AI Medical Scribe for Behavioral Health

Behavioral health clinicians carry a documentation load that no other specialty does. A 50-minute therapy session produces a richer narrative than a 15-minute medical visit, the note has to track affect and process and risk, and most clinicians are doing this for six to ten clients a day. The math does not work. Scriber is built to make it work.

The behavioral-health documentation problem

Most therapists and counselors do their notes after the clinic day ends. The American Psychological Association's 2023 practitioner survey found that more than half of psychologists in clinical practice spend over an hour on documentation outside of paid client time each day, and roughly a third spend two hours or more. That is unpaid work that arrives on top of an already emotionally demanding caseload. It is the single biggest contributor to clinician burnout in the field, and it is the reason private practices lose clinicians who otherwise love the work.

The problem is not just the time. It is the cognitive shape of the time. A medical SOAP note can often be written from a structured intake form and a few vital signs. A behavioral health note has to capture what the client said, how they said it, what changed in their presentation across the hour, what interventions you used, how the client responded, and what the plan is for next session. If you also do risk assessment, safety planning, or substance-use screening, that all goes in too. The note is a narrative, and narratives do not auto-fill from a template.

So clinicians cope in one of three ways. They write during the session, which fragments their attention and reads to clients as note-taking instead of presence. They write between sessions in a five-minute gap, which produces thin notes that miss clinical detail. Or they write at the end of the day from memory, which is the most common path and the most exhausting, because by 7 p.m. the details have started to blur and every note is a small act of reconstruction.

The stakes for getting this right are not trivial. Insurance audits look at note quality to confirm medical necessity. State licensing boards review records when there is a complaint. Court subpoenas pull notes that were written in a hurry and ask them to hold up against legal scrutiny. And in supervision and group practice, weak documentation creates real continuity-of-care risk when a client is seen by another clinician. Documentation is clinical work. It is also legal work and billing work. Doing it badly because you are tired is not a sustainable practice model.

An AI scribe addresses the time and the cognitive shape at the same time. It captures the session as it happens so you do not have to split your attention. It produces a structured first draft so you are editing rather than composing from memory. And it does this in a few minutes after the session ends, not at 9 p.m. on a Sunday.

What an AI scribe does for behavioral health

An AI scribe is a piece of software that listens to a session, transcribes it, and turns the transcript into a structured clinical note in the format you specify. Three things happen in sequence.

The first is ambient capture. With your client's informed consent, the scribe records the session audio. For in-person sessions this is microphone input from the device running Scriber. For telehealth, it is the audio from the call. The clinician does not have to type, dictate, or click anything during the hour. The capture runs in the background while you do the work.

The second is transcription. After the session ends — or, on engines that support it, in real time during the session — the audio is converted to text by a speech-to-text engine. Scriber does not lock you into one engine. We expose seven, including Azure Speech, Microsoft AI, Speechmatics, Deepgram, AssemblyAI, MAI, and xAI. Different engines have different strengths: some are faster, some are more accurate on specialized vocabulary, some are better at distinguishing speakers. Clinicians can pick what works for their caseload, and group practices can standardize on one. In our internal benchmarks, xAI returns a typical 50-minute session transcript in about three seconds, which is the fastest of the engines we tested.

The third is structured note generation. The transcript is passed through a clinical large language model with a prompt template tuned for the note format you chose — SOAP, DAP, BIRP, or one of the others described below. The model extracts the clinical content from the conversation and writes it into the format. It does not simply summarize. It identifies presenting concerns, mood and affect cues, interventions used, client response, risk language, and plan, and it places each in the right section of the note.

The output is a draft. The clinician reads it, edits anything that is wrong or missing, and signs off. In a typical workflow this takes three to five minutes per note instead of fifteen to twenty. Across a clinic day, that is roughly an hour of clinician time returned. The audio is discarded once the note is generated. Only the text note is kept, and only in the systems the clinician chooses to keep it in.

Note formats Scriber generates for behavioral health

Behavioral health uses more note formats than any other specialty, and the right format depends on the setting, the funding source, and the clinician's training. Scriber generates all of the common ones from the same captured session — you choose the format at the end and the same audio can be re-rendered into a different format if you decide later.

SOAP

Subjective, Objective, Assessment, Plan. The most widely used format across medicine and the one most insurers recognize. In behavioral health the Subjective section captures the client's reported experience, the Objective section captures observable presentation (affect, speech, behavior), the Assessment integrates clinical impression and progress, and the Plan covers interventions and next steps. Use SOAP when your EHR or payer expects a medical-style note.

DAP

Data, Assessment, Plan. A condensed format favored by therapists and counselors who find the Subjective/Objective split awkward for talk therapy. Data captures what happened in the session, Assessment is your clinical thinking about it, Plan is what comes next. Use DAP for routine therapy progress notes when you want the note to be readable as a session story.

