AI health advice has moved from a private experiment into everyday life. An Irish survey now suggests that six in ten people use AI for mental health support. That does not prove AI improves mental health. Instead, it shows how quickly people have accepted a new source of guidance.
The attraction is understandable. A chatbot answers immediately, does not appear judgemental, and can explain difficult language in simple terms. However, health decisions depend on context, examination, professional judgement, and an accurate account of symptoms. A fluent answer cannot supply what the system has never seen.
What the Irish survey found
The RTE report covers Irish findings from the AXA Mind Health Report 2026. Ipsos surveyed 19,000 adults across 18 countries, including 1,000 people in Ireland.
According to the report, 61 percent of respondents in Ireland had used AI platforms for mental health support. Among Irish users, 42 percent said they almost always followed the guidance without question. Meanwhile, 39 percent felt uneasy or concerned after using it. One third reported that AI advice had led to harmful behaviour.
Those findings deserve attention, although they also need careful interpretation. This was a self reported survey, not a clinical trial. The published news report does not explain every behaviour described as harmful. It also cannot prove that AI caused an outcome.
Even with those limits, one result is especially important. More Irish respondents had used AI for mental health than had consulted a healthcare professional. Therefore, AI is already functioning as an informal layer of health support, whether health systems planned for it or not.
The popularity of AI health advice may also reflect cost, waiting times, privacy concerns, or fear of judgement. Ease of access explains use, but it does not establish safety or effectiveness.
The global AXA Mind Health Report 2026 was produced with Ipsos. The Ipsos summary reports that 63 percent of respondents globally had used AI in relation to mental health.
Cyberchondria existed before AI
Before chatbots, people searched symptoms online and tried to diagnose themselves. That behaviour became known as cyberchondria. It describes repeated health searching that increases anxiety or distress instead of resolving uncertainty.
A person might search for a common headache, encounter information about a rare illness, and become frightened. They may then search again for reassurance. Each new result can introduce another serious possibility, creating a cycle of fear, checking, and short lived relief.
AI did not create that cycle. However, it can make the experience more personal and persuasive. A search engine presents links. A chatbot responds directly, remembers the conversation, and adjusts its wording to the user. Consequently, its answer may feel like a personal assessment even when it is only a prediction based on limited text.
AI health advice can also provide unlimited reassurance. For someone who struggles with uncertainty, another answer may calm anxiety briefly before the urge to check returns.
Recent research shows why this matters for young people. A 2025 adolescent study in the Journal of Medical Internet Research surveyed 159 people aged 14 to 17. Eighty three percent had searched for health information online. Higher cyberchondria scores were strongly associated with health anxiety, problematic internet use, compulsive behaviour, and difficulty tolerating uncertainty.
The study was small and cross sectional, so it cannot establish cause and effect. Nevertheless, it confirms that cyberchondria is relevant during adolescence. AI may now add conversation, reassurance, and apparent empathy to an already familiar pattern.
What the latest research says about AI health advice
One of the most important studies appeared in Nature Medicine in February 2026. Researchers tested whether large language models could help the public identify conditions and choose an appropriate action across ten medical scenarios.
The randomised study of 1,298 participants produced a striking contrast. When tested alone, the models identified the condition correctly in 94.9 percent of cases. However, they chose the correct next action in only 56.3 percent.
Performance fell further when real people used the same systems. Participants identified the relevant condition in fewer than 34.5 percent of cases. They chose the correct action in fewer than 44.2 percent. Their results were no better than those using other information sources.
This does not mean AI was wrong in every conversation. Instead, it shows that medical knowledge tests do not recreate real use. People may omit key facts, describe symptoms unclearly, misunderstand the response, or focus on the most frightening possibility.
Another 2026 study examined 888 responses to 222 common medical questions. Sixteen physicians reviewed answers from four public chatbots. The study in npj Digital Medicine found problematic responses in 21.6 to 43.2 percent of answers, depending on the model. Reviewers classified 5 to 13 percent as unsafe.
Models change quickly, so those exact rates should not become permanent labels. Still, the study identified a wider problem. General chatbots can produce unsafe advice with the same confident tone used for accurate information.
How physical health advice can go wrong
A doctor does more than match symptoms to a condition. They examine the patient, assess vital signs, review medicines, consider medical history, and order tests where necessary. A general chatbot cannot perform those tasks.
AI health advice can therefore appear more complete than the information behind it. Missing details can change both the likely explanation and the safest next action.
Misdiagnosis can therefore happen in two directions. The system may overstate a rare illness and increase fear. Alternatively, it may offer false reassurance and delay urgent care. Both outcomes matter.
Other risks include unsuitable medicine advice, missed interactions, incorrect doses, and outdated guidance. A person might also stop prescribed treatment after receiving a convincing explanation online. No general AI answer should override a doctor, pharmacist, or other qualified professional.
AI can also struggle when symptoms look similar. Chest discomfort might reflect anxiety, indigestion, or a serious medical emergency. Text alone may not reveal the difference. Anyone facing severe, sudden, or rapidly worsening symptoms needs professional assessment, not a longer chatbot conversation.
How mental health advice can go wrong
Mental health conversations create different risks. A supportive tone can help someone feel heard, yet warmth is not the same as clinical understanding. The system does not experience empathy, accept professional responsibility, or observe changes in behaviour.
