Rising use of AI chatbots as an entry point into mental health support raises concerns about care quality and regulation

As AI chatbots become a common first step for mental health support among young people, experts warn of the risks of replacing licensed therapy and emphasise their role as preliminary tools rather than substitutes.

AI chatbots are becoming an informal first stop for mental health support, especially among younger people who may struggle to access conventional therapy. Data cited by Psychology.com suggests 6% of US adults have already used AI for therapy and 17% have considered it, while a separate JAMA Pediatrics study found that about 19% of US adolescents and young adults have used chatbots for mental health advice. The appeal is straightforward: the tools are available at any hour, cost nothing up front and remove much of the fear of being judged.

That shift is changing expectations around care. Seph Fontane Pennock, founder of Psychology.com, argues that many people are not choosing software instead of therapy so much as choosing some form of disclosure over saying nothing at all. The JAMA Pediatrics findings support that view: 43% of young users said they sought chatbot advice at least monthly and 92% found it somewhat or very helpful. Yet more than half had not told anyone they were using these tools for mental health support, suggesting that the behaviour is still largely hidden from families, clinicians and researchers.

The central concern is that a chatbot is not a clinician. Pennock says these systems have no duty of care, no licence and no lasting therapeutic relationship with the user. They also cannot track the subtle changes that a human therapist would notice over time. That distinction matters because the relationship itself is a key part of why therapy works, not just the words exchanged in a single session.

Regulators are beginning to respond. Pennock notes that Illinois banned AI-delivered therapy in August 2025, while Nevada and Utah introduced restrictions the same year. The broader policy trend is clear: software that helps people understand their options may be tolerated, but systems that market themselves as therapists are facing pushback. That concern is reinforced by the mixed evidence on satisfaction, with Psychology.com saying both clinicians and the public still rate AI below human therapists for actual treatment.

The more useful question, Pennock argues, is what happens after the chat ends. A conversation with a bot may help a person name anxiety, panic or another pattern more clearly, but it can easily stop there. His practical advice is to use the exchange as preparation for human care: summarise the key points, bring them to a licensed therapist and use them as a starting place for proper treatment. That is also where fit matters. Research on therapeutic alliance suggests that the working relationship between therapist and patient can predict outcomes more strongly than the specific school of therapy being used.

None of this applies in a crisis. Pennock makes clear that chatbots are the wrong tool for acute distress or thoughts of self-harm. In the UK, the Samaritans can be reached on 116 123, and in the US the 988 Suicide and Crisis Lifeline is available by phone or text. Those services exist for moments when a conversational AI is not enough and immediate human support is needed.

The wider lesson is not that AI should replace therapy, but that it is already acting as a gateway for people who might otherwise remain silent. If the technology is used as an entry point into accountable, licensed care, it could help more people reach support sooner. If it becomes the destination, it risks leaving users with the feeling of having talked about their problems without ever moving towards treatment.