Can Patients Really Consent to AI Scribes in Psychotherapy?

There would seem to be a simple answer to the ethical problem raised by using an AI scribe in psychotherapy: ask the patient. After all, psychotherapy sessions cannot simply be recorded without the patient’s knowledge and permission. The American Psychological Association’s Ethics Code is explicit that psychologists obtain permission before recording the voices or images of individuals to whom they provide services. More recent APA guidance addressing AI scribes similarly emphasizes informed consent, privacy, confidentiality, data handling, security, and the clinician’s continuing responsibility for the clinical record.

So perhaps the matter is straightforward. Explain what the AI scribe does. Explain the potential risks and benefits. Tell the patient that they are completely free to decline and that their treatment will not be affected if they do. Obtain consent. Problem solved.

Except that psychotherapy has never been quite so simple. The difficulty is contained in the word consent itself. What does it actually mean for a patient to freely consent to something their therapist wants them to do? And what happens to that question when the thing being requested—the recording and processing of the patient’s most private speech—primarily benefits the person doing the asking?

The Peculiar Problem of Consent in Psychotherapy

Informed consent is one of the foundations of contemporary health-care ethics. At its simplest, it protects the individual’s right to decide what happens to them. The patient receives sufficient information about a proposed intervention, its risks, benefits and alternatives, and makes a voluntary decision about whether to proceed.

But informed consent has always been more complicated in psychotherapy than this model suggests. One difficulty is that psychotherapy cannot be adequately described in advance. We can explain our theoretical orientation. We can discuss confidentiality and its limits. We can describe fees, frequency, foreseeable risks and possible benefits. But neither therapist nor patient can know beforehand exactly what a psychotherapy will become. A patient cannot know what memories will emerge, what attachments will develop, what conflicts will become activated, what will be discovered—or what it will feel like to discover it.

The psychotherapy literature recognized this problem long before anyone imagined an AI scribe. Marta and Lowy, writing about informed consent in psychotherapy in 1993, emphasized that psychotherapy differs from many medical interventions because it is an evolving process involving active participation by the patient and because unconscious processes, transference and countertransference are part of the treatment itself. They argued that consent in psychotherapy is therefore better understood as interactive, iterative, repeated and revocable than as a single event occurring before treatment begins.

Beahrs and Gutheil similarly emphasized the complexities of implementing informed consent in psychotherapy, including the unpredictability of interactional outcomes and the risk that an overly legalistic approach to consent may itself interfere with the therapeutic alliance.

There is another complication. People usually seek psychotherapy at moments when they are, in one way or another, vulnerable. This does not mean they lack autonomy. It means that the conditions under which autonomy is exercised deserve particular attention.

Someone entering psychotherapy may be depressed, frightened, traumatized, ashamed, bereaved, confused, desperately lonely or in crisis. They may have difficulty asserting themselves. They may be highly sensitive to rejection. They may have spent a lifetime accommodating other people’s wishes. Indeed, some patients enter psychotherapy precisely because saying no to another person has never felt psychologically safe.

And now the therapist asks them for a favor.

“Would You Mind if I Used an AI Scribe?”

The therapist may not experience it as asking for a favor. They may regard the AI scribe simply as a clinical tool. Perhaps it allows them to concentrate more fully during sessions. Perhaps it reduces hours of documentation. Perhaps it improves the organization of their notes. Perhaps it means that instead of spending Sunday afternoon completing charts, they can spend that afternoon with their family.

These are understandable benefits. But notice something important about them. They are, to a considerable extent, benefits to the therapist. The person who stands to benefit from the patient’s agreement is therefore the same person asking the patient to agree.

This creates an ethical tension that a consent form cannot make disappear. Imagine that a therapist carefully explains the technology and then says: You are completely free to say no. It will make absolutely no difference to your treatment. This sounds reassuring. But is the therapist in a position to guarantee it?

There Are Two Unconsciouses in the Room

Psychotherapy is built upon the rather inconvenient discovery that human beings do not always know everything they feel, want or communicate. Psychotherapists accept this premise every day.

We notice the patient’s wish to please. We wonder about an unexpected hesitation. We become interested in why a patient apologizes before disagreeing with us. We pay attention when someone who describes themselves as “fine with everything” repeatedly organizes relationships around accommodating others.

We also attend to ourselves. Therapists are not neutral machines. We have emotional reactions to our patients and to what occurs in treatment—preferences, frustrations, wishes, anxieties, expectations and blind spots. In psychoanalytic language, aspects of these responses are often discussed under the broad heading of countertransference, but one need not work analytically to recognize that therapists bring their own subjectivity into every clinical relationship. Some of these reactions are conscious. Others may not be.