BIRP

Behavior, Intervention, Response, Plan. Common in community mental health and case management settings. It centers the clinician's intervention rather than the client's narrative and is well suited to behavioral and skills-based modalities. Use BIRP when your agency or funding source requires intervention-focused documentation.

GIRP

Goal, Intervention, Response, Plan. A variant of BIRP that anchors each note to a specific treatment plan goal. Use GIRP when payers require the note to demonstrate work toward documented goals, which is common with Medicaid managed care and many state behavioral-health authorities.

SIRP

Situation, Intervention, Response, Plan. A crisis and case-management variant that opens with the presenting situation rather than a goal. Use SIRP for crisis sessions, mobile crisis encounters, and unscheduled contacts.

Biopsychosocial assessment

A long-form intake document that covers biological, psychological, and social factors across the client's history. Used at the start of treatment to establish a working formulation. Scriber generates the full biopsychosocial structure from a 60- to 90-minute intake session.

Treatment plan

A goal-and-objective document that names the diagnosis, the treatment goals, the measurable objectives, the interventions, and the timeframe. Required by most payers within the first one to three sessions. Scriber generates a draft treatment plan from a discussion with the client about goals and from the assessment material already on file.

Intake note

The first session's clinical note. Shorter than a full biopsychosocial but longer than a routine progress note. Captures presenting problem, history, mental status, risk, diagnosis, and initial plan.

Mental status exam

A structured snapshot of appearance, behavior, speech, mood, affect, thought process, thought content, perception, cognition, insight, and judgment. Generated either as a standalone note for psychiatric encounters or embedded in the Objective section of a SOAP note.

Progress note

The general-purpose name for any of the above when used as a routine session note. Most clinicians settle on one format — SOAP, DAP, or BIRP — and use it for all routine sessions. Scriber will keep your default and let you override per session.

Telehealth and in-person, one workflow

Behavioral health is the specialty most committed to telehealth. The shift that began in 2020 did not reverse, and most therapists today have at least a hybrid practice. A scribe that only works for in-person sessions is missing half the work.

Scriber captures telehealth sessions through a direct Doxy.me integration. The clinician launches the Doxy room from inside Scriber, and the Doxy waiting room and call run in the same window as the Scriber recorder. Audio from the call is captured through the browser's tab-share mechanism, with the client's consent collected the same way it is collected for any telehealth recording. From the clinician's side, the session looks and feels like a normal Doxy call. From Scriber's side, the audio flows into the same transcription and note-generation pipeline that handles in-person sessions.

This matters because most AI scribes positioned for behavioral health do not have a real telehealth integration. They have landing pages that say "works with Doxy" or "works with Zoom," and what that means in practice is "you can copy and paste the note into the platform afterwards." The scribe is not connected to the call. The clinician runs the scribe on a phone next to the laptop, or shares a screen, or otherwise improvises. Scriber's Doxy integration is a real product feature: the call and the scribe live in the same window, the audio is captured natively, and the workflow is a single workflow.

For in-person sessions, capture is whatever microphone the device running Scriber has access to — laptop, phone, or external mic. The transcription and note pipeline is identical to telehealth. A clinician with a hybrid caseload uses the same product the same way regardless of where the client is.

HIPAA, BAA, and behavioral-health privacy

Federal HIPAA rules apply to AI scribes the same way they apply to EHRs. Any vendor that creates, receives, maintains, or transmits protected health information on behalf of a covered entity is a business associate, and a Business Associate Agreement is required before patient data can be shared. There is no clinical exception, no small-practice exception, and no telehealth exception.

Scriber signs a BAA with every user at signup. Not at the enterprise tier, not after a sales call, not as an upgrade. Solo clinicians on the lowest paid plan get the same BAA that group practices get, because the federal rule does not distinguish between them and neither do we. This is one of the most basic compliance requirements in the field and a surprising number of competitors gate it behind enterprise pricing. We do not.

Behavioral health adds two layers on top of baseline HIPAA. The first is 42 CFR Part 2, the federal rule that governs records of substance use disorder treatment. Part 2 imposes consent and disclosure requirements stricter than HIPAA. If your practice treats SUD, the records you produce are subject to Part 2, and your scribe vendor's data handling has to be compatible with Part 2. Scriber's no-audio-retention default and explicit consent-capture flow are built with Part 2 in mind.

The second layer is state mental health privacy law. Many states impose protections on mental health records that exceed HIPAA: stricter consent rules for minors, separate handling of psychotherapy notes, longer retention requirements, and in some states, a private right of action for breaches. California, New York, Illinois, and Massachusetts each have meaningful additional rules that practices in those states need to follow. Scriber is built to support the more restrictive state rules without configuration.

