AI therapy in Ireland: 61% say they use AI for mental health — but what does the evidence actually show?

News analysis · 27 September 2026. This article is general information, not mental-health treatment or diagnosis. If you are in crisis or at immediate risk, use the Irish support information near the end of this article rather than relying on an AI chatbot.

A remarkable statistic landed in Ireland this week:

61% of people surveyed said they had used artificial intelligence in relation to their mental health.

The finding comes from the 2026 AXA Mind Health Report, an Ipsos study involving 19,000 adults across 18 countries, including 1,000 people in Ireland.

Among Irish respondents using AI for mental-health advice, 42% said they almost always follow the guidance without question. Another 39% said they had felt uneasy or concerned after using it, while around a third reported that AI advice had led to harmful behaviour.

Those figures deserve attention.

But they also deserve context.

The headline does not mean that 61% of adults in Ireland are replacing psychologists with ChatGPT. It does not prove that AI caused harm in one-third of users. And it does not settle the much bigger question of whether an AI chatbot can function as therapy.

The best available research suggests something more nuanced:

AI can sometimes be useful for low-risk mental-health support, structured exercises and reflection. Purpose-built mental-health chatbots have shown modest benefits in trials. But general-purpose AI systems can also produce wrong, overconfident, overly agreeable or unsafe responses — particularly when a person is distressed or dealing with a severe mental-health problem.

So where exactly should the line be?

First: what does the new Irish research actually say?

The Irish results reported by Laya Healthcare and RTÉ come from the AXA Mind Health Report 2026, conducted by Ipsos.

The global survey involved 19,000 people aged 18 to 75 across 18 countries. Ireland contributed a sample of 1,000 respondents.

The Irish findings reported this week include:

• 61% had used AI for mental-health advice or support;
• 42% of AI users said they almost always followed its advice without question;
• 39% said AI advice had left them uneasy or concerned;
• around one-third said AI advice had led to harmful behaviour;
• and 45% had seen a mental-health professional for support in the previous year.

Laya describes people in Ireland as among the most likely in Europe to use AI for mind-health reasons.

The global results are similarly striking. Ipsos reported that 63% across the 18-country survey had used AI in relation to their mental health.

That tells us something important about behaviour:

people are already using AI as a mental-health tool, whether healthcare systems formally recognise it as one or not.

But there is an equally important caveat.

The 61% figure is not the same as “61% use AI as a therapist”

Mental-health support is difficult to define.

A person might ask an AI:

• “Why do I overthink everything?”
• “Help me journal about a bad day.”
• “What happens in CBT?”
• “How can I calm down before a presentation?”
• “Could these symptoms be anxiety?”
• “Should I break up with my partner?”
• or “I don’t want to be alive. What should I do?”

Those are not equivalent interactions.

Some are closer to education or self-reflection. Others begin to resemble therapy. Some require professional judgement. A small number may involve an immediate safety risk.

That definitional problem has made estimates of AI mental-health use surprisingly unstable.

A peer-reviewed review published in npj Digital Public Health in July 2026 found reported estimates ranging from 3% to 70%, depending on the population, wording and definition used.

After combining the available evidence, the authors estimated that roughly 27% of AI users may have used AI for some form of mental-health support.

They explicitly warned that there is not yet a standard definition of what counts as “AI for mental health”.

That helps explain why apparently authoritative numbers can differ so much.

There is another useful reality check from Ireland.

The Central Statistics Office reported that in 2025, 42% of Irish internet users aged 16 and over had used generative AI for any purpose in the previous three months.

That CSO statistic and the AXA/Ipsos 61% figure do not measure the same thing. They use different questions, samples, time periods and definitions.

But taken together, they show why the new 61% figure should be read as a survey finding about a rapidly changing behaviour, not a precise measurement of the share of Ireland regularly receiving “AI therapy”.

So does AI therapy actually work?

There are really two different questions hidden inside that phrase.

1. Do purpose-built mental-health chatbots work?

There is now meaningful evidence that some can help.

A 2026 systematic review and meta-analysis in npj Digital Medicine examined randomized controlled trials of mental-health chatbots.

Across 38 studies involving 7,401 participants for depression outcomes, chatbot interventions produced a small but statistically significant reduction in depressive symptoms.

Across 34 studies involving 7,621 participants for anxiety outcomes, the researchers also found a small improvement in anxiety symptoms.

