5 Best AI Scribes for Therapists in 2026

1. Mentalyc: best for a therapy-only documentation and insight platform

Mentalyc deserves the first place on a therapy-specific shortlist because behavioral health is the product's center, not a secondary template category. Its official FAQs describe support for individual, couples, family, child, and PMHNP workflows; editable templates; psychiatric evaluations and medication-management layouts; and a Chrome extension that inserts finished notes into web-based EHRs. That breadth matters when a practice uses more than one service type or wants to connect progress across sessions.

Its published privacy answers are unusually detailed for comparison work. Mentalyc states that a signed BAA can be generated from account settings, session data is not used to train models, and session audio is processed and deleted under the retention behavior it describes. Read the complete wording rather than compressing it into “audio is never stored”: the same FAQ explains temporary audio handling for note generation and separate behavior for some features. That distinction is exactly why practices should inspect the actual data flow instead of relying on a badge.

Choose Mentalyc when you want a behavioral-health ecosystem with therapy-specific templates, treatment-plan continuity, and a browser-based EHR handoff. Test whether its insights and level of detail match your documentation policy. A feature that can surface more session content is not automatically better if the clinical record should remain concise.

Verify it: Mentalyc official FAQs.

2. Upheal: best for therapists considering an ai-native mental-health ehr

Upheal is the most distinct option in this list because it is positioning itself as a mental-health EHR with an AI assistant, not merely a note generator. Its current product and pricing pages describe AI notes, treatment plans, customizable note sections, scheduling, forms, a client portal, telehealth, and migration support. For a solo or growing practice already reconsidering its EHR, evaluating those workflows together can be more useful than buying a separate scribe and then solving integration later.

The trade-off is scope. If the practice is satisfied with its EHR and only wants a draft note, an all-in-one platform may introduce more configuration, migration, and governance work than necessary. Conversely, if scheduling, forms, video, and documentation are fragmented, the broader platform may remove handoffs a standalone tool cannot. Upheal currently advertises usage-based pricing capped monthly, but verify the current US plan and what counts as a session before budgeting because the company has changed pricing models.

During a trial, test both the note and the system around it: client intake, consent capture, telehealth, template behavior, corrections, signing, export, and what happens when you leave the platform. Confirm how existing records migrate and how you retrieve them later. The best all-in-one experience is still a poor fit if it makes record access or termination difficult.

Verify it: Upheal official pricing and feature page.

3. Twofold: best for a simple cross-device behavioral-health workflow

Twofold is a strong generalist candidate when the practice wants to start quickly without adopting a new EHR. Its official behavioral-health page lists DAP, SOAP, BIRP, GIRP, and PIRP templates, and its main site describes in-person, telehealth, post-visit dictation, typed notes, audio upload, desktop, and mobile workflows. The published personal plan is $69 monthly or $49 per month when billed annually, with a seven-day trial; verify the checkout total and current offer because promotional pricing can appear on the site.

Twofold's security page states that it signs a BAA with healthcare organizations using the platform, does not use patient data for model training, processes audio to create the transcript and note, and does not retain the recording on its servers. Those are specific, useful procurement answers. They remain vendor statements, so a practice handling PHI should review the BAA, subprocessor list, retention controls, and exact account terms rather than treating this summary as due diligence.

Shortlist Twofold when device flexibility and a low-friction browser workflow matter. In testing, pay attention to whether its default behavioral-health notes preserve the difference between client report, observed behavior, clinician assessment, and plan. Also time the last step into your EHR: a fast draft can lose its advantage if formatting or field placement requires repeated repair.

Verify it: Twofold official behavioral-health and security pages.

4. Heidi: best for a flexible free starting point across clinical specialties

Heidi belongs on the list because its free plan currently includes unlimited AI documentation using standard templates and its product spans many clinical specialties and languages. That makes it a practical baseline: a therapist can learn what ambient drafting changes about the session without committing to a paid plan. Its paid Clinician tier adds advanced templates and other assistant features, while team capabilities sit higher in the product structure.

Heidi's official transcription and safety pages state that data is encrypted in transit and at rest, audio is not stored after processing, and the service aligns with several security and privacy standards. The company also emphasizes that the clinician reviews every note. A purchaser should still confirm US BAA availability for the exact plan and entity, retention of transcripts and notes, subprocessor terms, and whether advanced template controls needed by the practice require a paid tier.

Choose Heidi when a free, multi-specialty tool and language coverage outweigh the need for a therapy-only environment. Compare its output with the same session in a behavioral-health specialist. Look especially for unnecessary narrative, unsupported normal findings, speaker confusion in couples or family work, and whether your preferred DAP or BIRP structure remains stable after customization.

Verify it: Heidi official US pricing.

5. ClinicFrame: best for a focused desktop workflow for independent and small practices

ClinicFrame is the product behind this article. Its clearest fit is a clinician who wants one focused documentation application rather than a broader practice-management platform. The desktop app supports sessions in the room, telehealth system-audio capture without adding a meeting bot, and dictation after the encounter. DAP, BIRP, SOAP, and custom note templates let the same practice use different records for an intake, progress visit, or medication follow-up.

The main operational limitation is explicit: ClinicFrame does not currently offer direct EHR integration. The clinician reviews the draft, then copies the formatted note or exports it. That can be simpler for a solo practice with a stable workflow, but a larger organization should measure the copy step, wrong-chart risk, access governance, and the value of an integrated alternative. The right comparison is time to a correct note inside the system of record, not generation speed alone.

