Mentalyc

The documentation model is the product. Mentalyc sells AI written therapy notes, and its other features, the treatment planner, progress tracker, Alliance Genie therapeutic alliance insights and recommended CPT codes, are generated from the same session content by the same models. It works alongside the practice's own record system rather than replacing it, so nothing of substance remains without the models.

Mentalyc describes a draft and sign structure. Its FAQ says every note is a draft until the clinician signs it; notes, treatment plans and recommended CPT codes pass through the clinician before export; and clinicians who do not record can work from a dictated or typed summary. On the Team plan a practice can review members' notes, including unsaved ones.

Against that, Mentalyc markets AI-flagged risk detection and safety planning in every note "so nothing critical gets missed, even in a busy session". The sentence makes the system the reason nothing is missed: an absolute claim that it does not err, on the content where an error matters most. Signing off does not answer it. Reviewing a draft shows the clinician what the model wrote, not what it left out, and an omitted risk disclosure is exactly the error the sentence promises away. No edit rate, acceptance figure or miss rate for risk content is published, and nothing describes how errors are reported or corrected.

The technology description stays general: transcripts and notes are generated by machine learning algorithms, large language models and statistical models, with no human scribe, and notes can follow a clinician's own templates. No model, provider, version or description of training data is published, and nothing describes how a model change is tested or announced.

Mentalyc does name what affects note accuracy, audio quality, session length and whether the note format matches the therapy modality, without figures for any of them. Its claim of up to 90 percent less documentation time carries no method.

Mentalyc names its host, AWS, and does not name the provider of the large language models that its FAQ says generate transcripts and notes. The privacy policy lists categories of service provider, including hosting, email, marketing, payments and customer support, without naming them, and no sub-processor list is published.

The terms that govern data once it arrives are stated more fully than the list of parties: session data is never used to train AI models, transcripts are anonymized before processing, audio is deleted within three days, and the business associate agreement binds subcontractors to protections at least as strict as its own. Whether the model provider receives anonymized transcripts only, and what it retains, is not stated.

The evidence on offer is scale of use and testimonials: more than 30,000 clinicians, a 4.8 rating on G2, and therapists' quotes on time saved. Adoption at that scale is not evidence of benefit. Mentalyc's claim of up to 90 percent less documentation time, and its statement that most therapists export notes without substantive edits, carry no method, denominator or reference standard. Its product, pricing and security pages cite no study, controlled evaluation or accuracy assessment.

Retention, training use and deidentification are stated specifically enough to be tested, with one inconsistency the company leaves unresolved. Its security page FAQ says session audio is kept for up to three days in case a note needs regenerating and is then deleted automatically; the same page elsewhere says recordings are deleted immediately after the note is generated.

Transcripts are anonymized before processing, with a partial option that keeps first names for family sessions, and Mentalyc treats anonymized transcripts as outside the medical record. Session data is never used to train AI models, the clinician can delete any session data at any time, and account data is kept while the account is active.

The privacy policy lets Mentalyc use aggregated anonymized data to improve the service, and its about page names an anonymized research data set on effective interventions as part of its founding aim. Notes are designed to leave out sensitive disclosures outside the treatment focus. Nothing is published on safety engineering: no guardrails, hallucination checks, or handling of a note containing something that was never said.

An AI scribe processing protected health information is a business associate, Mentalyc states, and it signs a Business Associate Agreement with every account. The agreement is published in full on its site and accepted by checkbox at registration, so it applies on every plan, and a signed copy can be generated from account settings.

It requires agents and subcontractors with access to protected health information to agree in writing to restrictions and conditions at least as protective, and to return or destroy data at termination, including data they hold. Mentalyc's aggregate liability under the agreement is capped at $10,000.

For Canadian practitioners Mentalyc states compliance with PHIPA, PIPEDA, PIPA and PHIA and supplies PHIPA-aligned consent material; it also states compliance with Australia's Privacy Principles and New Zealand's Privacy Act.

SOC 2 Type II compliance and certification are claimed, and Mentalyc presents the attestation as its own. The report period, the auditor and the services in scope go unstated, and the security page links no trust portal or downloadable report, so neither the report nor its period can be checked from the site. Other stated controls are encryption in transit and at rest and hosting on AWS; certifications held by AWS belong to the host, not to Mentalyc.

