You open your phone at 2 a.m., unable to sleep. Racing thoughts. A familiar tightness in your chest. You type “I’m feeling anxious” into an app. Within seconds, a friendly, articulate response appears: breathing exercises, reassurance, a gentle nudge to reframe your thinking. It costs £8 a month. No waiting list. No awkward introductions.

It sounds almost too good to be true. And for millions of people in the UK, it’s no longer science fiction: it’s what happened last night.

The rise of AI therapy apps has transformed how people access mental health support, but questions remain about effectiveness.

AI therapy apps like Wysa, Woebot, and Replika have exploded in popularity. The NHS is piloting AI-driven triage tools. ChatGPT users regularly report using the chatbot for emotional support, sometimes without even realising they’re doing therapy. In 2026, AI mental health tools are not a fringe curiosity; they’re a mainstream option.

But do they actually work? And, more pointedly, what do they not do that a human therapist does? This guide cuts through the hype to give you an honest, evidence-based comparison between AI therapy and working with a qualified human practitioner in the UK.

What Is AI Therapy, Exactly?

When people say “AI therapy,” they’re usually describing one of three quite different things:

1. Chatbot-Based Mental Health Apps

Apps like Wysa (NHS-approved), Woebot (developed at Stanford), and Youper use scripted or AI-generated conversations to deliver principles of cognitive behavioural therapy (CBT). They ask questions, suggest exercises, and track your mood. Wysa has over 5 million users worldwide and is available through some NHS trusts.

These tools are built on structured clinical frameworks. They’re not free-form chatbots. When you tell Wysa you’re stressed, it doesn’t improvise; it guides you through a CBT-based exercise that a clinician designed.

2. General-Purpose AI as Accidental Therapists

This is the wild card of 2026. People are using ChatGPT, Claude, and Google Gemini as unofficial therapists: describing their problems, asking for advice, and receiving empathetic, well-articulated responses. OpenAI has added content warnings, but the behaviour is widespread. A 2025 study found that 1 in 8 ChatGPT users had discussed mental health concerns with the platform.

These tools weren’t designed for therapy. They have no clinical safeguards. But they’re free, instant, and non-judgemental, and for someone sitting alone at 2 a.m., that’s a powerful combination.

3. AI-Assisted Professional Therapy

Some private therapists now use AI tools to summarise sessions, track client progress, or suggest interventions. This isn’t “AI replacing the therapist”; it’s AI as a clinical assistant, similar to how a surgeon might use robotic tools. The human relationship remains central; the AI handles the admin.

The British Association for Counselling and Psychotherapy (BACP) has published guidance on AI in therapeutic practice, emphasising that AI should “support, not replace, the therapeutic relationship.”

What the Research Says: Do AI Therapy Apps Work?

The evidence is nuanced, and sometimes contradictory.

The Case For AI Therapy

• A 2023 meta-analysis published in Nature Digital Medicine reviewed 19 randomised controlled trials of AI mental health chatbots and found a small but statistically significant reduction in depression symptoms (g=0.29), comparable to some self-help interventions.
• Wysa’s clinical trials (published in JMIR Mental Health) showed a 24% reduction in anxiety scores among users with chronic pain over 8 weeks, compared to a control group.
• Woebot demonstrated a significant decrease in depression symptoms among college students in a 2022 Stanford trial, with effects persisting at a 6-week follow-up.
• AI tools have a unique advantage: availability. They’re accessible 24/7, never tired, never judgemental, and cost £0-£10/month versus £50-£120/session for private therapy.

The Limitations and Gaps

• Most trials are short-term (4-8 weeks) with small sample sizes; long-term outcomes are unknown.
• Dropout rates are high. A 2024 review found that 70-80% of users abandon mental health apps within 2 weeks, far higher than therapy dropout rates (~20-30%).
• Trials screen out severe mental illness, suicidal ideation, and substance use: the people who most need help. AI apps are not tested for crisis situations.
• There’s a publication bias problem: most studies are funded by the app companies themselves.
• Cultural and linguistic limitations: Most AI therapy tools are developed in the US and trained on English-language data. They perform poorly with regional dialects, non-Western cultural frameworks, and nuanced emotional expression.

In short: AI therapy apps can help with mild-to-moderate symptoms, roughly equivalent to a good self-help book, but delivered conversationally. For anything beyond that threshold, the evidence doesn’t support replacing a human therapist.

What Human Therapists Do That AI Cannot

This is the core tension, and it’s worth being specific about it.

