How BetterHelp, Talkspace, Grow Therapy, and SonderMind Use AI: Their AI Roles

The short answer: all four use AI, mostly behind the scenes

If you use BetterHelp, Talkspace, Grow Therapy, or SonderMind, AI may support matching, documentation, or other parts of the service. None of these platforms has replaced its therapists with chatbots. All connect users with licensed clinicians, and the sessions are delivered by humans. What the AI does is surround the session: it matches you with a therapist, drafts the clinical note afterward, summarizes your progress before the next appointment, and in some cases talks to you between sessions through journaling prompts, reflections, or app-based check-ins.

That distinction matters. AI handling paperwork is a different risk category than AI handling your feelings. The first mostly raises privacy and accuracy questions. The second raises safety questions, which is why platforms move much more carefully there and why regulators in states like Illinois have drawn hard lines around AI acting as the therapist.

The descriptions are based on dated company announcements, earnings-call coverage, and reporting cited below. Products change fast in this space, so treat this as a snapshot and check your own app's settings and consent screens to learn which features apply to your care.

For preparation that you control, the stress level test offers educational reflection and a mood tracker can organize observations to bring to your appointment. These tools cannot establish a diagnosis. The AI therapy guide explains how standalone chatbots differ from software used within a clinical service.

The practical distinction is between choosing a clinician, documenting an appointment, and receiving an automated reply between appointments. Ask which of these functions is active in your account. A platform may offer several AI tools without every tool applying to your care. Session transcription, therapist preparation, and optional journaling can involve different information and different consent choices.

BetterHelp: AI matching, plus a privacy record worth knowing

BetterHelp's most visible use of AI is matching. When you fill out the intake questionnaire, an algorithm pairs you with a therapist. On a 2023 earnings call, executives at Teladoc Health, BetterHelp's parent company, said the platform uses AI to optimize member-therapist matching across more than 100 criteria and was matching a patient with a provider roughly every 30 seconds. Teladoc has since said it is investing in AI across the business to improve efficiency and engagement, including for BetterHelp.

BetterHelp also carries the industry's most instructive privacy lesson. In March 2023, the Federal Trade Commission charged that BetterHelp had shared consumers' email addresses, IP addresses, and health questionnaire answers with Facebook, Snapchat, Criteo, and Pinterest for advertising, despite promising to keep that data private. The FTC finalized its order in July 2023: BetterHelp paid $7.8 million toward partial consumer refunds and is banned from sharing health data for advertising.

The episode predates the current wave of AI features, and the order now restricts what BetterHelp can do with your data. It is still the clearest reminder in this industry that the intake answers you give a therapy platform are sensitive data, and that what a company promises in its privacy policy and what it does are separate questions worth verifying.

Talkspace: smart notes, session prep, and an AI podcast

Talkspace has been the most public about building an AI toolkit for its therapists. Its smart notes feature generates draft summary notes of member sessions for providers to review. On January 15, 2025, the company launched Insights, which synthesizes a client's assessment scores and recent session details into a pre-session primer for the therapist, then generates a post-session update on progress and follow-ups. Talkspace says this work comes out of a dedicated AI Innovation Group and that client data stays inside its encrypted, HIPAA-covered platform.

The client-facing experiments are more unusual. Talkcast, rolled out in 2025, lets a therapist generate a short personalized podcast episode for a client after a session, recapping themes and suggesting things to practice. Speaking at a CNBC event in October 2025, then-CEO Jon Cohen said the podcast feature had improved engagement across early therapy sessions. That is a company-reported engagement claim, rather than independent evidence of symptom improvement.

Talkspace also runs AI on the safety side. Risk algorithms embedded in its platform scan conversations for signs of suicide risk and alert the therapist. Cohen told CNBC that within its New York City teen program, those alerts had contributed to human interventions for suicide prevention. Cohen has simultaneously been one of the loudest voices warning against teens using general-purpose chatbots for mental health support, calling the consequences disastrous and urging parents to pay attention. If you are in crisis or thinking about suicide, call or text 988 (US Suicide & Crisis Lifeline), available 24/7.

Grow Therapy: ambient scribes with dual consent

Grow Therapy, a marketplace that connects clients with independent therapists and handles their insurance and admin, announced AI-assisted clinical tools on February 20, 2025, rolling out to a network of more than 17,000 providers. The tools use ambient listening: software listens to the session and produces a draft clinical note for the therapist plus a plain-language after-visit summary for the client covering what was discussed and what comes next.

The consent design is worth copying industry-wide. The announcement says both the client and the provider must agree before the tools activate and describes the tools as HIPAA compliant. Verify recording and transcript retention in the current consent terms before agreeing. The launch announcement describes provider review and a human-centered approach. Consent to listening should be considered separately from storage: ask about temporary audio, transcripts, final notes, and each retention period.

Grow has kept building on the client side since then, announcing AI-assisted between-session reflections that surface themes from client journaling, and an AI coach chat feature developed and monitored under clinician oversight. Each step moves AI closer to the client, which makes the consent and oversight questions more important, and Grow's public framing has consistently kept the licensed clinician in charge of care. Availability can depend on location, account eligibility, and the specific service. Confirm access in your own account before treating an announced feature as part of your care.

