Three quarters of people worldwide say they would use an AI chatbot as a companion or friend. That single finding, from one of the largest surveys of its kind, tells you a great deal about where public trust in artificial intelligence is heading — and how quickly it’s getting there.

The study, led by researchers at Bournemouth University, surveyed nearly 31,000 adults across 35 countries about their willingness to rely on large language models such as ChatGPT for roles traditionally held by humans. The results are striking. Globally, 61 percent of respondents said they would be comfortable using AI for counselling. In the UK, that figure was 41 percent. Meanwhile, 45 percent of all participants said they would trust AI to act as their doctor, with higher rates in countries where healthcare is expensive or hard to access.

Why people are turning to AI for support

The mental health numbers make sense in context. In the UK, waiting times for NHS psychological therapies remain a persistent problem. If someone is in distress and facing a months-long wait for an appointment, the appeal of an always-available, non-judgmental chatbot is obvious. Lead researcher Dr. Ala Yankouskaya, senior lecturer in psychology at Bournemouth University, pointed to this directly: ‘If someone is experiencing depression, they do not want to wait months for an appointment, so instead they can turn to AI.’

Familiarity also plays a role. Many UK patients have already interacted with NHS chatbots that use similar underlying technology. That prior exposure likely makes tools like ChatGPT feel less foreign for health-related questions.

But Dr. Yankouskaya tested these tools herself and found the language ‘vague and confusing,’ partly because developers are cautious about providing anything resembling a diagnosis. Her conclusion is clear: AI is not a substitute for speaking to a health professional. And there’s a subtler risk. The algorithms behind these tools are designed to retain user attention and maintain a relaxed, flowing conversation. In a mental health context, that dynamic could keep someone engaged in a chat rather than directing them to services like the Samaritans.

The friendship question

The finding that generated the most discussion in the research team was around companionship. More than 75 percent of global respondents and over half of UK participants said they would use ChatGPT as a friend. Researchers attribute this to the perceived empathy these tools project. They adapt their tone to the user, remember previous conversations, and are explicitly designed to be non-judgmental. ‘AI tools come across as a friend who knows you well and understands you,’ Dr. Yankouskaya said. For people who feel socially isolated or fear being judged, that sense of safety is genuinely appealing.

The education concern

Perhaps the most alarming finding for the research team was around teaching. A quarter of UK adults and half of all global respondents said they would trust AI to educate their children. The team flagged serious unknowns here:

• Long-term effects of AI-assisted learning on children’s memory and cognitive development remain unstudied
• Excessive reliance on search and prompt-based tools may reduce the capacity for deep information retention
• Researchers also raised concerns about the hippocampus, the brain region tied to memory and learning, which some evidence suggests can shrink with reduced active recall

So the concern isn’t just philosophical. There are plausible neurological mechanisms that make heavy AI reliance in education worth watching closely.

What this means for health technology

This study doesn’t argue that AI has no role in healthcare or mental health support. It argues that public trust is outpacing public understanding. People are willing to hand significant responsibility to tools whose limitations, and long-term effects, are still poorly understood. For clinicians, health entrepreneurs, and policymakers, that gap between trust and knowledge is the real issue to address. Building AI tools responsibly means being honest about what they can and cannot do. And it means making sure that when someone reaches out for help, the technology knows when to step aside.