New research published by mental health charity Mind in July 2026 suggests growing numbers of people are turning to AI chatbots to support their mental health, with some relying on them instead of formal services, where inaccessibility, long waiting lists and rigid criteria for getting support are commonplace.
The report found that one in five (18%) people had used AI chatbots to support their mental health in the previous 12 months. Among that group, three in five (60%) said they had used chatbots in place of formal support, such as NHS talking therapies or counselling (Mind, 2026).
Of those who had used AI in place of formal support, around a third (31%) said they preferred it, while one in ten (11%) said existing formal support options did not meet their needs (Mind, 2026).
Earlier this year, when I was detained on an acute psychiatric ward with psychosis, I became one of those people. During my admission, I used ChatGPT to understand medication side effects, prepare for psychiatric conversations, collate evidence on ward conditions, track sleep and energy patterns, and process the meaning of a major episode.
ChatGPT gave me continuity, language and a way to interrogate my care. It also flattered, over-validated and mirrored my framing. Ultimately, however, it responded to me as a person asking a question, rather than first and foremost as a bipolar patient whose credibility had already been eroded.
My experience isn’t unique. On Reddit, one user wrote that AI had been “incredibly helpful to have [. . .] listen to me the way a friend would”. Further down the same thread, another user added, “I’ve had more connections and positive interactions lately IRL because of the ‘therapy’ I am having with AI.” As these examples show, AI is being used as a substitute for peer support or traditional talking therapy.
Perhaps the most interesting question isn’t why people are turning to AI chatbots, but what they’re looking for when they do, and what kind of support is proving difficult to find elsewhere.
In my experience, clinicians have often been quicker to diagnose than to explain. Sometimes time pressure leaves no room for follow-up questions. Meanwhile, prescribing decisions and the cadence of appointments can chip away at personal agency, sometimes leaving people waiting months to discuss debilitating side effects, let alone a dose reduction.
I’ve also had physical illness interpreted through the lens of psychiatric diagnosis, in ways that caused distress. A few years ago, I developed acute hyponatraemia following a gastro virus. Although the illness itself wasn’t caused by my bipolar diagnosis, my diagnosis shaped how the episode was understood afterwards. The focus shifted towards psychiatric explanations and assumptions about my behaviour, rather than simply responding to the experience I was describing. That weight of judgement is something I haven’t experienced with AI.
Now, AI doesn’t fix all those problems. How could it? But it does bring us closer to a style of interaction that makes room for curiosity.
In practice, the support AI chatbots provide can feel expansive. They create the space for people to ask question after question, rather than restricting use to a 10-minute appointment. They retain a thread of personal history – vital context that can get lost between ward staff, psychiatrists, community teams and GPs. And while these systems can make their own flawed inferences about personal insight or tone of voice, these aren’t recorded in clinical notes that may shape how the next professional perceives you.
Of course, AI chatbots can also become liabilities, and their unregulated use might be actively dangerous or detrimental to wellbeing in some cases. They will keep answering questions at 2am – a time when you might need sleep more than deepening analysis. Not all questions should be followed indefinitely. And there are times when we need to be challenged, not encouraged to continue a conversation. Equally, for someone experiencing or recovering from psychosis, engaging with an AI chatbot that’s constructing coherent narratives from fragments can be risky when pattern-making is already in overdrive.
The mental health sector is right to ask what AI can do to distressed people, but it must also be brave enough to ask what mental health services have done that makes distressed people seek out AI. Instead of viewing people’s use of AI as a troubling deviation from proper care, mental healthcare providers should pay closer attention to the human needs AI is meeting, whether that’s giving access to explanation, providing intellectual agency or taking people’s accounts of their own experiences seriously.
Yes, AI is an imperfect and potentially dangerous source of mental health support. It can give wrong information, distort conversations and encourage rumination. But it has much to teach us about better care.
About Hayley
Hayley is a writer and researcher interested in culture, tech and mental health. She’s created the library of mental health resources she wished she had after first being sectioned.