Disclaimer: Not medical advice, and not a substitute for clinical judgement. Confirm a tool’s regulatory status, such as FDA clearance or CE/UKCA marking, and how it handles patient data under HIPAA, GDPR or your local equivalent, before clinical use.
Some links in this guide are affiliate links, we may earn a small commission if you sign up, at no extra cost to you. Our recommendations are based on independent review; affiliate relationships do not influence which tools we cover or how we rank them.
Table of Contents
Where AI Belongs in Therapy: the Back Office, Not the Chair
AI has found a clear, useful place in therapy, and it is not the therapy itself. It is the paperwork. AI tools now draft progress notes, organize intake, and handle scheduling well enough to give clinicians real time back. Where AI has not earned trust, and where the law now agrees, is in the clinical work: diagnosing, treating, and being the thing a client experiences as care. In 2025 and 2026, several US states made that line legally binding, and the American Psychological Association put it plainly: AI “is not a safe or effective replacement for a qualified mental health provider” (APA, 2025).
This guide follows that line. First the administrative tools that safely save you hours, chosen for behavioral health specifically, then the boundary you should not cross, and how to adopt any of it in a HIPAA-compliant way. It fits within the wider AI healthcare landscape, alongside the parallel guides for doctors and nurses, but mental health carries risks that general medicine does not, so the bar here is higher.
The Clinical Bottom Line
- AI’s proven, safe role in 2026 is administrative: notes, intake, and scheduling. Even the new state laws that restrict AI therapy explicitly permit this use (Illinois IDFPR, 2025).
- The evidence for AI documentation is still thin: one small peer-reviewed trial (n=47). The “saves 70% of admin time” figures are vendor marketing, not independent research.
- AI as a therapist is now restricted. Nevada and Illinois barred it in 2025, more US states followed in 2026 (other countries are moving too), and over three-quarters of psychologists report patients bringing AI into their mental health (APA, 2026).
- Best move: use AI for the paperwork under a Business Associate Agreement, keep every clinical judgment human, and tell clients when AI touches their data.
The Admin Load AI Actually Fixes
The case for AI in a practice starts with the paperwork, because that’s what pushes clinicians out of the room. In the APA’s 2024 Practitioner Pulse survey, 62% of psychologists who had left or avoided insurance networks cited administrative burden, things like prior authorizations and audits, as a reason (APA, 2024). Notes are the daily version of that load: written after hours, essential to good care, and easy to fall behind on.
This is exactly where AI helps without touching the therapy. A tool that turns a recorded session into a structured draft note, which you then review and correct, removes busywork while keeping you in charge of the record. The key words are “draft” and “review”: the clinician stays accountable for every word that lands in the chart.
The Best AI Documentation Tools for Therapists
The tools built for behavioral health specifically are the ones worth your time, because they understand therapy note formats and clinical language. General medical scribes like DeepScribe and Augmedix (now part of Commure) can transcribe a session, but they lack behavioral-health fields such as mental status exams and risk documentation, so treat them as general-medicine tools adapted to psychiatry, not purpose-built for it.
| Tool | What it does | Standout |
|---|---|---|
| Mentalyc | AI progress notes in 100+ formats | Broadest template set (DAP, SOAP, BIRP, EMDR) |
| Upheal | AI-native EHR with notes and session analytics | Talk-ratio and sentiment insights, built-in dual consent |
| Blueprint | Notes plus symptom and outcome tracking | Free EHR, you pay only for AI use |
| Eleos Health | Ambient scribe and progress notes | The only option with a published clinical trial |
| Yung Sidekick | Progress and SOAP notes | Low cost (around $40/mo), CBT and EMDR focus |
| Twofold Health | Scribe for therapists and social workers | Structured note types (DAP, SOAP, BIRP, GIRP) |
One honest caveat on the evidence. Despite bold marketing, independent proof that these tools improve care is thin. The strongest study is a 2023 randomized trial of Eleos that found better symptom outcomes than treatment as usual, but it involved just 47 clients at a single clinic and was run by authors affiliated with the company (JMIR, 2023). Every “cuts documentation time by 70%” claim you’ll see is the vendor’s own number, not independent research. The time savings are real for many clinicians, but adopt on a trial, measure your own hours, and don’t take the marketing at face value.
Practice Management and Research Tools
Beyond notes, two categories round out a safe AI stack. For running the practice, SimplePractice and TheraNest handle scheduling, reminders, billing, and records, with AI layered into the admin flow. For staying current, research assistants like Elicit and Consensus search and summarize the literature, and a dedicated AI summarizer can compress a long clinical paper into its key findings.
A different category worth knowing is session-quality analytics. Lyssn scores recorded sessions for fidelity to methods like CBT and motivational interviewing, giving supervisors and clinicians structured feedback. Used for training and reflection, it supports the clinician rather than replacing judgment, which keeps it on the safe side of the line.
The Line You Can’t Cross: AI as Therapist
Here is where a 2023-era tool roundup becomes dangerous if you follow it into 2026. Consumer chatbots once recommended as “client support between sessions” have not aged well. Woebot shut down its app in 2025, Ginger merged into Headspace Health back in 2021, and Replika, an AI companion never designed for therapy, now faces an FTC complaint over deceptive health-adjacent claims. Steering a client toward any of them as a stand-in for care is exactly what regulators and clinicians are now warning against.
The clinical evidence is the reason. A 2025 Stanford study presented at the ACM FAccT conference found that leading AI chatbots expressed stigma toward conditions like schizophrenia and gave frankly dangerous responses, in one test, answering a person who had just lost their job and asked about tall bridges by listing bridge heights, missing the suicidal signal entirely. Bigger, newer models did not fix it (Stanford HAI, 2025). Real-world harm has followed too: in January 2026, Character.AI and Google settled lawsuits tied to teen suicides linked to chatbot use (CNN, 2026).
