AI Scribes for Mental Health: Notes, EHR Workflows and Review

An AI scribe for mental health drafts documentation for a clinician to review; mental health EHR software maintains the clinical record. The useful evaluation question is how those two functions work together: what enters the draft, which note type receives it, who can see it, and what happens after the clinician signs or corrects it.

For a therapy practice, that may mean moving a reviewed DAP note into an appointment record. For a psychiatry service, it may mean connecting the clinician’s supplied assessment and medication decisions with the appropriate encounter fields. Neither workflow is complete simply because a readable paragraph appears.

Use this guide to evaluate AI-scribe workflows alongside mental health EHR software: choose a note type, review the draft and verify the saved record. The examples and reusable evaluation record focus on note preparation, review and EHR transfer.

Sources checked October 8, 2026. Published by Vero, an AI documentation software provider.

Choose the record workflow before the scribe

Start with the problem your team needs to solve. If staff cannot find the current treatment plan, an AI drafting subscription will not by itself fix record organization. If the EHR works well but clinicians spend too long converting session information into a progress note, a drafting tool may be worth evaluating. If corrections disappear after saving, investigate version handling and transfer before adding another generator.

Separate three responsibilities:

• Drafting: capture permitted information, apply the selected template, preserve attribution and uncertainty, and let the clinician correct the output.
• Record management: attach the approved note to the correct patient and encounter, handle signatures and amendments, control access, and produce appropriate copies.
• Care workflow: complete the actual assessment, make clinical decisions, assign follow-up and verify that outstanding work reaches its owner.

An EHR may also include an AI scribe. These responsibilities still need separate acceptance checks. A single supplier does not make them a single action, and a standalone scribe does not necessarily replace scheduling, prescribing, billing or records administration. The EHR introduction explains the wider record system; this article concentrates on the mental health note moving through it.

For an initial scope, name the clinician role, encounter type, language, input method and receiving record. “Adult individual-therapy follow-up, English, clinician dictation after the visit, reviewed DAP text entered into the EHR” is a testable workflow. “AI for behavioral health” is not.

Documented software examples and record controls

These vendor-documented workflows, checked October 8, 2026, show how each product handles templates, correction and delivery. Evaluate the drafting route and receiving record together.

SimplePractice: appointment-linked progress notes

• Templates: its documentation lists Simple Progress Note, SOAP and DAP on Starter; prebuilt and customized templates are available on Essential and Plus.
• Signing and correction: clinicians sign and lock notes. Authorized users can unlock a note with a reason; the action is logged, but edits overwrite the original note, so preservation of the original information needs an explicit procedure. Client-signed notes cannot be unlocked.
• Delivery: the documented Client Portal route shares the five most recent progress notes; older notes can be exported.

For a demonstration, use the intended plan, correct a signed sample through the permitted route, and inspect the resulting record. The same help page also links SimplePractice’s AI Note Taker. SimplePractice progress-note documentation.

TherapyNotes: structured progress notes and revision history

• Templates: its progress-note workflow offers SOAP, DAP, BIRP and Play Therapy alongside its standard format, with template customization.
• Signing and correction: notes can be saved as drafts and electronically signed. Its editing documentation describes archived revisions, prior signatures and timestamps in Revision History.
• Delivery: selected notes can be downloaded together as a PDF. Native publication of every progress note to the patient portal is not established by the sources reviewed here.

Test what is carried forward from earlier notes and what needs a fresh clinician entry. A populated diagnosis, examination or risk field is not evidence that it was reassessed today. Sources: progress-note workflow, revision history, note downloads.

Vero: a draft-and-export workspace alongside the EHR

• Templates and input: Vero documents choosing a template and providing encounter information through recording, typed context or supported uploads.
• Review and correction: the generated note can be edited directly or revised through Vero Chat. The clinician reviews the complete result before adding it to the record.
• Delivery: the documented route is copy to the EMR or export through PDF, Word, print or email options. This source does not establish native write-back, EHR signing, co-signing or amendment controls.

