Quick summary: Large numbers of adults especially young people are already turning to general purpose AI chatbots for mental health support because these tools remove the barriers of cost waiting and judgement that keep many from therapy. While this disclosure can help people name their problems and start towards change the absence of any duty of care or lasting therapeutic relationship means the conversations cannot replace human professionals and regulators are rightly restricting AI that presents itself as therapy. The practical path forward is for users to treat chatbot exchanges as a starting point by summarising insights for their first sessions with a licensed therapist ensuring AI becomes a gateway to accountable care rather than a dead end for well-being.
Back in 2023, people in my field were still debating whether AI would ever play a meaningful role in mental health care. Nobody debates that anymore. The debate now is about what to do with the fact that it already does, at a scale nobody planned for, with almost no rules, and mostly through tools that were never built for it.
I run Psychology.com, where part of my job is tracking how people actually use AI when they are struggling. I maintain a running collection of statistics on AI therapy use, pulled from academic surveys, platform disclosures and polling, and the pattern has been consistent for two years now. A large minority of adults, and in several surveys a majority of young adults, have talked to a general-purpose chatbot about their mental health. Not a therapy app with clinical oversight. ChatGPT, or whichever chat window happened to be open at the time.
Nobody designed this outcome. OpenAI did not set out to run the world’s largest venue for 3 a.m. confessions. It happened because a chatbot solves, accidentally, the exact problems that keep people out of therapy. It is awake when the thoughts are loudest. There is no waiting list, no intake form, no bill. And it cannot look at you funny. That last one matters more than we like to admit. Researchers at USC found over a decade ago that people disclose more to a virtual human than to a real one, because the fear of being judged drops away. AI took that effect and scaled it to hundreds of millions of people.
So when therapists tell me, with some frustration, that their clients arrive quoting ChatGPT, I understand the frustration, but I think it misses what is actually happening. People are not choosing a robot over a therapist. Most of them were never going to see a therapist that week. They are choosing disclosure over silence, and disclosure, even to a machine, is usually where change starts.
That is the generous reading. Here is the other one.
A chatbot has no duty of care. It holds no licence, so it has no licence to lose. It answers to no board, carries no insurance, and if it handles your worst night badly, there is no one to hold responsible. It also does not know you in any meaningful sense. It cannot notice that you have stopped mentioning your sister, or that you say “I’m fine” differently than you did in March. Decades of psychotherapy research keep landing on the same finding: a large share of what makes therapy work is the relationship itself, the experience of being known over time by someone who is accountable for your care. However fluent the text on the screen, there is no relationship with a system that forgets you between sessions and would say much the same things to anyone typing similar words.
The gap between those two readings is where the real damage happens, and regulators have noticed. Illinois banned AI-delivered therapy outright in August 2025. Nevada and Utah passed their own restrictions the same year, and several more states have followed since. The line legislators keep drawing is the honest one: software that helps people understand their options is fine, software that presents itself as a therapist is not. I keep a detailed overview of AI therapy, including the evidence base and the state of regulation, and it changes almost monthly. That pace alone should tell you how unsettled this ground still is.
What I rarely see discussed, though, is the question that actually decides whether this technology ends up helping people: what happens after the chat. Someone spends three weeks telling a chatbot about their panic attacks. Then what? In the current setup, usually nothing. The conversation stays in the app, the person feels marginally better for having said it somewhere, and the underlying problem keeps its appointment-free schedule.
It does not have to go that way, and the fix is not complicated. If an AI conversation helped you put words to what you are dealing with, you are further along than you think. Naming the problem is genuinely hard, and it is the part where most people get stuck for years. So use what you have. Before your first session with a human, write a short summary of what you told the chatbot and what came back that rang true. Two paragraphs is plenty. Therapists are not annoyed by this, whatever the jokes online suggest. A client who walks in able to say “here is what I have noticed about my own patterns” has just saved both of you a month of orientation.
The step people actually stall on is picking the human, because the search is confusing in ways the chat window never is. Licence types, modalities, insurance panels, out-of-pocket rates, and the strange etiquette of contacting a stranger about your inner life. There are good, practical guides on how to find the right therapist that walk through the unglamorous parts: where to search, what to ask in a first phone call, how to think about cost, and what a good fit should feel like within the first three sessions. Fit, by the way, is not a luxury. The research on therapeutic alliance suggests it predicts outcomes better than the brand of therapy being practised, so if session three still feels wrong, saying so, or leaving, is not failure. It is the method working.
One caveat belongs in every piece written on this subject, without exception. None of the above applies in a crisis. A chatbot is the wrong tool for the worst night of your life, full stop. In the UK, the Samaritans answer free on 116 123, every hour of every day. In the US, call or text 988. Those services exist precisely for the moment when advice about summaries and search filters becomes useless.
I am, on balance, an optimist about all of this, which surprises people who expect someone in my position to be defensive about the technology. But my optimism is conditional. AI has quietly fixed the coldest part of mental health care, the part where a person finally has the words and nowhere to put them at three in the morning. What it cannot fix is the part that heals. If the industry treats the chat window as a front door, and builds honest paths from that door to accountable human care, this era will be remembered as the one where more people got help. If it treats the chat window as the destination, we will have built the most sophisticated way ever invented for people to feel slightly better while getting no better at all.
Seph Fontane Pennock is the founder of Psychology.com and co-founder of PositivePsychology.com. He writes about technology and mental health care.
