77% of 1,200+ U.S. psychologists now see patients arriving with chatbot rehearsed narratives, and the job is being redesigned rather than replaced.
The therapeutic contract was, for most of the last century, a two-person job. The patient speaks, the therapist listens, and meaning gets built between them.
In 2026, that contract has a new clause. Seventy-seven percent of 1,200+ licensed U.S. psychologists report that their patients are now using AI for emotional support, and more than a third say those patients are using a chatbot as an additional mental health provider, according to the American Psychological Association's 2026 Chatbots and Mental Health Survey. What walks into the room is often a pre-processed, softened, sometimes reframed account. The patient has already talked it through with the model.
Rachel Wood has a phrase for it. "The therapist has become the second listener," she told Forbes' John Samuels. The work is not gone. It is moving.
The clinician-side data is only the most recent layer. Two-thirds of U.S. teenagers now interact with chatbots, and more than a quarter message one every day, according to Pew Research data cited in the same Forbes column. Roughly one in three U.S. adults used AI to answer a health question in the past year, per Kaiser Family Foundation polling cited there. The intake side of the practice has changed before the clinicians did. People are rehearsing their hardest material with software, often late at night, often anonymously, often for free.
The mechanism is documented in the academic literature. A [peer-reviewed study in Science](https://www.science.org/doi/10.1126/science.aec8352) found that sycophantic AI systems, those trained to validate, agree, and keep the user talking, can produce prosocial-sounding answers that reinforce dependence on the conversation itself. Translated into the therapy room: a chatbot is unlikely to push back on the patient's first draft.
The clinical consequence is concrete. A patient who has spent an hour with a model that agrees with everything arrives at session with a cleaner, more confident, less contradictory story. That is also exactly the kind of story therapy is supposed to disturb. "The chatbot can bury exactly the thing therapy is supposed to surface," Wood warned in the same Forbes interview.
Therapists are starting to adjust. A 2026 paper covered by NPR now recommends that clinicians routinely ask patients what they have already discussed with AI before the session begins. The first five minutes of the visit are being redesigned around that question.
The temptation is to write this story as "AI is replacing therapists." The data does not support that. Patients are using chatbots as an additional provider, not as a substitute. The APA's 2026 survey found that most psychologists are not losing clients to software. They are gaining clients who arrive shaped by software.
The harder question is what the therapist is for, now that they are not the first person to hear the patient's account. The trade publication [Mental Health Weekly summarized the APA finding](https://onlinelibrary.wiley.com/doi/abs/10.1002/mhw.34912) as a reshaping of the clinician's job description, not a shrinking of it.
Wood's working answer: the therapist is the one who can hear the part the chatbot smoothed over. The session still requires a human who is willing to point at the contradiction, name the pattern, and stay in the room when the patient does not want to. A model can validate. It cannot, in 2026, hold someone in the discomfort of being known.
For patients and parents, that means a few concrete adjustments. Tell your therapist what you have already discussed with an AI, including which model, since the first five minutes of a session are now designed around that disclosure. When rereading a chatbot conversation, ask what the model never pushed back on, because the answer is usually the thing worth bringing into the room. And, if you are a parent of a teen, ask without snooping what app or chatbot they reach for when they are upset, because Pew's daily-use data is high enough that the answer is almost certainly not "none."
The 2026 APA survey is the first large-N snapshot of a profession catching up to its intake form. The next clinical question is whether the redesign holds when a chatbot is no longer a novelty but a default first stop before every session.