A few specific points worth naming. Audio is processed in memory and discarded as soon as the note is generated. We do not retain raw audio for training, model improvement, debugging, or any other purpose. The text note is retained only in the systems the clinician chooses — your Scriber account, your EHR if you export to one, your local files if you download. Encryption is AES-256 at rest and TLS in transit. The Azure environment Scriber runs in is covered by Microsoft's HIPAA BAA. We do not train models on your client data, we do not share data with third parties for marketing, and we do not retain content beyond the operational minimum needed to serve the product.

What clinicians use Scriber for

The most common use is routine therapy progress notes: 45 to 50 minutes of session, three to five minutes of editing, signed and exported before the next client. This is the backbone of the workflow and the one that returns the most time across a clinic day.

Therapy intakes are the second most common. A 60- to 90-minute first session produces a full biopsychosocial assessment plus a draft treatment plan, both of which would otherwise take the clinician an hour or more to write longhand. Scriber drafts both from the same session audio.

Treatment planning sessions specifically — the conversation with a client about goals at the start of treatment, or the periodic review every 90 days that most payers require — generate clean draft treatment plans. The clinician edits the goals and objectives for clinical accuracy and signs.

Supervision review. Supervisors of trainees and associate-level clinicians use Scriber notes as a starting point for case review. A supervisor reading a draft note can see the session content faster than reading a hand-written summary, and the trainee gets feedback on the clinical reasoning rather than on note craft.

Safety planning and risk assessment sessions are documented with the same workflow as any other session. When a client expresses suicidal ideation, the conversation that produces a safety plan is captured, transcribed, and rendered into a note that includes the risk assessment, the safety plan, and the clinical reasoning. Documenting these sessions well is one of the highest-stakes documentation tasks in behavioral health, and getting a complete first draft means the clinician spends review time on the clinical content rather than on whether they remembered every element.

Common questions

Is Scriber HIPAA-compliant for behavioral health notes? Yes. We sign a BAA with every user at signup, encrypt data at AES-256 at rest and TLS in transit, host on Azure under Microsoft's HIPAA BAA, and discard audio after the note is generated. There is no enterprise gate on compliance — solo clinicians on the lowest paid plan get the same compliance posture that group practices get.

Does Scriber generate DAP and BIRP notes? Yes, plus SOAP, GIRP, SIRP, biopsychosocial, treatment plan, intake, mental status exam, progress, and discharge summary. You choose the format at the end of the session, and the same captured audio can be re-rendered into a different format if you change your mind.

Can I use Scriber for telehealth sessions? Yes, with a direct Doxy.me integration. Launch a Doxy room inside Scriber and the call audio flows into the transcription and note pipeline natively. There is no separate device, no copy-paste step, and no need to run the scribe outside the call.

How accurate is Scriber for psychiatric terminology? Scriber supports seven STT engines, three of which are specifically tuned for medical and clinical vocabulary (Microsoft AI, Speechmatics, Azure). In our internal benchmarks, accuracy on psychiatric terminology — medication names, DSM-5 diagnostic language, mental status descriptors — is within about one percent of human transcription. Clinicians who do heavy psychopharmacology can switch to a medical-vocabulary engine for those sessions.

Is my client's audio stored anywhere? No. Audio is processed in memory and discarded after the text note is generated. We do not retain audio for training, debugging, or any other purpose. Only the generated text note persists, and only in the storage locations the clinician chooses.

AI scribe versus human scribe. A human scribe runs $30,000 to $60,000 per year per clinician. An AI scribe is roughly two orders of magnitude cheaper, available outside business hours, and consistent across sessions. A human scribe handles ambiguity better and can ask clarifying questions in real time. Most behavioral health practices that try a human scribe move to an AI scribe within 12 months on cost grounds.

AI scribe versus templates. A template fills in structure but not content. The clinician still has to write the narrative. A scribe captures the narrative and fills in the structure. Most clinicians who switch from templates to a scribe do not go back, because the scribe produces a longer, more detailed first draft than a template-driven workflow does.

What happens if my internet drops mid-session? Audio capture continues locally. When the connection comes back, Scriber uploads the captured audio and processes it. A 90-second outage in the middle of a session does not lose the session.

Try Scriber for behavioral health today

Scriber gives behavioral health clinicians their evenings back without trading away note quality, compliance, or clinical detail. The product is designed for the way therapists, counselors, psychologists, psychiatrists, social workers, and addiction counselors actually work — long sessions, narrative notes, hybrid in-person and telehealth caseloads, and real federal and state privacy rules.

Start a 7-day trial at scriberapp.com/register and run your next session through Scriber. If you want to see pricing for solo, group, and supervisor accounts, the pricing page lays out the options. A BAA is included at signup on every paid plan.