Home | Evidence and impact | Our programmes | Grown up

Friend or foe? How young people are using AI for mental health support

Expert feature
Teenage Girl At Home In Kitchen Connecting With Friends On Social Media Using Mobile Phone
  • Renate Samson
    • By Nuffield Foundation
  • 07/07/26
  • Digital lives
    Mental health

The impact digital life has on young people has become one of the critical public policy discussions of our age.

Striking a balance between the positive and negative experiences that technology has on young people’s lives can be complex. On the one hand, young people speak of the ability to connect via social media as being beneficial to their sense of self and community. On the other we hear of the negative impacts on their mental health and wellbeing. This can include overexposure to harmful content, the addictive nature of the never-ending scroll and the impact on sleep.

The UK government’s intention to block access to social media platforms for under 16s by the start of 2027 seeks to mitigate these harms. But while progress is being made in one area of digital life, another technological challenge is coming rapidly over the hill, namely the use of generative AI and chatbots.

In my work last year with the Nuffield Foundation looking at the digital life of 14-24 year olds in the UK, I heard many of the young people speak of how they were using generative AI in their daily lives; for research and revision, organising their time or supporting with recipes and exercise routines. Some of the young people spoke of the use of AI to support with mental health and wellbeing, with examples of using generic generative AI chatbots such as ChatGPT to chat about things on their mind, share worries or ask for support with friendships.

As these conversations were taking place, following sustained use of the very same generative AI models, additional research from organisations such as Internet Matters, the Center for Countering Digital Hate and Commonsense Media were showing how conversations young people have with generative AI can turn from helpful to harmful often very quickly.

The paradoxical nature of technology as being both a help and hindrance was coming into sharp focus. Further exploration of the impact generative AI might have on young people’s mental health – as a tool to support and as a tool to harm felt urgent. The following provides an overview of what I heard and thoughts on what we all need to do next.

Not all AI tools are the same

For many young people, conversations about mental health start with generic, easily available generative AI models such as ChatGPT, Claude, Grok, Meta and Gemini, or AI companion creators such as Replika, Character.Ai, Kindroid, Nomi or Anima. These models are not specifically trained to provide mental health support.

Specific AI for mental health products, however, are coming to market all the time. These products, designed to respond in an appropriate, therapeutic manner, are built on foundation AI models and are trained using relevant therapeutic data from mental health professionals. Awareness and access to specific products may come from a referral or as part of a service offered by a school or therapist.

Often conversations with generic generative AI models have innocuous or mundane beginnings. Just as people fall down rabbit holes on YouTube or take diversions in conversations with each other, a conversation with a chatbot may not begin as a cry for help but can quickly move in that direction. Identifying and understanding this is a vital part of further research into how people are using AI large language model chatbots.

Furthermore, while some mental health chatbots are specifically trained to pose therapeutic reflective questions, generic AI models are not. This combined with the inherently sycophantic tone of a chatbot, that reinforces and validates a user’s experience and feelings, can cause additional problems. Furthermore, many AI chatbots are designed to maintain engagement. Just as with the never-ending scroll of social media, generative AI chatbots do not want you to switch off. Ending a conversation may be in your interest but it’s not in theirs.

While specifically trained mental health or medical AI tools must adhere to existing standards, regulation and legislation, generic generative AI models do not. Not because they are exempt but because legislation regulating their use simply does not exist. If we are to see continued use of generic AI models for mental health and wellbeing purposes, we urgently need legal and regulatory scrutiny.

Why young people are using AI for mental health

In some instances, the very fact that AI is not a human is the allure to young people. An AI chatbot is considered by some to be a safe and non-judgmental space, a digital stepping-stone for some into human services.

Despite the awareness and understanding that AI is not real, emotional relationships and sometimes unhealthy bonds can still be formed. This is leading to concern that young people are not only sharing their innermost private thoughts with AI, but are also trusting that the responses the AI gives in return are accurate or legitimate, creating false impressions of care, concern, support or enablement.

Young people are also impacted by character bot creator tools such as Replika, in particular with regard to mental health and emotional development relating to sex and sexual relationships. Gendered AI companions – such as female chatbots who respond to conversations in submissive ways – may, I was told, impact the development of healthy and respectful relationships by creating blind spots for young men in how they perceive and engage with young girls and women. More research is needed in this area.

The sycophantic and agreeable tone AI chatbots can have, which is not the norm for social interaction with humans, is also a concern. Young people may begin to rely on AI to validate their emotions and feelings rather than face the challenge or questions friends or family might ask. These concerns have been raised by experts in this sector and young people themselves. They told us they were worried that the instantaneous response AI provides creates an unrealistic idea of friendship and places additional burden on young people to respond at a similar speed to friends and to provide supportive responses regardless.