Vero therefore fits the drafting step in this comparison. Test the actual destination separately, including whether copying the plan preserves its owner and deadline. Vero’s creating-a-note instructions.

Compare access as carefully as the editor

Ask who can view each record category, not just whether the software has a “private note” label. SimplePractice’s administrative-role documentation says an additional document-viewing permission includes progress and psychotherapy notes. TherapyNotes documents a separate Psychotherapy Note and a bulk-download warning that lets users deselect those notes; the reviewed page does not establish author-only visibility. Inspect role permissions and export contents with synthetic records. SimplePractice permissions; TherapyNotes psychotherapy-note handling.

Match the note type to the encounter

A format organizes information; it does not determine what happened. For a therapy progress note, useful distinctions include what the patient reported, the intervention actually provided, the patient’s response, the clinician’s interpretation and the agreed next step. Keep an intervention separate from a planned intervention and a patient’s report separate from a clinician’s observation.

SOAP organizes subjective information, objective findings, assessment and plan. DAP groups data, assessment and plan. BIRP organizes behavior, intervention, response and plan. A team can express the same supplied facts in different formats, but it should not invent content to fill every heading. If a clinician has not supplied examination findings, the software should not produce a normal examination to make the note look finished.

An intake or diagnostic evaluation needs the actual history, assessment and rationale appropriate to that service. Do not treat a short follow-up template as a complete intake. A medication-management encounter needs the prescriber’s actual reconciliation, assessment and decisions; a patient mentioning a medicine does not establish a current prescription or an instruction to change it. A treatment plan describes agreed goals and planned work, while the progress note records what happened in a particular encounter.

Check language that can easily change meaning: “previously,” “sometimes,” “patient reports,” “not assessed,” and “awaiting records.” Also review what a generated summary omits. Retaining every symptom while dropping the patient’s response to the intervention can leave a polished but less useful account.

The American Psychiatric Association’s July 2026 resource document identifies omissions, misattribution and loss of nuance as scribe-evaluation concerns, and emphasizes clinician review and approval. Its professional guidance covers integration, privacy, clinical validity and workflow. APA resource document, pages 45–52.

Worked example: a therapy progress note

Follow MH-EX-01 from its source facts to the saved record. Check attribution, time, unsupported additions and pending work against the supplied dates and actions.

Read the source first. Then compare the flawed draft, the source-aligned DAP excerpt and the incomplete saved copy. Open each review to see the mismatch and correction.

Why the receiving record changes the answer

The corrected excerpt preserves the agreed review date and the outside-report task. The saved-copy example loses the task even though all its remaining sentences are accurate. This is an omission, not a recognition error. A reviewer who checks only whether the opening paragraph “sounds right” will miss it.

The repair has two parts: restore the Plan field and verify that the records coordinator has an actual assigned task. Signing the note does not request the report. Likewise, writing that a clinician will review something does not demonstrate review. The clinical notes software guide develops these saved-version and amendment checks.

Adapt the structure without importing defaults

For a BIRP version of this excerpt, the patient-reported activity goes under behavior, the supplied planning work under intervention, the patient’s stated response under response, and the dated next steps under plan. Keep the clinician’s supplied interpretation visible in the locally approved structure. For SOAP, do not move the patient’s report into objective findings just to populate that section.

A useful blank prompt is: “Source and time → intervention completed → patient response → clinician interpretation → next action, owner and review date.” Add the service’s required assessment and record fields from the responsible clinician. Avoid prefilled denials, normal findings or copied-forward statements that have not been verified for this encounter.

Psychiatry example: reported medication use, decisions and transfer

In MH-EX-02, a fictional psychiatry case, “Medicine A” is an illustrative medication label. The source packet supplies the prescriber’s interpretation, decision and follow-up. The task is to preserve each in the note and receiving record.