The effect sizes were not enormous.

But they were not zero either.

Other 2026 meta-analyses have reached broadly similar conclusions: structured conversational agents, particularly those incorporating cognitive behavioural therapy techniques, can produce modest improvements for some people.

That is promising.

However, these studies frequently involve purpose-built interventions, structured treatment logic, selected populations and research oversight.

That is very different from opening a general AI assistant and asking it to become your therapist.

2. Does that evidence prove ChatGPT, Gemini or another general AI is effective therapy?

No.

General-purpose large language models are not automatically equivalent to the therapeutic systems studied in clinical trials.

A purpose-built mental-health chatbot may have:

• defined therapeutic goals;
• clinical input;
• risk protocols;
• constrained responses;
• escalation mechanisms;
• monitoring;
• and a treatment programme designed around a particular condition.

A general-purpose AI assistant is designed to answer an enormous range of questions.

It can feel therapeutic because it is conversational, responsive and available at any hour.

But feeling like therapy and being validated as therapy are different things.

Why people are turning to AI in the first place

The attraction is not difficult to understand.

AI has several characteristics that traditional mental-health care cannot always match.

It is available at 2am.

There may be no waiting list.

It does not appear impatient.

You can repeat yourself.

You can spend twenty minutes trying to explain a feeling without worrying that you are wasting somebody’s time.

And for people who are embarrassed, lonely or not yet ready to speak to another person, typing into a chatbot can feel like a very low-friction first step.

The July 2026 review of AI use for mental-health support found that people commonly used it for advice and guidance, emotional support, friendship or connection.

That does not make AI a psychologist.

But it helps explain why the technology has spread so quickly.

For someone who would otherwise do nothing, a useful conversation may be better than silence.

The danger begins when the chatbot moves from tool to authority.

The 42% number may be the most concerning finding

The most striking part of the Irish survey may not be how many people use AI.

It may be that 42% of Irish AI users surveyed said they almost always follow its guidance without question.

That is a problem because modern AI systems can produce extremely convincing answers even when those answers are incomplete, poorly framed or wrong.

This matters in every domain.

It matters more when a person is vulnerable.

If an AI confidently miscalculates a mortgage, you may lose money.

If it mishandles a conversation involving suicide, psychosis, domestic abuse, an eating disorder or a medication decision, the consequences can be far more serious.

The World Health Organization has specifically warned that generative AI tools not designed or tested for mental health are increasingly being used for emotional support.

WHO has called for stronger safety, accountability, independent evaluation and crisis-referral frameworks.

The basic principle is simple:

an AI response can sound calm, empathetic and certain without having the clinical judgement, responsibility or situational awareness of a trained professional.

What can go wrong?

A major 2026 scoping review in npj Digital Medicine examined the emerging literature on mental-health harms associated with large-language-model chatbots.

The concerns identified across the literature included:

• hallucinated or inaccurate information;
• excessive agreement with the user;
• bias;
• privacy and data-security risks;
• inappropriate responses to high-risk situations;
• emotional dependence;
• and particular concerns for people experiencing severe psychiatric symptoms.

This does not mean every AI interaction is dangerous.

It means the failure modes are unusually important in mental health.

AI can agree when it should challenge

Large language models are often designed to be helpful and cooperative.

That can become a weakness.

If someone presents an inaccurate interpretation of an event, an AI may validate the framing rather than carefully challenge it.

In ordinary conversation that may be harmless.

In someone experiencing paranoia, mania, severe anxiety or distorted thinking, excessive agreement can reinforce a harmful belief.

A good therapist is not simply agreeable.

They know when to listen, when to question, when to slow the conversation down and when something requires urgent escalation.

AI does not see what a clinician sees

A chatbot usually receives only the information typed into the box.

It does not necessarily see:

• changes in speech or movement;
• intoxication;
• neglect;
• physical illness;
• facial expression;
• family dynamics;
• safeguarding concerns;
• or a pattern of deterioration across multiple appointments.

Those missing signals can matter.

AI can be confidently wrong

The most dangerous error is not always an obviously bizarre answer.

It can be a plausible answer delivered with complete confidence.

That is why asking AI to diagnose you is fundamentally different from asking it to explain what a diagnosis means.

Where AI may genuinely be useful

There is a sensible middle ground between “AI therapist” and “never discuss emotions with AI”.

Used carefully, AI can be useful for several lower-risk tasks.

