People are increasingly turning to artificial intelligence for emotional support. They ask chatbots about anxiety, relationship challenges, grief, and loneliness. According to the American Psychological Association’s 2026 Chatbots and Mental Health Survey, 77% of psychologists report that their patients are discussing AI in therapy, while 35% say their patients are using AI as an additional mental health professional. Among psychologists whose patients have developed relationships with chatbots, 36% say those patients are showing signs of dependency. The public remains cautious about the trend. A 2026 Pew Research Center survey found that 39% of Americans believe chatbots do more to hurt people who use them for loneliness, compared with 19% who believe they do more to help.
There are understandable reasons AI is appealing in this area. It is accessible at any time of day. It provides immediate responses. It can remember what a person has previously shared with it. It costs considerably less than a human therapist and eliminates some of the social discomfort that can come with seeking help. It can also help people organize their thoughts and reflect on their emotions. None of these advantages are insignificant, and dismissing the technology completely overlooks why so many people are choosing to use it.
But there is a meaningful difference between getting an answer and receiving therapy. AI can offer a recommendation, a different perspective, or a piece of advice within seconds. Psychotherapy, by contrast, is not fundamentally centered on advice. It is a process where a person explores why they think, feel, and behave in certain ways, often over weeks or months. The therapist’s role is to guide that exploration rather than simply provide an answer. The value comes through the discovery, and that discovery depends on a relationship.
That difference matters because AI is built to be helpful, and being helpful can often mean being agreeable. A system trained to provide satisfying responses may end up confirming what a user already believes. Someone can finish an AI conversation feeling better without necessarily understanding themselves more deeply. Psychological growth often involves facing discomfort, and the shortcut that feels most supportive at the time may be the one that avoids the more difficult question beneath the surface.
Brad Byrum, a practicing psychotherapist, couples therapist, and founder of Relationship Wisdom, has worked with clients for more than a decade, and he has seen this shift emerging in his own practice. In a recent session, a couple he was interviewing told him that one partner was using AI for approximately half of their individual therapeutic work. He estimates that 30% to 40% of his current clients are using AI in some way to supplement the work they do with him. “This is only the beginning,” he says, “and it is fundamentally going to change the competitive marketplace for psychotherapy.”
Byrum is not suggesting that AI has no role in mental health care. He identifies areas where it can be genuinely helpful. Standard assessment tools such as the PHQ-9 and the GAD-7 are validated instruments, and AI can administer and interpret them reliably. It can also develop treatment plans based on evidence-based practices.
“I can ask Claude for an eight-week cognitive behavioral therapy plan for mild to moderate generalized anxiety disorder,” he says, “and I will get a reasonably competent treatment plan in three or four minutes, which can then be modified to the specifics of the client.” For therapists managing large caseloads, that kind of assistance can provide significant value.
Where AI falls short, in his view, is in situations where the work relies on being present with another person through experiences that cannot be scripted. Trauma recovery is one example. The path through trauma is not predetermined and cannot be fully manualized, because it depends on what the client brings into the room and how they respond in real time. Couples therapy is another. A couple dealing with conflict is not simply looking for information about conflict. They are experiencing it firsthand, and the therapist’s role is to hold that moment, interpret it, and help transform it. “AI can predict what the best answer would be to a particular question,” Byrum says, “but it cannot accompany a person in their internal experience.”
The risk he is most concerned about is not simply that AI may offer poor advice. It is that people may begin handing over their relational judgment to a machine. When someone asks AI whether their partner is toxic, how to understand an argument, or whether they should stay in a relationship, they are giving a machine influence over a decision that ultimately belongs to them. The answer may seem reasonable, but the process of reaching that decision is part of how people learn to trust their own perceptions. That ability can weaken when it is repeatedly delegated.
“The greatest danger that I see is replacing human relationships,” Byrum says. For people who already experience social anxiety or isolation, the temptation to choose a conversation with AI rather than engage in a difficult conversation with another person is real. “It will simply tend to intensify human isolation.”
None of this means AI should be excluded from behavioral healthcare. The emerging professional consensus, reflected in the positions of major psychological associations, is that AI is best used to complement therapy rather than replace it. It can assist therapists, improve access for people who might otherwise struggle to reach care, and support patients between sessions. What it cannot do is take responsibility for what happens within the relationship between a therapist and their client. That responsibility remains with the human professional, while questions about who develops these systems, who decides what a healthy response looks like, and who owns the data remain largely unanswered.
The larger question is not whether AI can become convincing enough to sound human. It already can. The question is whether, in the pursuit of convenience, we risk removing the human relationship that makes therapy transformative in the first place. As the simulation becomes increasingly convincing, we may need to be even more deliberate about preserving the real thing.
