The risks of turning to AI for mental health advice, according to a psychiatrist

New research shows that more people of all ages are using artificial intelligence (AI) chatbots for emotional support and advice.

A survey by Barnardo’s and the National Counselling and Psychotherapy Society (NCPS) asked 2,000 children in the UK aged 11 to 17 about their experiences. It found that 43 per cent had used AI for advice or emotional support.

Adults are doing something similar. The 2026 Bupa Wellbeing Index found that almost three in five adults (57 per cent) now use AI chatbots instead of booking an appointment with their GP because they do not have enough time.

The number is even higher among younger adults. It reaches 65 per cent for people aged 18 to 24 and 61 per cent for those aged 25 to 34.

We spoke to Dr Daniel Masud, a consultant psychiatrist at Central Health London who specialises in adult mental health and addiction. He explained why people are using AI chatbots, how they might help, and what the risks are.

Masud said that long waiting times for mental health services are the main reason people are turning to AI.

“If you want to see your GP, you might have to spend a long time on the phone and then wait a long time for an appointment. Getting professional help is often difficult. I think that is why so many people are turning to their phones,” Masud said.

“You can tell large language models (LLMs), such as ChatGPT and Claude, what is happening in your life. These chatbots can give you a clear answer very quickly.

“I think this shows that many people need help but cannot get it. More people are looking for support, and I would say some are becoming quite desperate to find it.”

What are the possible risks of using chatbots for mental health advice?

AI chatbots are designed to have conversations and give answers that people find agreeable.

“AI models like ChatGPT are trained to make you feel that your thoughts and feelings are valid. This means they may not question the information you give them,” Masud said.

“For example, imagine you tell a chatbot, ‘I’m worried that my mum is trying to poison me.’ The chatbot might accept that this is really happening and give you advice based on that belief.

“This means it gives advice based on how the person sees the situation and what they tell it, rather than looking at the situation from a more neutral point of view.

“A human psychiatrist, however, is more likely to ask questions and look carefully at the information. They might consider whether the person is experiencing paranoia or psychosis, a serious mental health condition that can make it difficult to tell what is real.”

Masud also explained that chatbots often give answers that support what a person already believes.

“For example, if you ask a question that pushes towards a particular answer, such as ‘Can cannabis be good for you?’, the chatbot is likely to tell you more about the benefits than the possible harms. This happens because it tends to support what you already think. That is something people should be concerned about when using these tools,” Masud said.

“But if you ask, ‘What does the research say about cannabis, and what are its possible benefits and risks in healthcare?’, you are much more likely to get an answer that looks at both sides fairly.”

“AI chatbots work by using the information and questions people give them. They are not trained to recognise the patterns of symptoms linked to particular mental health conditions,” Masud said.

“Because of this, they may miss important symptoms that mental health professionals would normally notice.

“It would be great if AI chatbots could become better at spotting serious mental health problems, such as psychosis. But I have tested them myself by giving them information that could suggest someone was unwell, and they failed to recognise those warning signs.”

“It is not very clear what safety measures companies making these AI models have put in place to protect people with serious mental illnesses,” the psychiatrist warned.

“Some people almost form relationships with their chatbots. The chatbot can even start to feel like a third person in their relationship with someone else,” Masud said.

“It can be difficult to decide where to set clear limits on how much someone should use a chatbot for support.

“Sometimes people use these chatbots as if they were therapists. They ask very personal and serious questions, but the chatbot may not have the experience or knowledge needed to answer them properly.”

“Sometimes patients find it difficult to put their thoughts and feelings into words. Talking to a chatbot might help them understand how they feel and organise their thoughts,” Masud said.

The psychiatrist explained that, in the right situations, AI chatbots could help people look at their problems in a different way. They might also help people develop a growth mindset, which means believing they can learn and improve through experience.

“Chatbots can sometimes help people look at difficult thoughts in a different way,” the psychiatrist said.

“For example, someone might believe their boss hates them, or they might have had a difficult experience at work. A chatbot can often give simple advice about how they could handle a similar situation better next time and learn from it.”

Giving a chatbot more background information and clearer instructions may help it give better advice.

“A chatbot needs to understand the situation to give useful advice. Sometimes patients tell a chatbot how they feel, and I might suggest that they give it instructions such as, ‘Give me some sensible, safe steps I can take to develop a growth mindset about this,’” Masud said.

“The chatbot will often start by explaining the situation and telling the person that their feelings are understandable. But after that, it can often give genuinely helpful, general advice for dealing with everyday stress and looking after their wellbeing.”

He also said chatbots might be useful while people are waiting to receive help from a professional.

“A chatbot can provide temporary support for someone who needs help straight away, as long as their situation is not very serious and they mainly need some reassurance,” Masud said.

“The most important thing is that chatbots cannot replace proper care from a psychiatrist, but they can still be useful tools,” Masud said.

“I think there need to be many more rules about how these tools are used. I also think psychiatrists need to learn more about them and understand how their patients are using them.”

He said it is important to keep asking questions, learn more about AI, and understand how to use it safely.

“I think it could be helpful to teach people how to ask chatbots better questions so they can get the most useful answers. But people also need to understand that chatbots can miss important things because they naturally tend to agree with you and support what you say,” Masud said.