SonderMind: AI Notes and a growing AI suite

SonderMind, which matches clients to in-network therapists and psychiatric providers, commercially launched its AI Notes tool on September 16, 2025. With patient consent, it transcribes the session and formats the clinical note for the provider to review and submit. The company says the tool cut average note completion time by 80 percent, from about 20 minutes to about four minutes, saving providers up to 90 minutes a day, and that 88 percent of its providers opted in, with more than 100,000 AI-generated notes submitted to insurers by launch. These figures come from SonderMind's launch announcement and describe its reported workflow results; they are not an independent comparison with other platforms.

In October 2025, SonderMind expanded this into a broader AI Suite: session prep and treatment planning support for providers, plus client-side app features like session takeaways, journaling, and goal tracking designed to extend support between weekly appointments.

SonderMind has also been specific about governance in its announcements: HIPAA and ISO 27001 compliance, encryption, features reviewed by an internal AI Governance Council of clinicians and compliance experts, and a stated policy that provider and client data is never sold or used for advertising. Those are company claims rather than independent audits, but they are the right claims to look for, and they give you concrete language to hold the platform to.

What stays human, and what to ask your platform

Across all four platforms, the pattern is consistent: AI drafts, summarizes, matches, and reminds, while licensed humans assess, treat, and carry responsibility for care. That split exists for good reasons. The evidence that the human relationship drives therapy outcomes is among the strongest in the field: a meta-analysis by Flückiger and colleagues found that the alliance between client and therapist robustly predicts outcome. The World Health Organization has likewise urged caution about deploying large language models in health care faster than they can be tested and governed.

Session transcription deserves your specific attention. When an AI scribe is on, your words are being processed by software, sometimes by third-party subprocessors, to produce the note. Reputable platforms require consent and encrypt the data. You still have the right to understand the pipeline before you agree.

Between-session AI features deserve a second look too. Journaling prompts, reflections, personalized recaps, and coach-style chat all extend the platform's reach into your week, which can help you practice skills, and also means more of your inner life flows through the company's systems. Treat these features as optional extras you evaluate one at a time rather than defaults you accept in a single tap, and remember that none of them is a crisis tool.

Five questions worth asking any platform: Is anything I say transcribed or processed by AI, and can I opt out without losing service? Is my data used to train AI models? Who besides my therapist can access AI-generated notes or summaries? What happens to transcripts if I delete my account? And does a human review every AI-generated note before it enters my record? A platform that answers these clearly is telling you something good about itself. One that cannot is telling you something too.

Is AI therapy HIPAA compliant?

HIPAA coverage depends on the organization and data flow. HHS explains that information in personal-use apps generally falls outside HIPAA unless the app is provided by a covered entity or business associate. A clinician's documentation system and a standalone consumer chatbot can therefore have different obligations. Ask which entity processes the information and which protections apply.

A HIPAA claim does not establish that generated notes are accurate or that a support feature improves health. Ask who reviews an AI summary, how you can flag an error, and whether raw audio or transcripts have different retention rules from your clinical record. Consent, confidentiality, accuracy, and clinical effectiveness each require a separate answer.

HHS distinguishes the medical record from a clinician's separately maintained psychotherapy notes. Patients generally have access rights to medical and billing records held by covered providers, with exceptions; the same access right does not extend to separately maintained psychotherapy notes. An app's label of "notes" does not settle which category applies. Ask where a generated summary is stored and whether it becomes part of your medical record. Source: HHS, Your Medical Records, cited below.

A consent checklist for your next appointment

Before a session, ask whether AI listening is active, what it produces, and whether declining changes access to the clinician. Request a plain-language explanation of any third-party processor and the retention rules for audio, transcripts, and final notes. Record the date and scope of your consent so you can revisit it when features change. The AI therapy privacy guide explains additional questions about sharing and retention.

After the session, read any patient-facing summary for factual mistakes about symptoms, medication, relationships, or agreed next steps. Ask the clinician how to request a correction if something is wrong. An automatically generated recap can sound fluent while reversing who said something or turning a tentative discussion into a firm conclusion.

Evaluate optional journaling and chatbot features separately from the appointment itself. Ask whether your clinician actually reads entries, how often someone checks alerts, and what response time to expect. If immediate help is needed, use a crisis or emergency service rather than waiting for a software flag to be reviewed.

If a summary gets something wrong, identify the specific passage and explain what you believe is inaccurate or incomplete. Ask the clinician whether the error appears only in the patient recap or also in the formal record. HHS explains that you can request an amendment to a covered medical or billing record; if the provider disagrees, you can submit a statement of disagreement for the record. A correction request is different from asking a chatbot to regenerate its answer. Source: HHS, Your Medical Records.

If you withdraw permission for an optional feature, ask when processing stops and what happens to information already collected. Distinguish disabling future listening from deleting a transcript or changing a completed clinical record. Request confirmation through the platform's support or consent process, and tell your therapist which optional features you intend to use so expectations about between-session contact are clear.