AI health advice becomes especially sensitive when a user feels isolated, frightened, or detached from reality. In those moments, agreement may feel comforting while quietly reinforcing risk.
People often need careful challenge as well as reassurance. A chatbot designed to remain agreeable may validate an inaccurate belief or reinforce a harmful interpretation. It may also miss indirect signs of psychosis, suicidal thinking, abuse, an eating disorder, or severe distress.
A 2025 Stanford study of therapy chatbots found increased stigma towards some conditions. Researchers also observed inappropriate responses to prompts involving suicidal thinking and delusions. However, the researchers did not argue that every use was harmful. They identified lower risk roles in journalling, reflection, coaching, and professional administration.
In March 2026, the World Health Organization warned that many generative systems used for emotional support were not designed or tested for mental health. Its experts called for independent testing, clinical input, crisis referral standards, and attention to emotional dependence.
Dependence deserves particular attention. An always available chatbot may become the first and only place someone discusses distress. Over time, that pattern could reduce contact with family, friends, teachers, or professionals. For a vulnerable person, constant reassurance may also strengthen compulsive checking rather than build confidence.
Where AI can genuinely help
AI health advice is not automatically useless or dangerous. The value depends on the task, the person, and the safeguards around the conversation. Used carefully, AI may help someone:
- Explain an unfamiliar medical term in plain language.
- Prepare questions for a doctor, therapist, or pharmacist.
- Organise a symptom timeline before an appointment.
- Summarise general information from a trusted health source.
- Support private journalling or structured reflection.
- Identify the type of professional service they may need.
These uses keep a human professional in the decision. They also treat AI as an information aid rather than a diagnostic authority.
A safer way to use AI for health questions
People do not need to avoid every health related AI tool. However, they should apply clear limits.
The safest role for AI health advice is to support preparation and understanding. Diagnosis, medication, crisis assessment, and treatment decisions still require qualified human care.
- Do not ask a general chatbot to diagnose you. Use it to prepare questions, not to reach a final conclusion.
- Never use AI for an emergency. Contact emergency services or an appropriate health service immediately.
- Do not start, stop, or change medication because of a chatbot response. Speak with a doctor or pharmacist.
- Ask the system to state uncertainty and provide reliable sources. Then open those sources and check the information.
- Use trusted Irish health information. The HSE mental health portal provides official information and routes to support.
- Protect private information. Do not upload medical records, prescriptions, names, addresses, or identifying details to a general chatbot.
- Watch for reassurance loops. Repeatedly asking the same question may be increasing anxiety rather than providing clarity.
- Bring useful output to a professional. A symptom list or question sheet can support a consultation, but it should not replace one.
What parents and schools should discuss
Young people may use chatbots because they feel embarrassed, frightened, or unwilling to burden an adult. A calm response will make future disclosure more likely. Punishment may push the behaviour further into private use.
Parents should explain that an AI can sound caring without understanding the child. Important health information must be checked with a trusted adult and a qualified professional. Families can also use CODA guidance on AI risks for children and children’s online privacy.
Schools should include AI health questions within digital literacy and safeguarding policies. Staff need a clear process when a student reports harmful advice, severe distress, or dependence on a chatbot. CODA’s online safety programmes for schools connect AI use with cyberpsychology, privacy, wellbeing, and practical safeguarding.
When to seek urgent help in Ireland
If someone may harm themselves or another person, phone 112 or 999. They can also attend the nearest emergency department. The HSE urgent mental health guidance provides current support information.
Text About It offers free support at any time. Text HELLO to 50808. Samaritans can be contacted on 116 123. Childline is available to children and young people on 1800 66 66 66 or through online chat.
For an urgent physical health problem, contact a GP or GP service outside normal hours. Call 112 or 999 when urgent medical attention is needed or a life is at risk. The HSE urgent care guide explains the available options.
The CODA View
The central issue is not whether AI can produce impressive medical answers. It can. The real question is whether a person can safely use that answer while worried, unwell, missing information, or seeking reassurance.
Earlier online searching created cyberchondria by placing frightening possibilities beside common symptoms. Conversational AI can intensify that effect because the response feels personal, immediate, and confident. It may also offer repeated reassurance without recognising the cycle it is feeding.
However, alarm is not the answer. AI can improve access to explanations, help people prepare for appointments, and support reflection. These benefits become safer when the system directs users towards trusted sources and qualified human care.
Health related AI needs stronger clinical testing, clear limits, crisis escalation, privacy protection, and independent oversight. Children and vulnerable users require particular safeguards. Most importantly, the public must understand that fluency is not diagnosis and confidence is not proof.
For that reason, AI health advice must remain connected to accountable human care. Convenience should open a route to support, not close it.
AI health advice should support human care. It should never quietly become its replacement.
About Children of the Digital Age
Children of the Digital Age provides practical online safety education for students, parents, schools, and organisations. Our work combines frontline experience, computer forensics, cybersecurity, and cyberpsychology. We explain both the technology and the human behaviour behind digital harm.
To discuss an AI safety or online safety session, visit Children of the Digital Age.
Last updated: 26 September 2026
This article provides general information. It does not provide medical diagnosis or individual treatment advice.