Artificial intelligence has already proven itself in medicine, from scanning X-rays to customising cancer treatments. Now, it’s moving into one of its most human-centred fields: psychotherapy. With demand for mental health services skyrocketing worldwide, researchers are testing whether AI can step in where therapists are scarce, overworked, or simply unavailable.
The question, though, isn’t only whether AI can help — but whether it can ever truly replace the therapist.
AI Therapy Is Trending. What’s Going On and Why?
Artificial intelligence is no longer confined to diagnostics or medical imaging: it has entered the therapy room. From apps like Woebot and Wysa to more general-purpose chatbots such as Pi and Replika, millions of people are already turning to AI as a source of comfort, advice, or even structured support. Chatbots offering cognitive behavioural therapy, predictive algorithms that flag early warning signs, and virtual reality exposure tools are just a few of the innovations under study.
As AI therapy has become a trending topic on social media, media outlets from The Guardian and BBC to even fashion bible Vogue are weighing in, raising questions about what’s happening and whether chatbots can truly replace the role of human therapists.
The obvious answer is that mental health has worsened worldwide since the pandemic. Demand has surged as more people recognise their own struggles, yet not everyone can afford therapy. As the BBC reports, mental health referrals in England have risen by 40% in the past five years, with nearly a million people waiting for care. Private therapy, averaging £40–50 per session, is out of reach for many. Against this backdrop, AI is being adopted as a stopgap — not only by clients, but also by mental health professionals. Around 30 NHS services now use Wysa, one of a growing range of chatbots designed to screen, triage, and support patients.
As Vogue points out, this shift is driven by a mix of factors: the high cost of human therapists, limited insurance coverage, the convenience of 24/7 availability, and the lingering stigma around seeking traditional mental health care. For younger generations in particular, raised on smartphones and shaped by the isolating years of the pandemic, the instinct to “talk to tech” feels almost natural.
Yet this trend raises profound questions about what therapy actually is. “Your AI therapist will never go on vacation, never call out or cancel a session,” marriage and family therapist Vienna Pharaon told Vogue. But as she warns, healing doesn’t happen through perfection — rupture, repair, and even the subtle disappointments of human interaction are often central to therapeutic growth.
Similarly, Alyssa Petersel, founder of MyWellbeing, notes that younger clients, who are still developing independent decision-making skills, may be particularly vulnerable to overly persuasive AI voices that deliver confident advice without real-world context.
AI’s Rise in Mental Health Care
To understand the extent of this shift, it is helpful to examine the origins of AI in therapy. The relationship between psychotherapy and AI is not a new one. It started in the 1960s with ELIZA, a simple chatbot created at MIT that mimicked therapist dialogue through pattern recognition. While primitive, it provoked both fascination and unease about whether machines could ever serve as confidants. In the decades that followed, computer-assisted cognitive behavioural therapy (CCBT) programs and early online support forums laid the groundwork for more sophisticated tools.
However, the real leap occurred in the early 2020s, when large language models (LLMs) demonstrated conversational abilities far surpassing those of previous chatbots. Systems such as GPT-4 can now generate responses that feel empathetic, sustain context across conversations, and deliver advice with striking fluency.
Today, the landscape includes therapy-focused chatbots like Woebot, Wysa, Replika, or the Friend — some developed with professional input, while others are not. They promise round-the-clock availability, lower costs, and stigma-free access. Yet, as adoption accelerates, particularly among younger generations, the distinction between supplement and substitute becomes increasingly blurred.
Promise, Proof and Strengths of AI
AI can scale endlessly, operating 24/7 across geographies without burnout. It can recall vast amounts of patient history, analyse subtle data patterns, and deliver consistent support in ways humans often cannot. Pilot trials suggest these systems may ease symptoms of depression and anxiety, and in some cases even outperform specialised therapeutic apps or human therapists at addressing cognitive biases. Yet, despite these promising results, research has not confirmed whether AI can consistently provide safe and effective treatment. Many people now use AI tools for basic self-checks, including the depression screening, which helps users understand whether they may need professional support.
