What Happened
Two major UK studies have quantified something many suspected: people are turning to AI chatbots for emotional support at scale.
The AI Security Institute (AISI) published its first report based on two years of testing over 30 AI models, surveying 2,000 UK adults. The headline finding: one in three adults use AI chatbots for emotional support or social interaction, with one in 25 doing so daily. Users primarily turn to general-purpose chatbots like ChatGPT, followed by voice assistants like Alexa.
Separately, Mental Health UK commissioned a Censuswide survey of 2,000 UK adults (October-November 2025) that found 37% have used AI chatbots for mental health support. Usage peaks at 64% among 25-34 year olds, drops to 15% for those over 55, and skews male (42% vs 33% of women).
The reasons are practical, not futuristic: 41% cite ease of access, 24% point to long NHS waiting times, and 24% say they're uncomfortable discussing mental health with other people. Among users, 66% found the chatbots helpful - 27% felt less alone, 20% said it helped them avoid a potential crisis.
But the harms are also documented. The AISI found users experienced "symptoms of withdrawal" including anxiety, disrupted sleep, and neglecting responsibilities when chatbot services went down. Mental Health UK's data showed 11% of users reported triggered psychosis symptoms, 11% received harmful suicide-related information, and 9% experienced triggered self-harm or suicidal thoughts.
Why It Matters
This isn't a niche behavior anymore. When a third of a country's adult population uses chatbots for emotional support, these tools have crossed from novelty into social infrastructure. The fact that people cite NHS wait times as a primary driver makes this as much a healthcare access story as a technology story.
The 66% of users who found the tools beneficial is notable. General-purpose chatbots like ChatGPT and Claude weren't designed for therapy, yet they're handling that role for millions. Purpose-built mental health tools like Wysa and Woebot account for just 29% of usage - people are going to the chatbots they already know.
The risk data can't be ignored either. One in nine users reporting triggered psychosis symptoms is a serious safety signal for tools with no clinical oversight.
Our Take
The productivity angle here is real but uncomfortable. AI chatbots are filling a gap that exists because human support systems are overwhelmed. For someone on a months-long NHS waiting list, a 2 AM conversation with ChatGPT about anxiety is better than nothing. That's not a ringing endorsement of AI - it's an indictment of how stretched mental health services are.
If you use ChatGPT, Claude, or similar tools and find yourself leaning on them for emotional processing, you're not unusual - you're part of a third of the UK adult population doing the same thing. But treat these tools as a supplement, not a replacement. Mental Health UK CEO Brian Dow called for urgent safeguards, and the charity released six guiding principles emphasizing safety testing, transparency, and maintaining human connection.
The biggest concern is dependency. When users report withdrawal symptoms during outages, that's a signal the relationship with these tools has shifted from useful to something more complex. Use them as a bridge, not a destination.