1. The Therapeutic Relationship

The single strongest predictor of therapy outcomes isn’t the modality, whether CBT, psychodynamic or person-centred; it’s the quality of the therapeutic alliance. Research consistently shows that the relationship between client and therapist accounts for 30-50% of therapeutic change, more than any specific technique.

An AI cannot build a relationship. It can simulate one. It can remember your name, mirror your language, express empathy convincingly. But it doesn’t care about you. It doesn’t think about you between sessions. It doesn’t feel a professional responsibility toward your wellbeing. This distinction is clinical, not merely philosophical: the therapeutic relationship works because it’s real.

2. Reading What’s Not Said

Human communication is overwhelmingly non-verbal. A skilled therapist notices when your voice wavers on a particular word, when you look away at a certain topic, when your body language contradicts your stated feelings. AI text interfaces capture none of this. Video-based AI could theoretically analyse facial expressions, but that introduces profound privacy concerns, and still wouldn’t match the intuitive attunement of an experienced human clinician.

3. Clinical Judgement and Risk Assessment

When a client says “I don’t think I can go on,” a human therapist hears the weight of that statement in context: the relationship history, the subtle cues, the pattern of previous disclosures. They know when to pause a session and conduct a formal risk assessment. They have a legal and ethical framework, a supervisor to consult, and emergency procedures.

AI chatbots have crisis detection algorithms that trigger pre-written responses and suicide hotline referrals. These are safety nets, not clinical judgement. In 2023, a Belgian man died by suicide after a series of conversations with a chatbot that failed to escalate his escalating distress. The AI didn’t miss a checklist item. It missed what it didn’t know it didn’t know.

4. Working With Complexity

Real human beings don’t arrive with a single tidy diagnosis. Someone seeking therapy might be dealing with childhood trauma, a difficult marriage, workplace bullying, chronic pain, and ADHD symptoms, all interacting in ways no algorithm can model. Human therapists integrate information across domains, adapt their approach mid-session, and hold multiple, sometimes contradictory, truths simultaneously.

5. Healing Through Relationship

Many therapeutic modalities, particularly psychodynamic therapy, attachment-based therapy, and person-centred counselling, are fundamentally relational. The mechanism of change is the relationship: experiencing a safe, boundaried, attuned connection with another human being, often for the first time. An AI can role-play that dynamic, but it can’t be it, because there’s no person on the other side.

Privacy: What Happens to Your Data?

This is where the conversation gets uncomfortable.

When you speak to a UK-registered therapist, your data is protected by GDPR, the Data Protection Act 2018, and professional confidentiality obligations enforced by bodies like the BACP and UKCP. Therapists keep notes that are stored securely, and you have the right to access them. Breaches can end a therapist’s career.

When you type your deepest fears into an AI chatbot, the picture is murkier. Most commercial AI therapy apps collect and store conversation data. Some anonymise it; many don’t. Privacy policies are often lengthy and permissive. Wysa’s policy notes that anonymised data may be used for “research and product improvement,” while ChatGPT conversations may be reviewed by human trainers unless you specifically opt out.

A 2024 Mozilla Foundation review gave privacy warning labels to 22 out of 27 mental health apps examined, citing vague data-sharing practices and weak encryption. The report’s title: “Mental Health Apps Have Terrible Privacy Protections.”

There have been no major UK data breach scandals involving AI therapy apps. Not yet. But the regulatory framework is playing catch-up, and the Information Commissioner’s Office (ICO) has signalled that mental health AI will be a priority area for enforcement.

The NHS Perspective: Where AI Fits

The NHS isn’t blind to AI’s potential, or its limits. Several NHS trusts have piloted AI tools, including:

• Wysa: commissioned by several NHS mental health trusts as a step 1 intervention (pre-therapy support while on waiting lists)
• Ieso Digital Health: an AI tool that analyses therapist language patterns in typed CBT sessions to improve outcomes
• Limbic: an AI clinical decision support tool used in NHS Talking Therapies (formerly IAPT) for triage

The key phrase is “while on waiting lists.” The NHS is using AI as a bridge, not a destination: something to offer people during the 18-week wait for in-person therapy, not a replacement for it. This pragmatic approach reflects the evidence: AI helps at the mild end, but clinical complexity demands human expertise.

AI Therapy vs Human Therapy: A Side-by-Side Comparison

Cost

AI: £0–£10/month. Often free for basic features.
Human: £50–£120/session private. Free through NHS but with waiting lists of 6-18+ weeks.