Supplied encounter facts

• P1 · Existing record: At this October 8 encounter, the existing prescription entry reads “Medicine A, one tablet each evening.” It is a historical prescribing instruction, not evidence of how the patient has actually taken it. No medicine strength is supplied in this exercise.
• P2 · Patient report: The patient reports missing the October 5 and 6 doses and taking the October 7 dose. They report unchanged anxiety and nausea after that evening’s dose. They prefer to discuss the symptoms before making a medication change.
• P3 · Prescriber’s interpretation: The prescriber records that nausea followed the reported dose, but its cause is unconfirmed. Interrupted use limits interpretation of treatment response. No resolution of anxiety or confirmed adverse drug reaction is supplied.
• P4 · Prescriber’s decision: No dose change, discontinuation or new prescription is authorized in the supplied decision. The prescriber plans a telephone review of actual use and tolerability on October 9. This is a planned clinical review, not a completed call.
• P5 · Follow-up and patient discussion: The patient agrees to the proposed review and prefers telephone contact. The coordinator must arrange it by 17:00 on October 8 and notify the prescriber before the end of that day if unable to arrange it. At drafting, the appointment is not yet booked.
• P6 · Assessment boundary: Examination, mental-status and risk-assessment findings are not included in this source packet. Required encounter documentation must come from the treating clinician; a template must not fill those gaps with normal findings or denials.

From those facts to a note excerpt

A flawed draft might say: “Taking Medicine A as prescribed. Medication caused nausea and has been discontinued. Anxiety stable; no safety concerns. Follow-up booked for tomorrow.” Each shortcut changes the source: reported missed doses disappear, a temporal association becomes causation, an unauthorized medication action appears, missing assessment becomes reassurance, and planned follow-up becomes a booking.

The source-aligned excerpt keeps the distinctions visible:

Reported use and symptoms: Existing prescription entry lists Medicine A, one tablet each evening. Patient reports missing October 5 and 6 doses and taking the October 7 dose. Reports unchanged anxiety and nausea after that evening’s dose; prefers discussion before a medication change.

Prescriber’s interpretation: Nausea occurred after the reported dose; cause unconfirmed. Interrupted use limits interpretation of treatment response.

Decision and follow-up: No dose change, discontinuation or new prescription authorized in the supplied decision. Prescriber plans telephone review of actual use and tolerability on October 9. Patient agrees and prefers telephone contact. Coordinator to arrange the review by 17:00 on October 8 and notify the prescriber before the end of that day if unable to arrange it. Appointment remains unbooked at drafting.

Keep symptom reports, clinical conclusions and medication instructions separate. “Unchanged anxiety” remains a patient report; patient-facing medication instructions must come from the prescriber. Preserve the dates supplied for this encounter.

Verify three destinations after transfer

Now suppose the receiving EHR preserves the narrative but maps “missed doses” to discontinued in its medication list and marks the follow-up completed. A correct paragraph has become an incorrect workflow.

1. Encounter note: Preserve the reported dates, symptoms, uncertain causation, prescriber decision and patient preference. Required assessment fields remain for clinician completion.
2. Medication record: Keep the historical prescription distinguishable from reported use. Remove the unsupported discontinuation through the EHR’s attributable correction route; do not create a replacement prescription or mark reconciliation complete merely because text was copied.
3. Follow-up task: Show the coordinator as owner, the October 8 scheduling deadline, planned October 9 clinical review, telephone preference and escalation condition. Keep the task outstanding until the corresponding work actually occurs.

Evaluation decision: fail this transfer. Compare all three destinations with P1–P6, correct the unsupported statuses and reopen the current record to confirm the repair. Retain the before-and-after evidence. Track clinical assessment and call completion in their own fields.

For Ontario physicians, CPSO’s documentation policy distinguishes encounter assessment, treatment and response, management/follow-up and prescriptions actually issued. Preserve these distinctions in the receiving record. CPSO Medical Records Documentation, provisions 10–11, checked October 8, 2026.