Patients are bringing this into your office whether or not you introduce it. In the APA’s 2026 survey of 1,242 psychologists, more than three-quarters had patients discussing AI for diagnosis, support, or companionship. Thirty-nine percent reported patients self-diagnosing with AI and a third saw patients using it to assist their treatment, while some clinicians reported patients developing dependency on chatbots and, in a smaller share, distorted thinking (APA, 2026).
Lawmakers moved quickly in response. Nevada barred AI from delivering mental or behavioral health care in June 2025, and Illinois followed in August with the WOPR Act, which prohibits AI from therapeutic decisions and client interaction while explicitly permitting administrative and supplementary use by licensed clinicians (Illinois IDFPR, 2025). By 2026, legal trackers counted at least seven states with restrictions, with Utah taking a lighter disclosure-only approach rather than a ban (Holland & Knight, 2026).
The takeaway is not “avoid AI.” It’s that the same laws restricting AI therapy carve out a clear, legal space for the administrative tools above. These particular statutes are US, so clinicians elsewhere should check their own regulator and data-protection law, but the principle, and professional bodies like the APA that state it, holds across borders. Stay in that admin space and you’re on solid ground anywhere.
A Safe-Use Boundary for Your Practice
The simplest way to stay compliant and ethical is to sort every possible AI use into two columns and never let the right column drift into the left.
| Let AI handle it (with a BAA) | Keep it fully human |
|---|---|
| Drafting progress notes for your review | Diagnosis and treatment decisions |
| Scheduling, reminders, and billing | Crisis and suicide-risk assessment |
| Organizing intake forms and history | The therapeutic relationship itself |
| Transcribing sessions (with client consent) | Interpreting what a client actually means |
| Summarizing research literature | Anything a client experiences as care |
If a use sits in the right column, no tool is compliant enough to make it safe. If it sits in the left, the question is simply whether the vendor meets your privacy and security bar.
Adopting AI Safely: HIPAA, BAAs, and Client Consent
The exact framework depends on where you practice, HIPAA in the US, GDPR in the EU and UK, PIPEDA in Canada, and equivalents elsewhere, but the checks below apply under all of them. Before any AI tool touches protected health information, three are non-negotiable. First, confirm the vendor meets your jurisdiction’s health-data standard and will sign the appropriate agreement (a Business Associate Agreement under US HIPAA, a data processing agreement under GDPR); without it, using the tool with client data is a violation, no matter how good the tool is. Second, understand how the vendor stores, retains, and trains on your data, and whether recordings can be deleted on request. Third, get informed client consent, especially for any session recording or transcription, and be transparent that AI is involved.
Match the tool to the practice, too. A solo clinician is also running a small business, and may want an all-in-one like SimplePractice with AI notes attached (our guide to AI tools for small business covers the rest of that admin load), while a group practice might pair a dedicated scribe with an existing EHR. Whatever you choose, adopt one tool at a time, measure the hours it actually saves, and keep the human relationship at the center. AI should clear your desk so you can do the work only a person can do.
FAQs on AI Tools for Therapists
Can AI replace therapists?
No. The APA states that AI is not a safe or effective replacement for a qualified mental health provider, and several states have made AI-delivered therapy illegal. AI’s appropriate role is administrative support, not clinical care.
Is it legal to use an AI therapy chatbot with clients?
In a growing number of US states, no. Nevada and Illinois barred AI from delivering therapy in 2025, and more followed in 2026. Outside the US the rules vary, so check your national regulator, but the direction is similar. Everywhere, these laws still allow AI for administrative tasks, so a note-drafting or scheduling tool is fine; an AI standing in as the therapist is not.
What are the best AI tools for therapy notes?
Behavioral-health-specific tools like Mentalyc, Upheal, Blueprint, Eleos Health, and Twofold are built for therapy note formats and clinical language. Choose one that signs a BAA, then trial it and review every note before it enters the record.
How do I confirm a tool is HIPAA-compliant?
Ask the vendor directly for a signed Business Associate Agreement and their security documentation (look for SOC 2). Confirm data storage, retention, and deletion practices, and whether your data is used to train models. No BAA means it isn’t safe for protected health information.
Should I tell clients when I use AI?
Yes. Transparency is both an ethical expectation and, in some states, a legal requirement. Explain what the tool does, get consent before recording or transcribing a session, and let clients opt out.
Sources and Further Reading
- American Psychological Association, Psychologists say patients are turning to chatbots (2026): survey of 1,242 psychologists on patient AI use.
- American Psychological Association, Using generic AI chatbots for mental health support (2025): the APA advisory and FTC engagement.
- Illinois IDFPR, Illinois prohibits AI therapy (WOPR Act) (2025): the admin-permitted, therapy-barred split.
- Holland & Knight, States continue efforts to regulate AI in healthcare (2026): the multi-state legal landscape.
- Stanford HAI, Exploring the dangers of AI in mental health care (FAccT, 2025): chatbot safety failures.
- CNN Business, Character.AI and Google settle teen-suicide lawsuits (2026): documented chatbot harm.
- APA, 2024 Practitioner Pulse Survey: administrative burden on psychologists.
- JMIR, RCT of an AI platform for behavioral interventions (2023): the one peer-reviewed outcomes trial (n=47).
Richard Johnson
Richard Johnson is an AI specialist at one of the world's largest technology companies, where he has spent the past three years helping organizations adopt AI. CognitiveFuture extends that work publicly: gathering the available evidence on each tool, from vendor documentation to independent reviews and user feedback, and cutting a crowded market down to the right choice for the job in front of you.