Availability

AI: 24/7, instant response, no appointment needed.
Human: Scheduled sessions, typically weekly, limited by therapist availability.

Evidence Base

AI: Small studies showing modest benefit for mild symptoms. No long-term data. Limited to structured CBT delivery.
Human: Decades of research across multiple modalities. Strong evidence for moderate-to-severe conditions, trauma, and complex presentations.

Relationship

AI: Simulated empathy. No genuine therapeutic alliance. No continuity of care if the service changes or shuts down.
Human: Real therapeutic relationship. The alliance itself is a mechanism of change. Continuity of care over months or years.

Suitability

AI: Mild anxiety, low mood, sleep issues, stress management, psychoeducation.
Human: All severity levels, including moderate-to-severe depression, anxiety disorders, PTSD, complex trauma, personality disorders, eating disorders, suicidal ideation.

Risk Management

AI: Crisis keywords trigger pre-scripted responses and hotline referrals. No clinical judgement. Several documented failures in real-world settings.
Human: Formal risk assessment, safety planning, clinical supervision, legal duty of care, emergency protocols, liaison with GPs and crisis teams.

Privacy

AI: Variable. Privacy policies often permit data use for product improvement. Some apps share data with third parties. Regulatory framework still emerging.
Human: GDPR-protected. Professional confidentiality obligations. Secure clinical notes. Breaches carry professional and legal consequences.

Personalisation

AI: Adapts to your language and stated preferences within pre-programmed boundaries. Cannot integrate complex life context.
Human: Adapts in real time across all domains: emotional, cognitive, relational, cultural, spiritual. Tailors approach based on deep clinical understanding.

Online therapy with a qualified human practitioner combines the convenience of digital access with the depth of a real therapeutic relationship.

When AI Therapy Makes Sense

There are genuine use cases where AI mental health tools are appropriate and helpful:

• While on an NHS waiting list. An AI app may help you manage symptoms during the 6-18 weeks you’re waiting for NHS Talking Therapies. It’s not a substitute, but it’s better than nothing.
• For psychoeducation. If you want to understand CBT principles before starting therapy, an app like Wysa can teach you the basics.
• For mild, situational stress. If you’re going through a difficult week at work and need coping strategies, an AI tool may be sufficient.
• As a supplement to therapy. Some therapists encourage clients to use apps for mood tracking or between-session exercises. Ask your therapist for recommendations.
• If cost is an absolute barrier. A free AI tool is better than no support at all, provided you understand its limitations.

When You Need a Human Therapist

You should seek a qualified human therapist if:

• Your symptoms are moderate-to-severe, interfering with work, relationships, or daily functioning
• You have thoughts of self-harm or suicide: this is an emergency, not a chatbot scenario
• You’ve experienced trauma, abuse, or adverse childhood experiences
• You have a diagnosed mental health condition (PTSD, OCD, bipolar, personality disorder, eating disorder)
• You’ve tried self-help or AI tools and haven’t improved
• You want to understand yourself deeply, not just manage symptoms
• You need a formal diagnosis or medical documentation, which AI cannot provide
• You’re taking or considering psychiatric medication, which requires professional oversight

Finding the Right Human Therapist: Where to Start

If you’ve decided that working with a real therapist is the right path, the next step is finding someone who fits. Here’s what matters:

• Qualifications: Look for accreditation by a UK professional body: the BACP (British Association for Counselling and Psychotherapy), UKCP (UK Council for Psychotherapy), BPS (British Psychological Society), or HCPC (Health and Care Professions Council) for clinical psychologists. These bodies set ethical standards and require ongoing professional development.
• Modality: Different problems respond to different approaches. CBT is effective for anxiety disorders; EMDR for trauma; psychodynamic therapy for deep-rooted relational patterns. A good therapist will discuss this with you in an initial consultation.
• Fit: Research consistently shows that the therapeutic relationship matters more than the specific technique. Many therapists offer a free or reduced-cost initial session. Use it to gauge how comfortable and understood you feel.
• Practicalities: Consider location, online vs in-person, session frequency, cost, and availability. A brilliant therapist you can never schedule with isn’t helping anyone.

Seek A Psych offers a free, personal matching service. Tell us what you’re looking for and we’ll connect you with a qualified therapist: no public directory, no scrolling through profiles. Every therapist listed is registered with a recognised UK professional body.

Related Articles

• Does CBT work for anxiety?
• UCL Case Formulation Model, Mental Health and CBT
• Ultimate Guide to Finding Therapy in London: 2026