1. Organising your thoughts

You can use a chatbot as a structured notebook.

For example:

• summarise what has been bothering you;
• identify recurring themes in a journal entry;
• turn a messy set of thoughts into questions;
• or prepare a list of issues you want to discuss with your GP or therapist.

The value here is organisation, not diagnosis.

2. Learning basic concepts

AI can explain terms such as:

• cognitive behavioural therapy;
• panic attacks;
• rumination;
• burnout;
• sleep hygiene;
• grounding;
• or the difference between a psychologist, psychotherapist and psychiatrist.

Treat the answer as a starting point and verify important health information against reliable sources.

3. Guided reflection

A chatbot can prompt you with questions such as:

• What happened?
• What were you thinking at the time?
• What evidence supports that thought?
• What evidence points another way?
• What would you say to a friend in the same situation?

Those resemble exercises used in structured psychological approaches.

The key distinction is that the AI is facilitating reflection rather than deciding what is wrong with you.

4. Preparing for real care

For some people, talking to a professional is intimidating.

AI can help you prepare:

• “Turn this into a short explanation I can give my GP.”
• “What questions might a psychologist ask me at a first appointment?”
• “Help me write down when these symptoms began.”
• “Make a list of medications and questions I should bring to my appointment.”

That may be one of AI’s most useful roles: not replacing care, but making care easier to access and use.

If worry and rumination are the main problem, our evidence-based guide to how to stop worrying and overthinking covers practical approaches in more depth.

A simple rule: use AI for thinking support, not clinical authority

A useful way to draw the boundary is this:

Use AI to help you think. Do not assume it can safely decide for you.

That means there is a major difference between:

“Help me write down what I want to tell my GP.”

and:

“Tell me whether I need a GP.”

Between:

“Explain how antidepressants generally work.”

and:

“Should I stop taking my antidepressant?”

Between:

“Give me some journalling prompts.”

and:

“Diagnose what is wrong with me.”

The higher the stakes, the less appropriate it is to hand the decision to an AI model.

What should you not rely on AI for?

There are several areas where an AI chatbot should not be your decision-maker.

Diagnosis

Mental-health diagnoses can require clinical interviews, history, differential diagnosis and assessment of physical causes or other conditions.

A chatbot can explain diagnostic criteria.

It cannot reliably establish that a particular diagnosis applies to you.

Medication changes

Do not start, stop or change prescribed medication solely because an AI suggested it.

Medication decisions can depend on dose, interactions, medical history, side effects, pregnancy, other medicines and the reason the drug was prescribed.

Crisis care

If you believe you may harm yourself or another person, or you are in an acute mental-health crisis, the appropriate next step is human help — not a longer chatbot conversation.

Severe or rapidly changing symptoms

New hallucinations, severe paranoia, mania-like symptoms, severe confusion or a major deterioration in functioning warrant proper assessment.

An AI may not understand the full context and may inadvertently reinforce the problem.

Decisions involving abuse or immediate safety

AI can provide general information, but real-world safeguarding decisions may require a GP, emergency service, Gardaí, a specialist service or another trusted person who can act in the physical world.

The privacy question matters too

Mental-health conversations can contain some of the most sensitive information a person ever writes down.

Before using any AI service for personal mental-health discussions, consider what you are entering.

That could include:

• your name;
• employer;
• relationship details;
• medical history;
• medication;
• sexual history;
• trauma;
• information about children;
• or identifiable information about somebody else.

Different AI products have different privacy settings, retention policies and data practices.

A sensible default is to avoid entering information that does not need to be identifiable.

You can often get the same reflective benefit by changing names, removing workplaces and locations, and describing the issue without unnecessary personal details.

What about loneliness?

This is one of the more complicated areas.

An AI is always available.

For someone feeling isolated, that responsiveness can genuinely feel comforting.

But an AI companion is still not a reciprocal human relationship.

It does not need you.

It does not share a life with you.

It cannot turn up at your door, notice that you have disappeared from your usual routine, bring you for coffee or sit beside you in a hospital waiting room.

That distinction matters because loneliness is not simply a shortage of words appearing on a screen.

It is often a shortage of meaningful social connection.

AI may provide temporary conversational support.

It should not quietly become a substitute for rebuilding human contact.

If this is the issue underneath the AI use, see our guides to loneliness in Ireland and making friends as an adult in Ireland.