Privacy is essential, but how do we monitor for safety?

Striking a balance between a safe space for young people to explore their feelings, while also ensuring that data is private and secure, is a challenge. For example, paid-for services are sometimes more data secure and rights-respecting, but access to free services – which many young people need – rely on paying with personal data.

Managing data within the counselling profession remains a concern. If generative AI tools are to be part of the therapeutic process by patients or practitioners, counsellors need to vastly improve their understanding of online privacy and security of confidential conversations and personal data.

Parental controls, including how to balance parental oversight and intervention with young people’s autonomy and agency, is similarly an area worthy of exploration. Behaviours online can be a sign of wider issues such as signs of abuse or sexual or extremist grooming. Identifying issues early may be beneficial. However, the challenge of how to safeguard in a way that is ethical, rights-respecting and privacy preserving is another complex paradox in this space yet to be tackled in a way that satisfies all parties.

We all need to understand AI more – especially mental health professionals

The difficulty of evaluating mental health even by trained practitioners is another challenge. No one is perfect at it. No form of therapy can fully protect a person from harm and no service is 100% safe.

When it comes to using AI to evaluate a person’s mental health, the idea it can do so more accurately than a human is yet to be seen. But if AI is to play a role in mental health care and therapy, training is needed across the spectrum of providers.

Research is currently being undertaken by academics at the University of Kingston exploring the level of knowledge and understanding counsellors have of AI. With more than 50% of counsellors aged over 50, we may find that comprehension of AI is a mixed picture.

If the findings do show limited engagement with, or understanding of AI there will be an urgent need for those working in mental health and therapeutic care to rapidly improve their AI and digital literacy in order to improve understanding of the impact on the sector and on their patients.

If generative AI is to be a part of our lives, understanding what it is, how it works and what its strengths and limitations are is necessary for us all. Young people feel they understand it better than their parents. Parents meanwhile say they are discussing AI with their children. Regardless, AI is new and we are all on a journey of learning, the more we understand and the greater our knowledge, the better equipped we will be to make informed choices. Ensuring conversation about AI is ongoing rather than a one off will be a responsibility for us all.

Conclusion

The use of generative AI tools by young people to support their mental health is already in motion. We are still in the foothills of understanding the technology, let alone its long-term impacts.

Research is needed to understand why young people are turning to AI in the first place. This could be due to access, greater trust in AI and the benefit of AI as a non-judgmental safe space outside of cultural norms. But what other reasons are there?

There is an urgent need to understand the regulatory gap between generic AI models and specific AI tools that define themselves as mental health products. We need to ensure a level playing field. Any product that can be used as therapeutic support must be subject to the same scrutiny, accountability, standards, and regulatory requirements.

Conversations about privacy are currently missing in the AI debate. Some people think we live in a post-privacy world, and the collection of data and monitoring of online activity is the price we pay for our digital lives. Yet, despite their visibility online, this is not how many young people feel. They care deeply about their privacy, with many feeling anxious that their every move, thought, or behaviour is being monitored by the technology surrounding them and the companies providing the services.

When treating or discussing mental health concerns, practitioners are taught that confidentiality is paramount unless there is a potential or immediate risk to life. However, confidentiality or privacy is not guaranteed when using digital, or AI connected products or services.

This is a live concern for practitioners, many of whom are not sufficiently confident or digitally literate to ensure the technology they are using to engage with patients is privacy preserving or data secure. Addressing this is important, ensuring products used in mental health services are rights respecting and data secure is vital.

Though some feel we are on a slope of inevitability that generative AI will be used by young people or prescribed to them as a support for their mental health, the assumption that chatbots are natural bedfellows for talking therapy may not manifest as we expect.

While many young people will turn to AI for short-term or long-term support, we must not ignore the voices of many of the young people I spoke with who were persistent in their intention to reject generative AI, be it for ethical, environmental and privacy reasons or simply because of the concern that this specific form of AI will undermine people and harm society.

Mental health care is not straightforward and relies heavily on finding the right person with whom a service user can work with. Access to a diverse range of therapeutic services should remain. If AI is to be a part of the ecosystem, its role will need to be clearly defined and understood by practitioners and service users alike. We must, as society, ensure it is a choice, rather than an inevitability.

Additional notes

This think piece was informed by a roundtable held in November 2025 chaired by Renate Samson as part of the Grown up? Journeys to adulthood programme. The roundtable was made up of expert stakeholders from academic, research institutes, youth advocacy, mental health care, NHS and AI innovators. The focus was on understanding the landscape and exploring the risks and possibilities for AI across mental health and wellbeing for young people.

Read more:

Related content

Profile

Latest