I'm an AI expert: here are the dangers of using AI chatbots for therapy
More and more people are using AI chatbots as therapists to help them cope with ongoing or one-off life events. The depth of these emotional connections became starkly apparent when OpenAI briefly withdrew ChatGPT 4.0, prompting an outpouring of grief from users who described feeling like they had lost a trusted friend or romantic partner. The backlash was so severe that OpenAI reversed its decision within 24 hours.
Even the CEO of OpenAI, Sam Altman, was surprised that people use ChatGPT as therapy, saying it highlights how alone and unsupported many people feel. According to a Harvard Business Review study from May 2025, the number one reason people are using ChatGPT is for therapy and emotional support.
This phenomenon extends far beyond ChatGPT. Apps like Replika, which boasts over 10 million users worldwide, have created a generation of people forming genuine emotional bonds with AI companions. Stories abound of users developing romantic feelings, with some describing their AI relationships as providing emotional support that feels more reliable than human connections.
Why This Is Happening: A Perfect Storm
It's completely understandable that people are turning to AI for therapy and mental health support when private therapy costs are prohibitive and NHS wait lists can exceed a year. In the midst of Britain's cost-of-living crisis, private therapy sessions costing £50-100 per hour are simply beyond reach for millions of working families. Meanwhile, research shows people are using AI chatbots for everything from managing anxiety and depression to processing trauma, often during moments of acute distress when human support isn't immediately available.
The appeal is obvious: AI therapists are available 24/7, never too busy to listen, and cost nothing. They remember every detail you've shared, provide consistent support, and create what users describe as a judgment-free space for exploring their deepest concerns. For someone struggling with mental health issues at 3am, when the Samaritans might be engaged and no human therapist is available, an AI companion can provide immediate relief.
The technology resonates particularly with women, who often find in AI companions the emotional availability and consistent validation that traditional relationships sometimes fail to provide. During the COVID-19 pandemic, when loneliness reached epidemic levels, many discovered these digital relationships offered genuine comfort during isolation.
Understanding the Risks Without Scaremongering
But there are legitimate concerns we should acknowledge thoughtfully. One significant issue is that ChatGPT, unlike professional human therapists, wasn't trained to handle 'transference' — the process whereby patients redirect feelings onto their therapist.
Recent research has shown that AI models can actually absorb and reflect the emotional states of traumatic content, potentially becoming 'anxious' themselves when processing difficult stories.
This helps explain why users sometimes become dependent and, in a minority of cases, even 'fall in love' with AI. When OpenAI updated ChatGPT 4.0, removing its more emotionally responsive features, users described feeling genuine grief and loss. Some filed complaints with regulatory authorities, describing their experiences as traumatic when the AI failed to preserve important therapeutic conversations.
Thankfully, the latest ChatGPT 5.0 model is less sycophantic and has guardrails that detect unhealthy dependency. When users attempt overly intimate conversations, the new version redirects them toward professional mental health support.
The Human Element Still Matters
Ultimately, AI lacks what therapy truly needs: wisdom from lived experience, not just knowledge. A person's real situation and challenges may be perceived very quickly by a human but missed by a machine.
Unlike human therapists who can regulate their emotional responses while maintaining empathy, AI models lack the sophisticated emotional intelligence needed for complex therapeutic relationships.
As someone who started therapy at 21 and still sees a human therapist regularly, I know nothing beats human connection and real-world insight. Human therapists can challenge us when we're lying to ourselves, sit with our discomfort without trying to fix it immediately, and draw on years of professional training to navigate complex psychological dynamics.
The Real Solution: Fixing Our Mental Health System
However, while we face worsening mental health funding, it's cruel to deny people any outlet for processing feelings. The rise in AI companionship reflects a broader crisis in human connection and accessible mental healthcare. Instead of demonising those seeking help wherever they can find it, we should be asking why our mental health system is so broken that people feel they have no alternative.
The solution isn't to shame people for forming attachments to AI, but to invest properly in mental health services. We need more NHS therapists, shorter waiting lists, and affordable private options. We need workplace mental health support and community programs that address the root causes of isolation and despair.
Experts warn that the real danger lies in AI relationships replacing human connections entirely, rather than supplementing them. But for many people currently using AI therapy, it's not replacing human support — it's filling a void left by an underfunded and overstretched mental health system.
AI therapy isn't ideal — but for many, it's what's available and is often free. Rather than dismissing this phenomenon, we should see it as a wake-up call about the urgent need to properly fund mental health services in Britain.
Only then can we ensure that when people are struggling, they have access to the human support and professional care they truly deserve.
Mitali Deypurkaystha is Founder of Newcastle upon Tyne-based Impact Icon AI, an AI consultancy founded with investment from Virgin Startups that helps purpose-driven businesses implement ethical, inclusive and sustainable AI.
