Answers

AI scribes can be HIPAA compliant when the vendor signs a Business Associate Agreement (BAA), encrypts audio and notes in transit and at rest, offers audit logs, and provides data-retention controls. HIPAA compliance is not a certification the vendor holds — it is a posture the practice must verify per plan. Never send real client audio to a scribe without a signed BAA on the plan you purchased.

Sources: HIPAA scribes guide · Methodology — rubric

Only a subset. 42 CFR Part 2 imposes stricter re-disclosure limits than HIPAA on records from federally assisted SUD programs. A compliant scribe must support consent capture, prohibit re-disclosure without patient authorization, and offer data-segregation controls. Most general medical scribes do not meet this bar; behavioral-health-native vendors are more likely to.

Sources: SUD scribes guide · Insights — compliance posture

Written informed consent should disclose that a session is recorded, name the AI vendor, describe what data is stored and for how long, note the vendor's HIPAA and Part 2 posture, and offer a documented opt-out with no clinical penalty. Verbal consent alone is insufficient in most states and payer contracts. Consent language belongs in the intake packet, not the first session.

Sources: Informed consent guide

Only scribes with strong speaker diarization produce usable notes for couples and family work. Twofold Health, Blueprint, and Upheal all support multi-speaker sessions; diarization accuracy on our benchmark ranged from ~85% to 94% at three-plus speakers. Scribes designed for medical dictation typically collapse speakers and are unsuitable for relational modalities.

Sources: Couples & family therapy guide · Speaker attribution benchmark

Yes, but the field is narrow. Group therapy demands stable diarization across 4–10 speakers, per-member note synthesis, and treatment-plan tie-ins. Twofold Health and a small number of SUD-focused vendors ship group-therapy templates. Most consumer or medical-visit scribes do not — running a group session through them produces one blended note that will fail a payer audit.

Sources: Group therapy scribe use case

Partially. Child and adolescent sessions add three challenges: consent must include the guardian, play-therapy content is behavioral rather than verbal, and confidentiality around the guardian requires careful note structure. Prefer vendors that support developmental-language templates and per-note redaction. Never rely on the scribe to make disclosure decisions on the clinician's behalf.

Sources: Child & adolescent guide

There is no single 'best' — the top three scribes on our rubric (Twofold Health 9.0, Blueprint 8.9, Upheal 8.7) cluster within 0.3 points and each wins a different practice profile. Twofold is our disclosed editorial pick for practices covering therapy plus substance-use treatment. Blueprint is the strongest measurement-informed therapy pick. Upheal leads for telehealth-first solo therapists. Match the tool to the practice, not the leaderboard.

Sources: Full 2026 rankings · Methodology

Solo-clinician plans on tracked vendors range from about $79 to $199 per month in 2026, with the median near $129. Group plans are typically 20–35% cheaper per seat. Community-mental-health and Medicaid-oriented deployments negotiate custom pricing tied to encounter volume. Free tiers exist but usually cap monthly sessions and exclude the BAA required for real client audio.

Sources: Pricing trend dataset · Compare tool

There are free tiers (limited sessions per month, no BAA, no EHR push) and open-source ASR you can self-host. Neither is appropriate for real client audio in production. If you are testing the workflow, use synthetic transcripts on a free tier; if you are documenting real clients, budget for a paid plan on a vendor that will sign a BAA.

Sources: Free vs paid analysis

Native EHR pushes vary widely. Twofold Health integrates natively with SimplePractice, TherapyNotes, and Alleva. Blueprint and Upheal cover the mainstream therapy EHRs. Many vendors still require copy-paste or manual upload for niche systems. Check the EHR coverage matrix under Insights before signing a contract — an unsupported EHR erases most of the time savings.

Sources: EHR coverage matrix

In our 200-session blind benchmark, hallucination rates ranged from about 1.2% to 4.8% of note claims depending on vendor and visit type. Rates are highest on intakes (dense history) and group work (speaker confusion), lowest on straightforward individual follow-ups. Every note should be clinician-reviewed before signing; treat the scribe as a first draft, not a final artifact.

Sources: Hallucination rate benchmark

On our five-visit-type benchmark, top vendors scored 8.4–9.1 out of 10 against a clinician-authored reference note — strong enough that the AI draft is usually faster to edit than to write. Weaker vendors scored 6.0–7.5, requiring so much rewriting that time savings collapse. Purpose-built behavioral-health scribes outperformed general medical scribes on every visit type.

Sources: Note quality benchmark

Policies vary and matter. Some vendors delete audio within 24 hours of note generation; others retain it 30 days or longer for model improvement unless the customer opts out. For behavioral-health work, prefer vendors that delete audio by default and let you disable any training-data use in the BAA or DPA. Assume the retention policy is your policy — you are the custodian.

Sources: Security & data handling guide

Run a two-to-four week pilot on 10–20 sessions across your real visit mix. Score each note on faithfulness, format adherence, and edit time. Compare against your unassisted baseline. Include at least one intake, one couples or family session, and (if applicable) one group. Decide on measurable criteria before starting; if the vendor cannot meet them, do not extend the pilot on trust.

Sources: Buyer guides