reading of the Privacy Rule — the transcription itself generates PHI. The moment ambient audio is being processed by the vendor's system, the BAA requirement is already in effect.
As of May 2026, BAAs are automatically provided on paid tiers for Freed, Mentalyc, Twofold, Berries, and JotPsych. Free tiers at several vendors do not include BAAs — using a free tier in clinical practice is not HIPAA compliant, irrespective of how the tool is marketed.
Before adding any AI notes tool, four things to verify:
- Locate the actual BAA document or BAA workflow — not the "HIPAA compliant" badge, the signed agreement.
- Confirm the BAA covers all the features you intend to use; some vendors' BAAs cover the core tool but not add-ons or integrations.
- Review the data retention policy: how long is audio stored before deletion, who at the vendor can access it, and where is it processed geographically.
- Confirm in writing that the vendor does not use your session recordings or transcriptions to train its AI models.
The HIPAA requirements that apply to therapist websites are a related layer — if a tool also generates content for your website or marketing, the same PHI boundary questions arise. And if you haven't reviewed how the 2025 HIPAA Security Rule overhaul tightened obligations around encryption and access controls, that context matters for any third-party tool holding clinical data.
The Consent Problem That Ended One Therapy Relationship
In May 2026, NPR published a two-part investigation into the expanding use of AI recording tools in therapy. The reporting introduced Molly Quinn, a 31-year-old librarian from Fayetteville, Arkansas. Quinn had signed her practice's standard intake paperwork, which included a brief clause about AI documentation tools. No one had spoken with her about it directly. After she later learned that her sessions had been recorded without a verbal discussion, she did not return to that practice (NPR, May 2026).
A bioethics researcher at New York University, interviewed for the same reporting, named the gap between paperwork and consent clearly: consent forms alone do not produce informed consent. "Folks scroll through these forms without reading them," the researcher said. The recommendation from ethicists and from the APA's June 2025 ethical guidance converges on the same standard: AI documentation tools should be disclosed through direct verbal conversation, not simply included in intake paperwork.
What that conversation should cover, per the APA guidance:
- What is recorded, and in which session mode (live ambient, post-session dictation, or upload)
- How the recording or transcript is processed, and by which vendor
- Who can access the generated note and the raw transcript
- How long the audio is retained before deletion
- That the client can decline, and what the documentation process looks like if they do
There is also a legal layer beyond HIPAA: several states have two-party recording consent laws requiring that all parties agree before a conversation is recorded. In those states, the ambient recording mode of any AI notes tool requires explicit prior consent — not just a clause in a form, but acknowledgment from the client that they understand the session will be captured.
The therapeutic relationship is the mechanism through which therapy works. A client who learns that their sessions were monitored without a genuine, direct conversation has experienced a rupture in that relationship, even if the intake paperwork was technically compliant. The direction the field is moving — and that the APA guidance reflects — treats the consent conversation as a clinical matter, not an administrative checkbox.
The Major Tools Compared
Pricing in this market shifted considerably through 2025 and 2026 as the leading tools moved away from freemium models toward structured subscription pricing. The table below reflects current pricing as of September 2026.
| Tool | Pricing (monthly) | BAA on paid plans | Best for |
|---|---|---|---|
| Freed.ai | $39 (40 notes), $79 (unlimited), $119 (EHR push + CPT/ICD-10 coding) | Yes | Multi-specialty practices; expanded to AI phone receptionist (scheduling, 90+ languages) in April 2026 |
| Upheal | $1/session, capped at $69/mo per provider | Yes | Mental health-specific; usage-based model benefits lower-volume solo practices |
| Mentalyc | $19.99 (40 notes), $39.99 (100), $69.99 (160), $119.99 (330) | Yes | Therapy specialists; tiered by session volume; strong template library for clinical formats |
| SimplePractice Note Taker | Add-on to SimplePractice subscription (Claude-powered) | Via SimplePractice BAA | Practices already on SimplePractice; notes write directly to chart, no copy-paste step |
For practices on SimplePractice, the native Note Taker is the path of least friction — one BAA, one workflow, one system. The limitation is portability: if you ever move off SimplePractice, the tool stays behind. For practices using TherapyNotes, Jane App, or a lighter EHR, a standalone tool gives you notes you can move anywhere.
Freed has expanded its footprint beyond documentation: its April 2026 AI receptionist feature handles scheduling and inbound calls in over 90 languages, which positions it as a broader operations tool for practices that want to consolidate AI vendors. That breadth may matter if you're evaluating AI as a practice-wide decision rather than a documentation-only one. For a direct comparison of SimplePractice versus purpose-built practice platforms, the question of which integrated features are load-bearing versus which are convenience is worth examining.
Who Owns the Note?
The APA's June 2025 guidance establishes the accountability framework plainly: "The signing clinician is responsible for everything in the clinical record, regardless of how it was drafted." An AI-generated note does not distribute liability toward the vendor. Your licensing board's expectation — that what is in the chart is accurate, clinically defensible, and represents your professional judgment — applies identically whether the first draft came from your own pen or from a language model.
The practical implication is that AI-generated notes require genuine clinical review, not a scroll-and-approve workflow. Language models make errors. In a general productivity context, an AI error costs a few minutes to fix. In a clinical record, it can affect a client's ongoing care, an insurance determination, a disability proceeding, or a licensing board investigation. The errors that carry the highest risk are subtle ones: a symptom attributed to the wrong session, a clinical impression that inverts what was meant, a medication reference that doesn't match what the client actually reported.
This doesn't argue against using these tools — it argues against deploying them in a way that uses their speed advantage to compress rather than improve the review step. The goal is a better-quality note in the same time, not a rubber-stamped draft generated in 30 seconds. The tools that make this realistic produce note formats the clinician already knows how to review, integrate into the EHR where chart context is visible, and delete recordings after processing so the note is the only artifact that persists.
The APA guidance also notes that several states have enacted or are considering transparency laws requiring that AI use in clinical documentation be disclosed to clients on an ongoing basis — not just in the initial consent conversation. Checking the current requirements in your state's licensing board guidance is advisable before deploying any of these tools.
What the Admin Time Savings Actually Buy
Therapists who have adopted AI notes consistently report the same reallocation: the hours previously spent writing or dictating notes in evenings or between sessions are available for something else. For solo practitioners, that time typically returns as either additional clinical availability, professional development, or the practice-building activities that don't happen when documentation extends into every non-session hour.
AI documentation tools address one major source of administrative overhead — what happens after the session ends. They don't address intake conversations, the question of how a prospective client found the practice, what made that person choose to reach out rather than scrolling past, or what happens when someone visits a therapist's website at 11pm and can't leave a voicemail that gets returned for three days.
That front-door problem is where the marketing work happens. A practice that has automated its post-session documentation now has the capacity to think about it. The combination of tools that has emerged in private practice in 2026 tends to look like AI documentation for the back end of the session, and an AI visitor agent on the practice website for the front door — the real-time intake conversation that converts a late-night visit into a scheduled consultation without requiring the therapist to be available after hours. See how that works in practice.
The APA survey data implies that most of the therapists who haven't adopted AI documentation tools yet are watching the market rather than opposed to it. The compliance framework above — BAA, consent conversation, clinical review — is what separates a thoughtful adoption from a rushed one. The tools are maturing. The professional standards are catching up. The gap between the two is where practices that move carefully are finding meaningful competitive ground.