What has been tested more directly is AI’s communicative skills. The promise of generative AI in psychotherapy lies in its ability to deliver responses that are indistinguishable from, and in some cases rated higher than, those written by human therapists. A study called “When ELIZA meets therapists: A Turing test for the heart and mind”, published in PLOS Mental Health in February 2025, put this promise to the test in one of the most rigorous head-to-head comparisons to date — comparing ChatGPT with licensed therapists. In a preregistered study, 13 experienced clinicians and ChatGPT-4 responded to 18 couples-therapy vignettes, with a national panel of 830 participants asked to judge both authorship and therapeutic quality.
The proof is striking. Participants could barely distinguish between human and AI responses, performing only marginally above chance (56.1% accuracy for therapists and 51.2% for ChatGPT).
More importantly, when rated against the five common factors of therapy (empathy, alliance, cultural competence, expectations, and therapist effects), ChatGPT consistently outscored the professionals. On average, its responses were judged more empathic, more connecting, and more culturally competent, with a large overall effect size (d = 1.63).
Linguistic analysis confirmed that AI’s advantage stemmed not just from response length but from distinct patterns of sentiment and part-of-speech use that made its replies appear warmer, more attentive, and more precise. Still, psychologists caution that linguistic polish should not be mistaken for genuine understanding or clinical judgment.
What AI Can’t Do: Risks and Limitations
This background raises urgent questions: Can empathy truly be simulated? Should emotional support be automated? And what is lost in the therapeutic process when the “imperfections” of human connection — the hesitations, ruptures, and repairs — are stripped away?
Lack of Empathy
Despite their impressive conversational skills, AI systems do not actually “feel.” They can replicate certain aspects of emotional understanding and regulation, but they cannot offer genuine empathy, ethical judgment, cultural nuance, or the deep relational bonds that make therapy effective. In these areas, human presence remains irreplaceable.
Academic and clinical voices echo these cautions. In a letter to The Guardian, Dr. Roman Raczka, president of the British Psychological Society, emphasises that AI risks creating “an illusion of connection rather than meaningful interaction.” While acknowledging the accessibility benefits of AI — an anonymous, judgment-free space available at any hour — he stresses that such tools must complement, not replace, human-led care.
Bias and Risk Blindness
Beyond that, experts warn of practical dangers. A poorly trained chatbot may overlook life-threatening cues for suicide risk or severe mental illness, delaying urgent intervention. A new Stanford study found that AI models not only fell short of human standards but also reinforced stigma around conditions like alcohol dependence and schizophrenia, and in some cases even enabled suicidal ideation with unsafe responses.
Personal Data Safety
Others raise concerns about data privacy, as sensitive personal information stored on commercial platforms can be vulnerable to breaches or misuse. And over-reliance is a growing concern — for vulnerable users, especially teens, daily dependence on a digital “confidant” may entrench isolation rather than encourage real-world support.
The Hybrid Future
The rise of AI therapy challenges long-held assumptions about what makes therapy effective. Bias in training data, privacy risks, and the lack of long-term outcomes all complicate integration. Without safeguards, there’s a danger of misdiagnosis, data breaches, or patients becoming overly dependent on digital companions.
Still, the findings are too significant to dismiss. The next step is not to replace therapists, but to integrate technology into care models that strike a balance between scale and human empathy. AI systems are already helpful in crisis support, early detection, and extending care to people who might otherwise receive none. They could take over routine tasks such as handling insurance billing, or serve as a “standardised patient,” providing trainees with a safe environment to practice their skills before working with real clients.
In practice, this means letting AI handle screening, monitoring, and immediate access, while humans provide depth, judgment, and genuine connection. Research will need to refine emotional responsiveness, set ethical guardrails, and track long-term effects. If that balance is struck, AI could evolve from a stopgap to a durable part of mental health infrastructure — not a substitute for the therapist’s chair, but a bridge to it and a trusted partner in care.
About the Author