Why, then, would we imagine that these dynamics suddenly cease operating when we ask a patient to consent to a technology that we would very much like to use?

A patient may hear the therapist say, It is entirely your choice, while simultaneously perceiving—correctly or incorrectly—that the therapist hopes the answer will be yes. The patient may wonder whether refusing will be inconvenient. They may worry about appearing mistrustful. They may know that nearly every other patient has agreed. They may not want to disappoint the therapist.

And beneath these relatively ordinary social concerns may lie questions that are much more emotionally charged: Will you think differently about me? Will I become the difficult patient? Will you be annoyed with me? Will you still like me? Will you still want to take care of me?

The therapist may sincerely believe that refusal will change nothing. But the therapist has an unconscious too. Perhaps they do feel disappointed. Perhaps there is a flicker of irritation. Perhaps they find themselves explaining the technology again several sessions later, convinced that the patient simply did not understand how safe it was. Perhaps none of this happens. The point is that we cannot solve the problem merely by declaring that it will not happen.

When a “No” Becomes a “Yes”

An acquaintance recently told me about an experience involving her child’s therapist. The therapist asked for permission to use an AI scribe during the child’s sessions. The mother declined. The question was raised again. She declined again. Eventually, after being asked repeatedly, she agreed.

Her description of what happened was striking: she felt that she had simply been worn down into consenting.

There is, of course, no way to generalize from a single anecdote. But the story exposes something important about the ethics of consent. At the end of this process, the therapist had obtained a yes. But was the ethical problem therefore resolved?

The mother’s original preference had been perfectly clear. She did not want the technology used. Nothing necessarily changed in her evaluation of the technology. What changed was her willingness to continue resisting the request.

This raises a question that deserves considerably more attention as AI scribes become normalized: When does repeatedly seeking consent become pressure to consent?

The distinction matters because ethical consent requires more than eventual acquiescence. Recent bioethical writing about ambient AI has begun to raise precisely these broader concerns. Discussions of what have evocatively been called clinical “listening walls” have questioned whether conventional approaches to consent are adequate for technologies that continuously capture clinical conversations, emphasizing unresolved questions involving autonomy, privacy, trust and governance.

But psychotherapy introduces another dimension that even this literature can miss. The person requesting consent is not simply a health-care provider. They may be one of the most psychologically significant people in the patient’s life.

But Some Patients Say They Like AI Scribes

There is now emerging research suggesting that many patients are comfortable with ambient AI scribes. Some report appreciating the clinician’s increased attention when the clinician is no longer occupied with documentation. These findings are worth taking seriously.

For example, a 2026 Stanford study surveyed 2,202 patients following outpatient visits in which an ambient AI scribe had been used. Seventy percent found the scribe helpful, and nearly three-quarters wanted their clinician to use it again in future visits. The authors appropriately described their findings as evidence that ambient scribes may be acceptable to many patients in clinical practice and called for further research into issues including privacy, trust and communication.

But we need to be extremely careful about what findings like these tell us about psychotherapy. These were outpatient medical visits, not psychotherapy sessions. Evidence that patients may find ambient scribes acceptable in general medical care cannot establish that the technology is psychologically neutral when introduced into psychotherapy.

Indeed, another study points directly toward the difficulty. Lawrence and colleagues’ 2025 study of informed consent for ambient AI documentation involved 103 patients and 18 clinicians in ambulatory medical care and found that patients’ comfort depended upon factors including trust in their clinician, understanding of the technology, and perceptions of its benefits and risks. The authors concluded that the consent conversation itself is a nuanced and consequential part of patients’ acceptance of the technology.

The distinction between general medicine and psychotherapy is crucial. Psychotherapy is not simply medical care involving unusually private information. The interpersonal encounter itself is the treatment.

In ordinary medical care, conversation often functions primarily as a vehicle for communicating information. The physician needs to know where it hurts, when the symptom began, what medications the patient takes, and what has changed since the last appointment.

In psychotherapy, what is said matters—but so does the manner in which it becomes possible to say it. A hesitation may matter. A silence may matter. The decision to tell the therapist something today that could not be said last month may matter. A joke made immediately after an unbearable disclosure may matter. So may the thing that is almost said and then withdrawn.

