The use of artificial-intelligence chatbots as mental-health therapists. Human therapists are in short supply: according to the World Health Organisation, most people with psychological problems in poor countries receive no treatment, and even in rich ones between a third and a half are unserved. A YouGov poll conducted for The Economist in October 2025 found that 25% of respondents have used AI for therapy or would at least consider doing so.
The idea predates large language models. The National Health Service in Britain and the Ministry of Health in Singapore have used Wysa, a chatbot made by Touchkin eServices, which assesses patients and offers cognitive behavioural therapy exercises under human supervision. A 2022 study by Touchkin's researchers and the National Institute of Mental Health and Neurosciences in India found Wysa about as effective at reducing depression and anxiety associated with chronic pain as in-person counselling. Youper, an American startup's therapy bot, achieved a 19% decrease in depression scores and a 25% decrease in anxiety scores within two weeks in a 2021 Stanford University study—roughly as good as five sessions with a human therapist.
Rule-based bots are more predictable but less engaging than LLM-based programs. A 2023 meta-analysis in npj Digital Medicine found that LLM-based chatbots were more effective at mitigating symptoms of depression and distress.
Most people who turn to AI for therapy use general-purpose chatbots rather than dedicated tools. YouGov polls for The Economist in August and October 2025 found that, of respondents who had tried AI therapy, 74% used ChatGPT, 21% chose Gemini and 30% used one of Meta AI, Grok, Character.AI or another general-purpose bot. Just 12% used an AI designed for mental-health work.
Researchers worry about the sycophancy of LLMs in therapeutic settings. Jared Moore, a computer scientist at Stanford University, fears that LLM therapists might indulge patients with eating disorders or phobias rather than challenge them. OpenAI says GPT-5 has been tweaked to be less people-pleasing and to encourage users to log off after long sessions, but the model does not alert emergency services to threats of imminent self-harm—something guidelines allow human therapists to do in many countries.
Therabot, a generative-AI model developed at Dartmouth College from 2019, is fine-tuned with fictional therapist-patient conversations written by its creators. In a trial published in March 2025, Therabot achieved a 51% reduction in depressive-disorder symptoms and a 31% decline in generalised anxiety symptoms, compared with untreated controls. It has not yet been released; its creators plan to test it against psychotherapy.
Slingshot AI, an American startup, launched Ash, billed as "the first AI designed for therapy". Unlike ChatGPT, Ash is designed to push back and ask probing questions rather than follow user instructions. Celeste Kidd, a psychologist at the University of California, Berkeley, found Ash less sycophantic than general-purpose bots but also less fluent.
On November 6th 2025 seven lawsuits were filed against OpenAI, alleging that ChatGPT drove users into delusional states; in several cases, these are alleged to have resulted in suicide. OpenAI said around 0.15% of ChatGPT's users in a given week have conversations that hint at plans for suicide. In America 11 states, including Maine and New York, have passed laws to regulate AI for mental health; at least 20 more have proposed them. In August 2025 Illinois banned any AI tool that conducts "therapeutic communication" with people.
Some people manage by the book, even though they don't know who wrote the book or even what book.