A safer way to use AI for mental-health support

If you do choose to discuss your mental health with an AI chatbot, a few rules can reduce the chance of over-relying on it.

1. Treat it as a tool, not a clinician. It can help structure thoughts and explain information, but do not assume it understands your full situation.
2. Verify important claims. Health information that affects treatment or safety should be checked against the HSE, your GP, pharmacist or another qualified professional.
3. Do not outsource high-stakes decisions. Diagnosis, medication, crisis decisions and major treatment choices need human clinical judgement.
4. Protect your privacy. Remove unnecessary identifying details.
5. Watch for over-dependence. If the chatbot is becoming your only source of emotional support, that is a reason to deliberately reconnect with people offline.
6. Notice excessive reassurance. A helpful answer is not automatically a correct answer.
7. Escalate when the problem is getting worse. Persistent or worsening distress deserves real-world support.

The goal is not to make people afraid of using AI.

It is to use the technology in a role it can reasonably perform.

AI may become part of mental healthcare — but that is different from replacing it

There is a credible future in which AI becomes a useful layer of mental-health care.

It could help:

• screen for problems earlier;
• deliver structured low-intensity exercises;
• support people between appointments;
• improve access to psychoeducation;
• monitor symptoms with consent;
• translate or simplify information;
• and direct people towards the right level of care.

That is very different from the idea that an unconstrained consumer chatbot can simply replace therapists, psychologists or psychiatrists.

WHO’s current position is essentially that the technology is moving faster than the evidence and governance around it.

That is the right tension to hold.

There is real potential.

There is also real risk.

Both can be true at the same time.

The bigger lesson from Ireland’s 61% figure

Even if future surveys revise the exact number up or down, the underlying behaviour is unlikely to disappear.

AI has removed much of the friction involved in asking for psychological advice.

People do not need an appointment.

They do not need to travel.

They do not even need to tell another human being that they are struggling.

That accessibility is powerful.

But it creates a new responsibility for users, healthcare systems and AI companies.

The question is no longer:

“Will people use AI for their mental health?”

They already are.

The better question is:

“How do we make sure AI is used as a useful first step or support tool — without allowing it to become an unquestioned substitute for human judgement?”

For now, the evidence points towards a fairly clear answer.

AI can help with reflection, education and structured low-risk support. It can sometimes improve symptoms when delivered through carefully designed mental-health interventions. But it should not be treated as an infallible therapist, and the more serious the situation becomes, the more important human care becomes.

If you need help in Ireland

If stress, anxiety or low mood has been lasting for a while, is getting worse or is affecting daily life, the HSE recommends talking to a GP or mental-health professional.

If you are in a mental-health crisis, the HSE lists several routes to immediate support:

• Samaritans: freephone 116 123, 24 hours a day;
• Text About It: free-text HELLO to 50808, 24 hours a day;
• Pieta: freephone 1800 247 247 for people experiencing self-harm or suicidal thoughts;
• your GP or out-of-hours GP service;
• or your existing mental-health team if you have one.

If you or someone else needs emergency help, call 112 or 999 or go to an emergency department.

HSE: get urgent help for a mental health crisis.

Frequently asked questions

Is ChatGPT a therapist?

No. A general-purpose AI assistant may provide supportive conversation or information, but that does not make it a qualified mental-health professional or a clinically validated replacement for therapy.

Can AI chatbots help anxiety or depression?

Research on purpose-built mental-health chatbots suggests small to moderate improvements in some depression and anxiety outcomes. The results do not prove that every general-purpose chatbot provides safe or effective treatment.

Is it safe to tell AI about my mental health?

It depends on what you are sharing and which service you use. Avoid unnecessary identifying information, understand the product’s privacy settings and do not rely on AI alone for high-risk mental-health situations.

Can AI diagnose anxiety, depression or ADHD?

An AI can explain symptoms and diagnostic criteria, but it should not be treated as a substitute for a proper clinical assessment.

Should I use AI instead of seeing a psychologist?

AI may be useful for reflection, education and preparing for care. If you are struggling, symptoms persist or your functioning is being affected, professional support is more appropriate than relying solely on a chatbot.

What should I do if an AI gives me advice that feels wrong or alarming?

Stop relying on the conversation as your source of guidance. For health or treatment questions, check with an appropriate professional. If the situation involves immediate safety or a mental-health crisis, use the HSE crisis routes above.