No medical device claim is made, and neither Mentalyc's terms of use (March 2026) nor its product pages state a position on device status or say whether the output may be relied on for clinical decisions. Several outputs are clinical judgments generated by software rather than transcribed: risk assessments and mental status exams populated from the session, AI-flagged risk detection and safety planning in every note, marketed as ensuring nothing critical gets missed, SMART treatment plans built from session content, and recommended CPT codes.

The exposure lies in generated risk assessments and mental status exams entering the record beside the clinician's own documentation, and Mentalyc does not say whether each risk flag cites the session passage it rests on. Everything reaches the record only after the clinician edits and signs. The rules that govern the product are payer documentation requirements, HIPAA and the Canadian privacy statutes, state licensure, consent law and 42 CFR Part 2.

Mentalyc writes notes to document medical necessity for payer audit, supplies a client consent template, and publishes Part 2 guidance that places patient consent on the provider and states it issues no Part 2 certificate.

Notes are described as accurate, ethical and compliant, and the approach as ethics committee approved, with no name for the committee and no word on what it reviewed. No bias evaluation, method or subgroup result is published. The groups where performance is plausibly at risk follow from the product's own inputs and from the accuracy factors Mentalyc itself names (audio quality, session length and modality): accent, dialect and language in transcription, child and play therapy sessions, and couples, family and group sessions where several speakers must be told apart. Risk detection is the output where a gap between groups would matter most.

Review before commit is the mechanism: Mentalyc's FAQ says every note is a draft until the clinician signs it, and notes are edited and saved before export to another system. Nothing stands behind the output: no accuracy or error figure, no published limitations, and no warranty or remediation commitment for a wrong note.

The business associate agreement caps Mentalyc's aggregate liability at $10,000, and its terms of use disclaim any warranty of accuracy and make stopping use the sole remedy for dissatisfaction. Two design choices bear on risk content, where an error matters most in psychotherapy notes that payers and courts may read.

Mentalyc markets AI-flagged risk detection and safety planning in every note "so nothing critical gets missed", and states that sensitive disclosures outside the treatment focus are not captured. No accuracy figure stands behind the risk flags, and nothing describes how the system decides what falls outside the treatment focus.

Mentalyc works alongside the practice's record system rather than replacing it. Its Chrome extension fills a finished note into the client's chart in one click, and Mentalyc names the systems it works with: SimplePractice, TherapyNotes, TheraNest, Jane, Osmind, Tebra and Kareo, plus other web based record systems. That is one directional writing through the browser, with no EHR login or API connection, so nothing is read back from the record.

Sessions can be captured alongside any video platform, including Zoom, Google Meet, Doxy.me and SimplePractice telehealth. Mentalyc names no electronic prescribing, laboratory, health information exchange or FHIR connection.

The service is hosted on AWS, with US based data residency stated and encryption in transit and at rest. Sessions can be recorded live in person or alongside any video platform, uploaded, dictated or typed, so clinicians who do not record can still use it, and session audio is deleted within three days.

For Quebec, the privacy policy states that transfers outside the province use standard contractual clauses or data processing agreements; for New Zealand, it states that standard contractual clauses are not currently relied on. Tenant isolation and the AWS region are not stated.

Tiers are published with figures, alongside a 14 day free trial with no credit card. Solo plans are priced by monthly note volume: Mini $19.99 a month for 40 notes and transcripts, Basic $39.99 for 100, Pro $69.99 for 160 and Super $119.99 for 330, with annual billing saving $60 to $240 a year. Higher tiers add features such as supervision notes, group therapy notes, EMDR, play and psychiatry formats, custom templates and computed CPT codes.

The Team plan is advertised as unlimited notes with team management, at a list price of $119.99 per seat a month shown at $59.99 per seat for new teams. That is enough to size the cost against a caseload without contacting anyone.

Coverage is stated clearly: mental health documentation for therapists, counselors, social workers and psychologists, covering individual, child, couple, family and group sessions, EMDR, play therapy and psychiatry formats, and supervision notes, with plans for solo clinicians and group practices. Mentalyc states that it serves more than 30,000 clinicians.

Nothing published validates the coverage by setting: no study, named deployment or outcome is tied to any session type or practice size. Its published Part 2 guidance places consent for substance use disorder records on the provider.