What mental health scribe research shows

Individual therapy: adoption and feedback, not measured time savings

A March 2026 retrospective mixed-methods study examined Talkspace Smart Notes. In its use analysis, 162 full-time and 1,366 contractual providers submitted 286,911 notes. Separately, its ratings analysis reported optional feedback on 30,519 of 466,139 notes (6.5%); those are different denominators. It concerned eligible individual-therapy workflows; couples therapy and psychiatry were excluded. Provider ratings were largely favorable, but administrative documentation time was not measured, and all authors were Talkspace employees. The findings support further workflow evaluation, not an independent accuracy ranking or a claim of proven patient benefit. McCrudden and colleagues, JMIR Formative Research.

Primary care: more symptom documentation is not the same as better care

A January 2026 retrospective matched analysis examined 20,302 adult annual primary-care visit notes across two health systems. AI-scribed notes had greater model-estimated psychiatric symptom documentation, but lower adjusted odds of a documented psychiatric intervention than contemporaneous unscribed visits: 0.83, 95% confidence interval 0.72–0.95. The composite measure concerned depression diagnosis, antidepressant prescribing or referral for psychiatric evaluation and care. This is an association with possible residual confounding, not proof that scribes caused undertreatment. It is also primary-care evidence, not a psychotherapy trial. Castro and colleagues, JAMA Psychiatry.

For a clinic pilot, the practical lesson is to measure more than note length or user satisfaction. Count clinically material additions and omissions, unfinished follow-up, correction effort and failed captures. Include patient choice and the effect on the session. Do not transfer another product’s findings to Vero or your own configured workflow.

Privacy depends on the record and jurisdiction

United States: progress notes are not automatically psychotherapy notes

HIPAA uses psychotherapy notes for separately maintained notes by a mental health professional documenting or analyzing counseling conversations. It distinguishes these from routine chart information such as medication monitoring and summaries of diagnosis, symptoms, treatment and progress. With limited exceptions, disclosure of psychotherapy notes requires patient authorization, including disclosure to another clinician for treatment. Calling every therapy progress note a psychotherapy note would blur that distinction. HHS OCR explanation, checked October 8, 2026.

42 CFR Part 2 is another scope question, not a label for all mental health care. It applies to federally assisted programs providing substance use disorder diagnosis, treatment or referral, with some requirements also applying to recipients of Part 2 records. Compliance with the 2024 changes was required by February 16, 2026. Determine whether the proposed records are in scope before choosing the consent and disclosure workflow. HHS Part 2 guidance, checked October 8, 2026.

Have the privacy lead map the data flow and applicable agreements before patient use. Address recording consent, vendor contracts and authorization to disclose specific records as separate requirements.

Ontario: apply the relevant professional and privacy requirements

For Ontario physicians, CPSO advises explaining AI use and obtaining patient consent before recording conversations with AI. Physicians remain accountable for their documentation and existing privacy obligations. Psychologists, social workers, psychotherapists and clinicians in other provinces should use their own regulator’s requirements. CPSO AI guidance, checked October 8, 2026.

Make the patient’s choice operational

Prepare a product-specific explanation of what is captured, where it is processed, who can access it, how long copies remain and how the clinician uses the draft. Verify those details against the configured service; do not promise immediate deletion or therapist-only access without support.

If a patient declines recording, use the approved manual route. If they ask to stop, confirm the capture state and continue without recording. Establish how the earlier captured material is handled under the applicable policy. For a telehealth session, test the patient’s audio channel as well as the clinician’s microphone; a transcript of only one participant can create a misleading account. Group, family, interpreter and minor-patient workflows need their own scope and access review.

Mental health documentation evaluation checklist

Run these checks with synthetic information before proposing a patient-data pilot. Record the application/version, template, input method, device, tested language, receiving EHR and date. Retain the raw draft, correction record and saved result where available. When the product does not expose a transcript or capture trace, record that evidence limit instead of claiming the missing layer passed.

