AI and Mental Health: A Useful Tool, but Not a Replacement for Human Counselling

Artificial intelligence has changed enormously in a very short period of time.

When I first wrote about AI and therapy, I was much more cautious about its role in mental health. I still do not believe AI should replace a human counsellor, but I do think the conversation has become more nuanced.

AI can be useful. For some people it can provide information, help organise thoughts, encourage reflection, explain psychological concepts or offer practical strategies at a time when a counsellor is not immediately available.

The question is no longer simply whether AI has a place in mental health support. I think it clearly does. The more important question is where is it helpful, and where do we still need another human being?

What AI Can Do Well

One of the obvious strengths of AI is accessibility.

You can ask a question at 2 am, explore something privately, or get help putting words around what you are feeling without needing to wait for an appointment.

AI may be useful for things such as:

• learning about anxiety, boundaries, communication or coping strategies

• journalling and reflecting on thoughts or feelings

• preparing questions or thoughts before a counselling session

• practising skills learned in counselling

• identifying patterns you may want to explore further

• helping you put an experience into words

• suggesting practical ways to manage stress or difficult emotions

There is also growing research into AI tools that have been specifically designed for mental health treatment rather than general-purpose chatbots.

For example, a 2025 randomised trial of a purpose-built generative AI mental-health chatbot called Therabot found significant improvements in symptoms of depression and anxiety compared with a control group. Participants also reported feeling a surprisingly strong sense of connection with the program. More recent reviews suggest that some AI conversational tools, particularly those based on established approaches such as CBT, may provide real benefits for some people. That is important evidence, and I do not think mental-health professionals should simply dismiss it because AI is not human.

But Not All AI Mental Health Support Is the Same

There is an important difference between a mental-health tool that has been deliberately developed, clinically informed and tested, and asking a general AI chatbot to become your therapist. Most general AI systems were not designed to provide mental-health treatment.

They are designed to generate helpful-sounding responses based on patterns in enormous amounts of information. That means they can give very good answers, but they can also misunderstand context, give inaccurate information or agree with something that really needs to be questioned.

Research has already demonstrated some concerning examples. In testing, AI chatbots have sometimes failed to respond appropriately to suicidal thinking or have reinforced delusional beliefs instead of challenging them safely. Researchers have also raised concerns about bias, over-validation and the risk of people becoming emotionally dependent on AI systems.

This does not mean AI is dangerous every time someone uses it for emotional support. It does mean we need to be thoughtful about what we ask it to do.

Sometimes We Need More Than Reassurance

One limitation I see with AI is that it can be very good at validating us. Validation is important. Most of us need to feel heard and understood. But good counselling is not simply someone agreeing with everything we think or feel. Sometimes we need someone who can gently notice a pattern we cannot see ourselves, question an assumption, recognise when we are avoiding something difficult, or help us look at a situation from another person's perspective.

There are times when the most helpful response is not the most comfortable one.

A counsellor is also taking responsibility for the work they are doing with you. We have professional boundaries, ethical responsibilities, training, supervision and a duty to recognise when something is outside our scope or when additional support is needed.

AI cannot hold that responsibility in the same way.

The Relationship Still Matters

Counselling is not only about information. If information alone solved most emotional and relationship difficulties, many of us would probably be doing much better already.

We can understand intellectually that we should set boundaries, stop overthinking, communicate differently or be kinder to ourselves and still find it incredibly difficult to actually do those things.

Part of counselling is being in a relationship where another person is paying attention to you over time. A counsellor remembers your story, notices patterns, hears what you say and sometimes what you struggle to say. We can notice changes in tone, expression, hesitation or emotion that may tell us something important is happening. We can also sit with grief, shame, uncertainty or pain without needing to immediately solve it. That human relationship is still a significant part of what makes counselling different from receiving good advice.

Where I Think AI Fits Best

For me, this does not need to become an argument about AI versus counselling.

I think AI is likely to become another useful tool. It may work particularly well alongside counselling, much like books, worksheets, journalling, apps and other resources already do.

You might use AI between sessions to help reflect on something we discussed, generate questions you want to bring back, practise a communication skill or better understand a concept. It may also provide a useful first step for someone who is not ready or able to access counselling yet.

What I would be cautious about is relying on it as your only source of support when you are experiencing significant mental-health difficulties, suicidal thoughts, psychosis, trauma-related concerns, abuse, severe relationship problems or other complex situations.

Those are times when human professional support becomes particularly important.

The World Health Organization has taken a similarly balanced position. It recognises the potential of AI to improve access to health information and care, while emphasising the need for evidence, safety, accountability and proper oversight when AI is used in mental health.

My View

I do not think AI is going away, and I do not think it should. Used thoughtfully, it can be an incredibly useful resource. But I also do not think we should confuse a system that can produce empathic language with a person who genuinely knows us, takes responsibility for our care and is present with us in a real relationship.

For many people, the best future may not be choosing between technology and human support at all.It may be learning how to use both well.

References

Heinz, M. V., et al. (2025). Randomized Trial of a Generative AI Chatbot for Mental Health Treatment. NEJM AI.

Hang, Y., et al. (2026). The effectiveness of CBT-based NLP-enabled AI conversational agents for mental health intervention: a systematic review and meta-analysis. npj Digital Medicine.

Moore, J., et al. (2025). Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers. ACM FAccT 2025.

World Health Organization. (2026). Towards responsible AI for mental health and well-being: experts chart a way forward.