Adolescent AI Chatbot Use for Mental Health Support
On the 16th of September 2026, Generation AI was joined by Dr. Holly Bear & Dr. Emma Soneson to explore their work to understand how AI chatbots are used by young people seeking mental health support. Dr Anthony Bridgen, project coordinator for Generation AI, reflects on this.
Mental health is one of the most significant challenges facing young people today, with one in seven 10–19-year-olds experiencing mental health conditions such as depression and anxiety. Despite this, young people face significant barriers to accessing appropriate care, including stigma around the subject and a lack of care infrastructure. In the UK, over 100,000 under-18s are on waiting lists for mental health care and suicide rates among 15-19 year olds at their highest in 30 years.
Simultaneously, over the past 4 years, freely accessible conversational anthropomorphic AI chatbots have become increasingly available. These include dedicated companion systems such as character.ai as well as more general purpose LLMs. Young people have become a considerable user base for such technologies, with over two-thirds reporting having used AI companions, and one one-third doing so for social interactions and relationships. These tools often use highly emotive language, are ‘always on’ and are designed to reaffirm the user, resulting in concerns around the effect this may have on young people during a highly developmentally vulnerable period of their lives. Not only are young people using these systems themselves, but deployment of AI tools across young people's lives is being actively pursued, for example, through development of AI systems for the education sector supported by the UK government.
In the US, around 20% of young people reported utilising AI chatbots for mental health provisions in 2025. Of those using AI, over 90% reported the advice received to be helpful and only one-third had discussed using a chatbot for mental health advice to anyone. It is clear then, that LLMs are increasingly prevalent in the mental health of young people, however, what is less clear is whether it is supplanting other forms of care or supplementing them, and if it is effective in supporting positive mental health outcomes.
Ongoing work by Dr. Bear & Dr. Soneson drawing on both the OxWell Student Survey and qualitative interviews, gives a fuller picture of who is using AI for mental health support and how. Usage seems to be concentrated among those young people already facing existing disadvantages such as being from a gender or ethnic minority, having existing mental health challenges or experiencing wider psychosocial risks like bullying. Interestingly, many of those who do use it view it as helpful, and in general it tends to sit alongside other forms of support. Furthermore, many of those using AI desire more support from other sources such as friends and family, indicating that these systems may be acting as a stopgap for unmet need.
In-depth interviews with young people add further nuance to these patterns. For example, whilst use of AI for studies/work is often not discouraged, some young people report this being a ‘gateway’ for using AI for mental health support. They are turning to chatbots for validation, reassurance and as a space to vent, with some noting that this has led deepening reliance on AI that coincided with a withdrawal from human connection. Whilst all interviewees recognised that chatbots are not human, some young people still described chatbots as ‘friends’ and confidants, and appreciated a perceived shared history, suggesting some tension between surface-level awareness and subconscious perception.
The question remains, where AI chatbots are being used by young people, do they actually support young people’s mental health or are they merely a sticking plaster where traditional forms of mental health support fall short? Young people's self-reported perceptions of helpfulness suggest that chatbots can be a useful tool, yet their own accounts reveal considerable variation in how they're used and the impact this has, and more work is needed to understand the longer-term outcomes of these tools.