The checklist identifies the responsible system and role. A drafting tool does not fail merely because the EHR owns signing, but the complete workflow still needs a working signature route. “Not applicable” requires a reason and approving owner in the internal record—for example, an interpreter test excluded from an English-only initial scope, not from every future use.

MENTAL HEALTH DOCUMENTATION — EVALUATION RECORD
Complete clinic findings and evidence in your approved internal workspace.
Product/edition, template and configuration: [fill in]
Reviewer/role, jurisdiction, test date, device/app version: [fill in]
Encounter types, languages, source and receiving EHR: [fill in]

MH-01 — Patient, encounter and source
Responsible system/owner: Complete workflow · treating clinician
Test: Open two synthetic encounters with similar names; include a historical source.
Expected result: Only the intended encounter receives the note; sources retain dates and speaker identity.
Evidence required: Source packet, encounter identifiers and saved-note comparison.
Status: Not run
Actual finding / evidence reference / test date: [fill in]
Correction effort / transfer effort / failed attempts: [fill in]
Remediation / owner / due date / recheck result: [fill in]

MH-02 — Unassessed fields and uncertainty
Responsible system/owner: Drafting tool + clinician
Test: Omit examination/risk findings; mark an outside report unavailable.
Expected result: No normal examination, denial, diagnosis or reviewed-report claim is invented.
Evidence required: Template configuration, raw draft, corrections and clinician review.
Status: Not run
Actual finding / evidence reference / test date: [fill in]
Correction effort / transfer effort / failed attempts: [fill in]
Remediation / owner / due date / recheck result: [fill in]

MH-03 — Intervention, response and plan
Responsible system/owner: Complete workflow · clinician/operations lead
Test: Run MH-F3 through MH-F6, then transfer the reviewed excerpt.
Expected result: Intervention, response, owner, date and escalation survive; the task exists separately.
Evidence required: Reviewed draft, saved Plan field and task status.
Status: Not run
Actual finding / evidence reference / test date: [fill in]
Correction effort / transfer effort / failed attempts: [fill in]
Remediation / owner / due date / recheck result: [fill in]

MH-04 — Consent and stopping capture
Responsible system/owner: Capture tool + clinical/privacy leads
Test: A synthetic participant declines initially, then a second rehearsal requests stop mid-session.
Expected result: Manual documentation remains possible; no new audio is captured after verified stop.
Evidence required: Patient explanation, observed control state, capture trace and fallback procedure.
Status: Not run
Actual finding / evidence reference / test date: [fill in]
Correction effort / transfer effort / failed attempts: [fill in]
Remediation / owner / due date / recheck result: [fill in]

MH-05 — Note type and recipient access
Responsible system/owner: Receiving EHR + privacy/records lead
Test: Use synthetic progress and separately maintained notes; inspect permitted and restricted roles and an export.
Expected result: The clinic’s approved access and disclosure rules hold in the record, portal and export.
Evidence required: Role matrix, view/export results and approved record-category policy.
Status: Not run
Actual finding / evidence reference / test date: [fill in]
Correction effort / transfer effort / failed attempts: [fill in]
Remediation / owner / due date / recheck result: [fill in]

MH-06 — Signature, revision and re-generation
Responsible system/owner: Drafting tool + receiving EHR
Test: Correct a draft, regenerate, sign a sandbox note and attempt an amendment.
Expected result: No silent replacement of an approved correction or signed record; the amendment route is attributable.
Evidence required: Before/after versions, author/date history and amendment procedure.
Status: Not run
Actual finding / evidence reference / test date: [fill in]
Correction effort / transfer effort / failed attempts: [fill in]
Remediation / owner / due date / recheck result: [fill in]

