Calm AI Therapy Editorial · 9 min read

77% of Therapists Say Their Patients Are Already Using AI. Here Is What Worries Them.

The APA surveyed psychologists in 2026 about patients bringing AI into therapy. The numbers are striking, the concerns are specific, and most of them are fair. We go through them one at a time.

Published September 16, 2026

The American Psychological Association ran a survey in 2026 asking psychologists what they were seeing. Seventy-seven percent said their patients report using AI. More than a third said patients were using it as an additional mental health provider, not as a search engine or a journal, but as someone else in the room.

That is a large number, and it arrived faster than anyone planned for. It is worth sitting with what the same psychologists said about it, because the concerns are specific and most of them are correct. We build one of these products. We would rather engage the criticism than route around it.

The first concern is data. Two thirds of psychologists said they were worried about breaches. They are right to be. Mental health data is the most sensitive category there is, and the industry's record is poor. Our answer is architectural rather than reassuring: your conversations are stored in your account, not pooled into a training set, and you can delete all of it from Settings in one action. We do not train models on what you write, and we do not sell it. That is a claim you should check rather than take, which is why the delete button is one click and not an email request.

The second is accuracy. Around sixty percent were concerned about inaccurate or biased outputs, and about the absence of rigorous testing. This is the hardest one to answer honestly, because it is true of every product in this category including ours. A language model states false things with the same confidence it states true ones. What can be done is narrowing what the thing is allowed to do: Aura does not diagnose, does not name conditions, and does not touch medication, because those are the places where a confident wrong answer does real damage. She is built to ask before she teaches, and to be a place to think out loud rather than an authority to defer to.

The third is the therapeutic alliance, and it is the one we have least to offer on. The relationship is the part of therapy that the evidence supports most strongly, and it is made of warmth, attunement, being known over time, and a person choosing to sit with you. An AI has memory and availability. It does not have the rest of it. Anyone selling you an AI as a replacement for that relationship is selling you something they cannot deliver.

What the survey also shows is why people are doing it anyway. Waiting lists are long. Sessions cost money. The worst hour is usually not during office hours. A tool that is there at three in the morning and remembers what you said last week is not competing with a good therapist. It is competing with nothing, which is what most people have at three in the morning.

That is the honest position. Not a replacement, not a treatment, not a clinician. Something that is there in the gap, that does not forget you, and that tells you plainly when what you need is a person.

The same psychologists, notably, are using these tools themselves. Around half reported using AI at work for notes and correspondence. The profession is not reflexively hostile to the technology. It is specifically worried about it sitting across from a patient unsupervised, which is a narrower and much more reasonable objection than the headlines suggest.