ClinicFrame states that a signed BAA is included with every account, on every plan, and that patient content is never used to train AI models, ours or any third party's. Because we publish this comparison, treat those statements the way you would any vendor's: read the agreement you are asked to sign and confirm the retention, subprocessor, and data-use terms before using PHI. Test ClinicFrame when its capture modes and note formats fit, and make contract review a release gate for real patient use.

Verify it: ClinicFrame therapy workflow.

Therapy-specific platform or general medical scribe?

A therapy-specific platform is usually the better starting point when behavioral health is the entire practice, the team relies on several mental-health note formats, or treatment-plan continuity and therapy-focused controls matter. A general medical scribe can be the better operational fit when the organization also includes prescribers, primary care, or other specialties; when one procurement and governance process must cover multiple teams; or when the clinician already has an approved custom template that produces a concise record.

Do not decide from the template menu alone. Ask each finalist to generate DAP or BIRP and SOAP from the same session. A strong product should change which information it emphasizes, not merely move the same generic summary under different headings. Also compare the amount of sensitive narrative retained, the handling of multiple speakers, the separation between observed behavior and client report, and the final handoff into the record. The best therapy tool is the one that reliably documents the clinical work without turning the chart into an unnecessary transcript.

1. Choose a therapy-specific platform when behavioral-health depth and treatment-plan continuity outweigh cross-specialty standardization.
2. Choose a general medical scribe when one tool must support therapy, medication management, and other clinical services.
3. Choose an all-in-one mental-health EHR only if scheduling, forms, telehealth, portal, and documentation need to move together.
4. Keep manual documentation or dictation available for encounters where consent, environment, risk, or session sensitivity makes ambient capture inappropriate.

What a therapy AI scribe must get right

A therapy note should support continuity, professional accountability, and the requirements that apply to the service without becoming a replay of the session. Under HHS guidance, separately maintained psychotherapy notes have a specific meaning and additional protections; ordinary progress notes in the medical record are not automatically psychotherapy notes. That is a record-design question, not just a template label. Review the HHS explanation of mental-health information and psychotherapy notes with your own counsel or compliance lead.

The draft should also preserve evidentiary boundaries. Client report, collateral report, observed behavior, clinician intervention, assessment, and plan are not interchangeable facts. A fluent model may join them into a sentence that sounds clinically reasonable but changes who said what. The test is not whether the prose sounds professional; it is whether a reviewer can trace each important statement to the encounter and whether unsupported conclusions are absent.

• Correctly separate reported mood from observed affect and documented risk assessment.
• Represent the intervention actually delivered and the client's response without embellishment.
• Connect progress and barriers to the treatment plan without inventing improvement.
• Keep unnecessary trauma, family, sexual, substance-use, or third-party detail out of the general record.
• Handle names, pronouns, medications, scores, dates, and quotations with explicit review.

How to run a fair three-session pilot

Do not compare products with one unusually clean demo. Prepare three representative, de-identified simulations or properly authorized encounters and reuse the same underlying facts: a routine individual progress visit, a complex intake, and the hardest session type your practice actually sees, such as telehealth, family work, or a medication follow-up. Configure each approved template before the trial so you are testing the intended workflow rather than a vendor default.

Score the final draft before you start editing, then measure the work required to make it record-ready. Count clinically meaningful corrections, not commas. An omitted safety action, wrong speaker, invented intervention, incorrect medication, or excessive sensitive detail should weigh far more than formatting. Repeat the trial after changing a template; some products improve substantially with configuration, while others remain generic.

1. Confirm consent and use only content your practice is authorized to process.
2. Run the exact in-person or telehealth device setup you will use in practice.
3. Score completeness, unsupported content, attribution, sensitive-detail control, and formatting.
4. Time review, correction, signing, and transfer into the EHR—not only note generation.
5. Record failures by session type and decide which encounters should remain dictation or manual documentation.

Privacy questions to ask before using PHI

Start with the BAA checklist for AI scribes, but do not stop there. Ask which legal entity signs it, whether it covers your exact plan during a trial, which subprocessors touch audio, transcripts, and notes, where data is processed, how access is logged, and what happens on termination. Ask separately about audio, transcript text, generated notes, support access, backups, and model improvement. “We delete recordings” does not answer how long the transcript or note remains.

Consent is a workflow, not a sentence copied from a blog. Requirements vary by jurisdiction, setting, recording law, professional rule, and organizational policy. Explain the tool in plain language, give the client a meaningful opportunity to decline, and provide a fallback. The ClinicFrame patient-consent guide can help structure the conversation, but it is not legal advice.

Choose the note format before choosing the vendor

DAP, BIRP, and SOAP organize attention differently. DAP separates encounter data, assessment, and plan. BIRP makes the intervention and response relationship more explicit. SOAP fits many integrated medical environments but can create a weak or overfilled Objective section in talk therapy. The DAP vs BIRP vs SOAP comparison explains the trade-offs. A custom template is useful only if the practice can define what every section should and should not contain.

Ask each vendor to generate two formats from the same session. Compare whether changing labels changes the actual content selection or merely rearranges the same summary. For group practices, test shared templates, version control, supervision, and how changes affect existing clinicians. A technically customizable tool can still create governance problems if every user silently builds a different record.