MH-07 — Audio path and language
Responsible system/owner: Capture tool · health-IT and specialty reviewers
Test: Rehearse in-person or telehealth audio, including an interpreter if in scope.
Expected result: Each required speaker is captured and attributed; interruption or missing audio is detectable.
Evidence required: Device/app versions, audio route, tested language and separate findings per setup.
Status: Not run
Actual finding / evidence reference / test date: [fill in]
Correction effort / transfer effort / failed attempts: [fill in]
Remediation / owner / due date / recheck result: [fill in]

MH-08 — Retention, exit and support copies
Responsible system/owner: Both suppliers + privacy/records lead
Test: Export a synthetic record, remove test access and rehearse a support request without patient content.
Expected result: Approved copies remain usable; access removal works; retention/deletion exceptions are known.
Evidence required: Contracts/data map, export contents, access test and deletion/backup schedule.
Status: Not run
Actual finding / evidence reference / test date: [fill in]
Correction effort / transfer effort / failed attempts: [fill in]
Remediation / owner / due date / recheck result: [fill in]

Decision and permitted scope: [hold / bounded pilot / approved workflow, with rationale]
Unresolved requirements and restrictions: [fill in]
Clinical + privacy/health-IT approval names and dates: [fill in]
Next review and change triggers: [fill in]

MH-03 — fictional completed row
Configuration: example EHR sandbox; manual transfer of the MH-EX-01 source-aligned excerpt.
Reviewer: example treating clinician. Test date: October 8, 2026.
Expected: intervention, response and all follow-up work survive transfer.
Finding: the saved Plan field retained October 15 follow-up but lost the report coordinator’s October 12 recheck and notification condition.
Evidence: SAMPLE-SOURCE-01, SAMPLE-DRAFT-02, SAMPLE-SAVED-03 (fictional references).
Status: Fail.
Remediation: clinician restores the missing sentence; records coordinator creates the corresponding task.
Recheck: SAMPLE-SAVED-04 contains the complete plan; SAMPLE-TASK-01 shows the owner, due date and unresolved report status.
Decision: transfer defect repaired. Complete remaining checks and clinical fields, then obtain workflow approval.

Include at least one reviewer familiar with the actual specialty and encounter type, plus the practice’s privacy/health-IT owner. Review language-specific findings with appropriately qualified speakers; do not pool interpreter and single-language results into one reassuring average. Retain failed attempts and abandoned sessions in the report.

Set stop conditions before testing. An invented risk denial, wrong-patient note, missing plan or unauthorized recipient should trigger a hold and investigation. Base release approval on the complete evaluation record, including remaining gaps, their owners and the permitted configuration.

Budget for the reviewed, saved note

Separate the EHR subscription from any scribe entitlement. Ask whether custom templates, additional clinicians, supervision, export and support change the plan or quote. Include configuration, privacy review, staff training, clinician correction and transfer effort. An apparently cheaper drafting tool can cost more if every note needs manual reformatting.

For a hypothetical budget, assume an EHR costs $160 per month, a scribe costs $60, and $120 of setup is spread over three months. That is $260 per month before staff time. At 120 encounters, three minutes of clinician review plus one minute of transfer per encounter takes eight hours. Valued at an assumed $90 per hour, that adds $720: $980 per month, or $8.17 per completed encounter after rounding.

Measure the current workflow with the same tasks and volume before interpreting a difference as a saving. Do not count time as “saved” when it simply moves to a supervisor or records coordinator. Keep clinical quality and permission failures separate from the financial calculation; a low unit cost does not resolve them.

A practical first implementation decision

Choose one encounter type, one approved input route and one receiving note format. Use the fictional packet to calibrate the reviewers, then test the actual configuration with synthetic material. Establish the manual fallback and verify the patient explanation, permissions and retention arrangement before any authorized patient-data pilot.

A bounded decision names what is approved and what is not: for example, individual follow-up progress-note drafting only, with other encounter types and multi-person capture still unassessed. Recheck after a template change, new language, new recording method, new recipient or changed retention setting. The goal is not the longest note. It is a faithful, usable record with a clear author and intact follow-up.