Push-to-Talk Dictation vs Always-Listening AI Scribes for Therapists

They are different products solving overlapping problems. The distinction matters because the compliance, ethical, and clinical implications are not the same.

Push-to-talk dictation is a tool the clinician uses to convert their own speech into text. The clinician presses a hotkey, speaks the note in their own words — usually between sessions or after the session ends — and the words appear as text in the EHR or note app. The client is not on the microphone. Nothing is recorded. The clinician is the author of the note in every meaningful sense.

AI scribes (Heidi, Freed, Mentalyc, Supanote, Abridge, Nabla, DeepScribe, SimplePractice Note Taker) record the entire session, send the audio to a cloud server, run speech recognition on both speakers’ voices, and use a large language model to generate a structured clinical note — SOAP, BIRP, DAP, or modality-specific. The clinician reviews and signs the LLM’s draft. The audio and transcript of the session pass through the vendor’s infrastructure; how long they are kept varies by vendor.

Both reduce documentation time. The trade-offs they ask the clinician to accept are very different.

Therapy session recording is governed by a combination of state wiretap law, professional ethics codes, and clinic policy. Each layer adds friction.

Under state wiretap law, the United States is a patchwork of one-party and two-party consent states. In one-party consent states (most of the country), a single party to the conversation can authorize recording. In two-party (or all-party) consent states — including California, Florida, Illinois, Maryland, Massachusetts, Montana, Pennsylvania, and Washington — every participant must agree before recording is lawful. Recording a client without their consent in those states is a criminal matter, not a paperwork lapse.

Under professional ethics, APA Ethics Code Standard 4.03 requires psychologists to obtain permission before recording the voices or images of the people they serve. The ACA Code of Ethics (B.6.b) and the NASW Code of Ethics (standard 1.03(h)) set similar consent requirements for counselors and social workers. Adopting an AI scribe means securing documented consent from every client, every time, with renewed consent if the recording purpose, vendor, or retention changes.

Under subpoena exposure, a recording or verbatim transcript that still exists can be requested in custody disputes, malpractice claims, board complaints or criminal cases, like any other record. Many scribes delete audio once the note is drafted, so check your vendor’s retention policy. Producing a summary note you wrote is a different experience from producing a client’s own recorded words.

A minority of states require all-party (also called two-party) consent to record a private conversation. California (Penal Code 632), Florida, Pennsylvania and Washington are well-known examples. The statutes differ in their details, including how they treat in-person conversations versus phone calls, so check your own state; the Reporters Committee for Freedom of the Press publishes a state-by-state recording guide.

A therapist licensed in any of these states who uses an AI scribe must document explicit consent from every client before the recording starts. Many clinics handle this with a one-time consent form added to intake paperwork; others handle it with a verbal consent recorded at the top of each session. Both methods work if implemented consistently. The risk arises when a new client signs the intake but later withdraws consent verbally, or when a returning client recorded before is asked again under a new vendor or retention policy.

A clinician who uses push-to-talk dictation has none of this surface area. The clinician records themselves, between sessions, stating their own clinical reasoning. There is no patient voice on the microphone, no recording to consent to, and no two-party consent question to manage. This is not a minor advantage in two-party-consent states.

It happens, and it is the client's right. Questions about whether sessions are recorded, and by whom, have become a normal part of intake conversations as more practices adopt AI tools.

When a client declines AI scribe use, the clinician has two options: revert to typing the note themselves for that client, or use a non-recording dictation tool. Most clinics that adopt AI scribes maintain a mixed workflow: scribe-enabled for clients who consent, manual for the rest. Push-to-talk dictation makes the manual case faster without changing the clinician’s ethical posture.

Sapience Med is designed to be the same tool a therapist uses for every client, regardless of consent posture, because the tool does not record the client at any time.

The reasons clinicians give tend to fall into four groups.

Authorship clarity. Push-to-talk produces the clinician’s own words, in their own clinical voice. AI scribe output is an LLM rendering of the session, which the clinician must review for accuracy and tone. Generated drafts can contain small errors or statements the clinician would not have written. The discovery liability is the same either way: the clinician signs the note and owns it. Push-to-talk minimizes the gap between intent and output.

Patient trust. Many clients, especially those seeking therapy specifically for privacy-sensitive issues (trauma, sexual orientation, immigration status, custody disputes), explicitly do not want their sessions recorded. A clinician who avoids ambient recording communicates a posture; one who records and discloses it communicates a different posture. Both can be ethical; only one signals maximum-restraint.

Subpoena minimization. Audio or a verbatim transcript of a session says far more than a summary note and is harder to put in context. If no recording is ever made, there is nothing beyond the note to request.

Cost and complexity. As of October 2026, AI scribes that therapists use and that sign BAAs are listed from $0.99 a session (Blueprint) and $19.99 a month for a 40-note plan (Mentalyc) up to $150 a month (Heidi Health, billed monthly), and the price recurs for as long as you use them. Sapience Med is a one-time $149 license and needs no recording-consent workflow.

AI scribes are not the wrong tool everywhere. There are legitimate clinical contexts where ambient recording with explicit consent makes sense:

• Inpatient or hospital settings where every encounter is already documented in detail and standardized SOAP output is the deliverable.
• High-volume primary care or behavioral health intake where ambient drafting saves enough time to be the difference between sustainable practice and burnout.
• Group practices that have already implemented consent workflows, BAA review, and EHR integrations with the AI scribe vendor.
• Clinicians who specifically want LLM-generated structured notes (e.g. PCL-5 or PHQ-9 templated extraction).

For solo and small-group mental health practices where the clinician already writes thoughtful notes and just wants to write them faster, push-to-talk dictation is the better fit. That is the audience Sapience Med is built for. Read the HIPAA architectural rationale for the compliance dimension, or see how Sapience Med works in SimplePractice for setup.