Patients reveal sexual fantasies, betrayals, hatred, envy, shame, aggression, traumatic memories, resentment toward their children, thoughts about death, and feelings about the therapist sitting in front of them. Sometimes they arrive intending to disclose these things. Often they do not. A thought emerges. Something becomes speakable. The patient surprises the therapist—and sometimes surprises themselves.

An ambient recording system is therefore not simply another documentation device operating in the background. It introduces another listening presence into a setting in which the experience of being listened to is already central.

We urgently need research specifically examining AI scribes in psychotherapy: not merely whether patients say they are satisfied with them, but whether their presence affects spontaneity, disclosure, inhibition, trust, silence, the therapeutic alliance, the patient’s feelings and expectations about the therapist, the therapist’s own responses, and the subjective experience of privacy. Evidence of acceptance in general outpatient medical care cannot answer these questions.

A Visit to the Veterinarian

I had an unexpectedly illuminating experience of this recently in a setting about as far removed from psychoanalysis as one can imagine. We had just adopted a rescue dog and took him for his first veterinary appointment.

On the wall was a notice informing us that appointments would be recorded and that anyone who did not wish to be recorded should tell the clinic. It was an opt-out arrangement. We were already there. We were slightly anxious about our new dog. And neither my husband nor I had any particularly sensitive information to reveal. The dog himself, as far as we knew, had no secrets whatsoever. So we didn’t object.

But during the appointment I noticed something. The conversation felt different. I felt slightly more conscious of myself speaking. More interestingly, I experienced the veterinarian as somewhat more guarded. Her answers seemed unusually careful and conservative. I had the impression that she was less inclined to think aloud—to entertain possibilities or offer the kind of tentative, speculative thought that people sometimes share when they are simply talking together.

I cannot know whether the recording actually changed her behaviour. But I know that it changed my experience of the encounter. Someone—or something—was listening. There was, suddenly, an audience. The dog remained impressively indifferent.

It struck me afterward that this had happened during an interaction in which virtually nothing deeply personal was at stake. What happens when the subject is incest? Or an affair? Or a sexual fantasy? Or the wish that one’s child had never been born? Or the patient’s growing realization that they are furious with their therapist?

Knowing that one’s words are being captured may have no effect whatsoever. But surely we cannot simply assume that.

“We Can Turn It Off Whenever You Want”

Another reassurance commonly offered is that the patient can ask for the AI scribe to be turned off whenever particularly sensitive material arises. Again, this sounds reasonable. But it presumes a rather peculiar model of psychotherapy.

Imagine a patient approaching something they have never previously been able to say. There is a silence. They begin speaking. Then they stop: Actually, before I tell you this, could you turn off the AI?

Perhaps some patients could do exactly that. But for others, the act of identifying the coming disclosure as especially secret may itself be exposing. More importantly, the interruption may disturb the very psychological process through which something previously unspeakable was becoming speakable.

Psychotherapy does not reliably announce its most sensitive moments in advance. Sometimes we discover what a session was about only after we have said it.

Compliance Is Not the Same as Ethics

Professional organizations are beginning to respond to AI much more seriously. The American Psychological Association has published guidance concerning artificial intelligence in psychological practice and, more recently, specific guidance for psychologists evaluating AI scribes. That guidance identifies informed consent, privacy compliance, data handling, security, accuracy and professional responsibility as matters clinicians need to consider. APA also explicitly notes that most AI scribes function through ambient listening to clinical sessions and that the clinician remains responsible for reviewing AI-generated documentation before it becomes part of the clinical record.

These developments are important. The APA Ethics Code itself is also unambiguous about recording: Standard 4.03 states that psychologists obtain permission before recording the voices or images of people to whom they provide services.

But regulation and professional guidance have inevitably been trying to catch up with technologies that entered clinical practice with extraordinary speed. And regulation can take us only so far.

It is comparatively easy to ask whether a vendor encrypts data. We can determine whether recordings are retained. We can investigate whether data are used for model training. We can require disclosure. We can design a consent form. All of these things matter enormously.

But none tells us what happens psychologically when a therapist who wants to use a technology asks a vulnerable person, who depends upon that therapist and may be deeply attached to them, for permission to do so.

That question cannot be outsourced to legislators, regulators—or AI vendors. Psychotherapists will have to think about it ourselves.

Because psychotherapy is not radiology. It is not surgery. It is not even an ordinary conversation between a physician and patient. We work within a relationship in which attachment, dependency, authority, fantasy, resistance, compliance, the patient’s expectations and feelings about the therapist, and the therapist’s own emotional responses—sometimes discussed in psychoanalytic language as transference and countertransference—are not unfortunate contaminants of an otherwise straightforward clinical exchange. They are part of what we are working with.

