Should you turn to AI for mental health advice? Here are the risks to consider

Research suggests that people of all ages are increasingly turning to chatbots for advice and emotional support.

A joint survey conducted by children’s charity Barnardo’s and the National Counselling and Psychotherapy Society (NCPS), involving 2,000 children aged 11 to 17 across the UK, found that 43% had used AI for advice or emotional support.

Similarly, the 2026 Bupa Wellbeing Index revealed that nearly three in five adults (57%) are now turning to AI chatbots instead of booking a GP appointment due to time constraints. This figure rose to 65% among 18–24-year-olds and 61% among those aged 25–34.

We spoke to Dr Daniel Masud, a consultant psychiatrist at Central Health London who specialises in general adult psychiatry and addictions, to explore the reasons behind this trend and to examine the potential benefits and risks of using chatbots for mental health advice.

What do you think is driving people to turn to AI chatbots for mental health advice and emotional support?

The psychiatrist thinks that the biggest driving factor is long waiting times for mental health services.

“If you go and see your GP, you might have to spend lots of time on the phone and wait a long time for an appointment. Professional support is often quite difficult to get, so I think that’s why lots of people are turning to their phones,” says Masud.

“You can tell large language models (LLMs) like ChatGPT and Claude what’s going on and these chatbots will get something coherent back to you very quickly.

“So, I think it’s a symptomatic of there being unmet need and an increased demand, and I would say that people are getting pretty desperate for support.”

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

Chatbots can be too validating and not objective enough

The models are designed to be conversational and agreeable.

“LLM models like ChatGPT are trained to validate you, so won’t necessarily question the information you are giving it,” says Masud.

“For example, if you told it ‘I’m worried that my mum is trying to poison me’, it would genuinely believe that that was the case, and would advise you accordingly.

“So, it’s gives advice through the prism of what that patient’s seeing/telling it rather than having a more objective view.

“Whereas, a human psychiatrist is more likely to be a little bit more curious and inquisitive about the actual information that is being given, and query whether or not this person might be paranoid or potentially psychotic.”

The psychiatrist adds that chatbots also have a tendency to lean into your own bias.

“If you ask it a question which is quite leading, such as “Can cannabis be good for you?”, for example, it’s likely that the chatbot will come up with more positives than relevant negatives. This is because it’s more likely to lean into your own bias, and that’s one of the things that people will be concerned about when they’re using it,” explains Masud.

“Whereas, if you were to say, “What’s the evidence base for cannabis, and what are the positives and negatives in healthcare to consider?” it’s much more likely to give you a much more balanced output.”

Important symptoms can be missed 

“These AI chatbots rely on logic and the prompts that they are given and are not trained to recognise patterns of specific mental health conditions,” says Masud.

“So, what sometimes ends up happening is that important symptoms that would have otherwise gone through the conduit for mental health services are  missed.

“It would be great if LLMs were able to better detect serious mental health problems, such as psychosis, but I’ve tested it myself and put in prompts that could be suggestive of illness, but it didn’t manage to pick those up.”

Not enough safeguarding

“There isn’t a lot of clarity about what safeguards are in place for severe mental illnesses from these large language model companies,” warns the psychiatrist.

Risk of developing a pseudo-relationship with the chatbot

“Some people are almost develop relationships with their chatbots, where it can become like a third person in a relationship,” says Masud.

“It’s quite difficult to establish firm boundaries around the appropriateness of how much somebody’s using that model to support them.

“People will sometimes use these chatbots for therapy and will ask it lots of quite deep questions, that the chatbot will not necessarily have the experience or answers to.”

Do you think that using AI chatbots can be helpful for generic wellbeing advice for managing everyday stress or low mood?

“Sometimes patients don’t really feel like they’ve got the language to be able to coherently tell you what’s going on or what they’re thinking and feeling, and talking to a chatbot might be able to help them understand their own perspective and map those feelings out,” says Masud.

The psychiatrist says that AI chatbots have the potential to help people reframe their thinking and adopt a growth mindset under the right circumstances.

“Chatbots can sometimes help people reframe thoughts that are difficult that they might be having,” says the psychiatrist.

“For example, if they think their boss hates them, or they have encountered a difficult scenario at work. The chatbots can often offer some light advice about how they might handle that situation next time and grow from that.”

Providing chatbots with context and prompts might help improve the quality of the advice.

“It has to have context to give you sound advice. Sometimes patients will tell the chatbot how they are feeling and I might ask them to give it prompts such as ‘give me some reasonable, safe steps to take to adapt a growth mindset to this’,” says Masud.

“Often the chatbot then will give a bit of a spiel about what’s happening and do a validation piece, but then often gives some really helpful generic advice for general stress and wellbeing situations.”

He notes that chatbots can a useful tool while people are waiting for professional support.

“It can act as a stopgap for someone who needs immediate support who might not be in a very serious situation but might just need some reassurance,” says Masud.

In conclusion:

“The main point is that these chatbots don’t replace proper psychiatric care, but can certainly be a useful tool,” says Masud.

“I think that there needs to be much more regulation around it and I also think that psychiatrists need to engage with it more and understand how their clients are using it.”

He says staying inquisitive and learning how to use it safely is key.

“I think it might be good to teach people how to prompt it to get the most value from it, while also recognising where things can get missed because of the chatbot’s natural propensity to validate you,” says Masud.