The Phone on the Side Table: Consent and Privacy When a Therapist Uses an AI Scribe
A small phone lies face down on the side table. A laptop sits open but untouched. Somewhere between the greeting and the first real sentence, a client may begin to wonder whether the room is being recorded, and by what. Ambient AI documentation — commonly called an AI scribe — has become a familiar presence in US health care, and it is now appearing in private therapy practices that never used to think of themselves as technology adopters.
The awkwardness is structural rather than personal. Most clients assume that what happens in a therapy room stays in the therapy room, and that assumption is broadly correct in spirit but imprecise in mechanics. Notes have always been written, stored, shared with insurers, and occasionally subpoenaed. What is new is the microphone.
What an AI Scribe Actually Does
An ambient documentation tool captures session audio, converts speech to text, and uses a language model to draft a clinical note — typically a progress note summarizing topics discussed, interventions used, and the client's presentation. The clinician reviews and edits that draft before it enters the record.
That pipeline matters because it separates two things clients often conflate: the note and the audio. The note is a familiar artifact of care. The audio is a raw recording of a conversation, and practices differ enormously in how they handle it. Some vendors process audio in memory and discard it within minutes. Others retain it for a defined window so a clinician can re-listen. A smaller number store it indefinitely or use it to improve their models. None of this is visible from the waiting room.
Why the Audio Question Is the Important One
A progress note is a summary written in clinical language. An audio file is the session itself — the pauses, the tremor in a voice, the names of family members, the details a client chose to say aloud once and never repeat.
Under the HIPAA Privacy Rule, a technology vendor that handles protected health information on a practice's behalf is a business associate and must be covered by a business associate agreement. That requirement is a floor, not a guarantee of good practice. It says nothing about retention periods, nothing about whether a model is trained on the data, and nothing about whether the client was told.
Recording law is a separate layer. A number of states — California, Florida, Illinois, Maryland, Massachusetts, Montana, New Hampshire, Pennsylvania, and Washington among them — require the consent of all parties to record a conversation, with variations in how strictly that applies to clinical encounters. A practice operating in one of those states should be asking for permission explicitly. A client in any state is entitled to ask regardless.
Reading the Notice of Privacy Practices for Disclosure Language
Every practice distributes a Notice of Privacy Practices, and most clients skim it for the section on insurance and never return. It is worth reading a second time with a specific question in mind: does this document name the tools the practice uses?
Look for a heading such as “Business Associates,” “Technology Vendors,” or “Electronic Health Record and Documentation Systems.” Strong notices name categories of vendors and state plainly that session audio may be processed. Weak notices use only generic language about “third-party service providers” and never mention audio at all.
Three phrases are worth flagging when you find them:
- “Ambient documentation” or “AI-assisted scribing” — a named disclosure, which is a good sign.
- “De-identified” — ask what that means in practice, since de-identification standards are technical and vary by vendor.
- “May be used to improve services” — this often signals model training. Ask directly whether session content is used that way.
If the notice says nothing about documentation technology, that is not proof of concealment. It may simply be out of date. It is, however, a reason to ask.
Six Questions Worth Asking Before the First Session
A short, matter-of-fact inquiry at intake is easier than a confrontation later. Consider these:
- Do you use an AI tool to help write notes during sessions?
- Is the session audio recorded, and if so, how long is it kept?
- Who has access to the audio, and is the vendor under a business associate agreement?
- Is session content used to train or improve a model?
- Can I see how the practice describes this in its privacy notice?
- Is there an alternative if I prefer not to be recorded?
Asking these questions is not an accusation. Clinicians who have adopted these tools thoughtfully tend to welcome the conversation, because it gives them a chance to explain safeguards they are proud of.
Requesting a Non-AI Arrangement
Some clients will decide the tool is fine. Others will not, and that preference deserves to be honored without penalty. The most effective request is specific, brief, and framed around the mechanics of documentation rather than suspicion of the clinician.
A workable phrasing: “I would rather my sessions not be captured by an AI tool. Can we arrange for notes to be written the traditional way, or would you prefer I find a clinician who works that way?”
Most practices can accommodate this. Documentation method is a workflow choice, not a clinical one, and a therapist who types or writes notes afterward loses nothing in quality of care. If a practice cannot offer an alternative — some larger organizations standardize on a single system — that is useful information, and it is reasonable to ask for a referral rather than to proceed uneasily.
What to Do If the Answer Is Vague
Vagueness is the signal to slow down. A clinician who cannot say where the audio goes, or who dismisses the question as unimportant, is telling you something about how the practice handles details generally. You do not need to win the argument. You need to decide whether this is the right room for you.
Clients routinely vet therapists on credentials, specialties, and fees. Asking about the microphone on the side table belongs in the same conversation — not as a test of technological virtue, but as an ordinary question about what happens to the most sensitive material you will ever speak aloud.