Some Uncomfortable Questions

Before asking a patient to consent to an AI scribe, perhaps clinicians should ask themselves something first.

Why do I want this patient to agree? Who benefits most from the technology? How will I actually feel if the patient says no? Would I ever ask again after receiving a clear refusal? If so, why?

Have I presented the technology neutrally—or have I unconsciously been making the case for it? Would a patient who tends to comply with authority experience this request differently from another patient? Could the request itself become part of the patient’s feelings and expectations about me and our relationship? Could my own wish for the patient to agree influence my emotional responses to the patient—what an analytically oriented therapist might think about in terms of countertransference?

Might the patient speak differently because the session is being captured? Might I speak differently?

And perhaps most importantly: If the technology primarily makes my professional life easier, why should my patient have to alter the conditions of their psychotherapy to accommodate it?

These questions do not necessarily lead to the conclusion that AI scribes should never be used in psychotherapy. They lead somewhere less comfortable. They ask us to recognize that obtaining consent is not necessarily the same thing as establishing that consent was psychologically free.

Can the Patient Really Say No?

There is a danger, whenever new technology enters professional practice, that ethics becomes procedural. We disclose. We explain. We obtain the signature. We document that consent was obtained. And having completed the appropriate steps, we experience the ethical problem as settled.

Psychotherapy should make us particularly suspicious of that conclusion. Our entire discipline rests, in one way or another, upon the recognition that what people say consciously and what is happening psychologically are not always identical.

A patient’s yes may certainly mean yes. But sometimes a yes can also contain accommodation, dependency, anxiety, resignation, identification, fear of conflict, or the wish to preserve an important relationship.

That does not render the patient’s decision meaningless. Nor does it give the therapist the right to decide what the patient “really” wants. It does, however, place a greater ethical responsibility upon the person with more power in the relationship.

Perhaps, then, the most important question is not: Did I obtain my patient’s consent to use an AI scribe?

It is: Have I created a relationship in which my patient can comfortably refuse me?

And perhaps there is one further question we should be willing to ask ourselves: When my patient says no, can I leave it at no?

Where We Stand

For us at Note Designer, this is ultimately a question about what kind of space we believe psychotherapy needs to be. We do not believe that AI should be listening to psychotherapy sessions, and we have deliberately chosen not to build recording or transcription into Note Designer. Our AI tools begin only after the clinician has decided what information belongs in the clinical record; they can assist with rewriting that information, but they are never present in the therapeutic encounter itself. This reflects a distinction we care deeply about: there is a difference between using technology to assist with the work that follows a psychotherapy session and introducing that technology into the session itself. As clinicians, we believe there remains something worth protecting in the privacy of two people thinking and speaking together, without an additional listener in the room.

American Psychological Association. (2017). Ethical Principles of Psychologists and Code of Conduct. Standard 4.03: Recording.

References and Further Readings:

American Psychological Association. (2025). Ethical Guidance for AI in the Professional Practice of Health Service Psychology.

American Psychological Association. (2026). Guidance for the Evaluation of AI Scribes.

Beahrs, J. O., & Gutheil, T. G. (2001). Informed consent in psychotherapy. American Journal of Psychiatry, 158(1), 4–10. https://doi.org/10.1176/appi.ajp.158.1.4

Kumah, E., Antwi, J., Boakye, D. S., Botchwey, C. O.-A., & Anyimadu, E. (2026). The ethics of listening walls: Patient autonomy and consent in the age of ambient clinical AI. npj Digital Medicine, 9, 506. https://doi.org/10.1038/s41746-026-02973-z

Lawrence, K., Kuram, V. S., Levine, D. L., et al. (2025). Informed consent for ambient documentation using generative AI in ambulatory care. JAMA Network Open, 8(7), e2522400. https://doi.org/10.1001/jamanetworkopen.2025.22400

Marta, J., & Lowy, F. H. (1993). Informed consent: A prerequisite for psychotherapy? Canadian Journal of Psychiatry, 38(8), 547–551. https://doi.org/10.1177/070674379303800804

Shah, S. J., Murtagh, K. N., Lin, S., Garcia, P., et al. (2026). Patient perspectives on clinicians’ use of ambient AI scribes. JAMIA Open, 9(3), ooag104. https://doi.org/10.1093/jamiaopen/ooag104