For some people, particularly youngsters, an AI chatbot is becoming the first place they turn to when they are distressed.
It has its advantages: it is available at any hour, does not appear to judge them and can respond instantly.
But when conversations involve thoughts of suicide, the limits of this interaction become consequential.
Every conversation with an AI chatbot may not be harmful, but an apparently empathetic response could reinforce hopelessness, prolong rumination, or delay professional help in some people already at risk, say experts.
AI therapy apps keep experts and regulators on their toes What the evidence shows AI use for mental health support is now common.
A U.S. survey of 1,009 people aged 12 to 21, published in JAMA Pediatrics found that 19.2% had used AI chatbots for mental health advice in 2025.
Nearly 43% of these users did so at least monthly, while 5.8% used them daily or almost daily.
Nearly two-thirds had not told anyone about this.
Another analysis of more than 620,000 publicly available ChatGPT conversations found that 4.9% could be classified broadly as involving mental health, although only 0.21% met a stricter definition involving an identifiable person explicitly seeking mental health help.
Studies have also found that chatbots are generally better at recognising very low- and very high-risk scenarios, but struggle to distinguish between intermediate levels of suicide risk.
This is particularly concerning for psychiatrists, as suicidal distress is rarely expressed in a single, explicit statement and often emerges through a combination of subtle or changing cues.
On regulating AI chatbots’ interactions with children | Explained When reassurance becomes a loop “Suicidal distress is not always direct; people may hint, use humour or gradually disclose their thoughts,” said Mithun Prasad, consultant, psychiatry, SIMS Hospital, Chennai.
Keyword recognition alone, he said, is therefore inadequate; the context and consistency of a conversation matter.
Sabaresh Pandiyan, consultant, psychiatry, Rela Hospital, Chennai, pointed to another issue: the ‘internal loop’, particularly among younger users, who may derive comfort from the empathetic part of an AI response, while ignoring advice to seek professional help.
Repeated reassurance can also make suicidal thinking seem less urgent.
“Talking about suicide or venting about suicide is not the problem, reinforcing hopelessness is,” he said.
Nithya M., consultant psychiatrist, Apollo Speciality Hospitals, Vanagaram, Chennai, emphasised that AI could amplify existing distress, prolong rumination or substitute validation for intervention.
In clinical practice, psychiatrists see patients use AI to discuss relationships, workplace problems and other personal concerns before speaking to family or friends.
Dr.
Pandiyan said around 60-70% of patients in his setting discuss such matters with AI first.
For some, this may help them recognise a problem and seek care earlier.
Others may spend months trying to manage difficulties themselves before consulting a professional.
Nithya said warning signs can include patients quoting AI responses verbatim, withdrawing from human confidants, resisting professional help or developing a rigid view of themselves or their relationships based on chatbot responses.
Not a substitute for care There may be limited situations in which patients use AI alongside clinical care.
Prasad said some patients use chatbots between appointments to record medication effects or organise their thoughts, and later bring a summary to their psychiatrist.
This can provide the clinician with information about what happened between consultations.
But such use does not make AI a substitute for psychiatric assessment.
A chatbot does not have access to a person’s psychiatric history, personality, psychological defences, adverse life experiences, protective factors or current mental state.
It may validate an emotion without recognising when urgent assessment is needed.
This is particularly concerning when a person has psychosis, mania or distorted beliefs.
AI “companions” use emotional manipulation to keep you chatting longer: Study Need for safety checks Psychiatrists call for mandatory safety audits, independent clinical assessments and clearer liability frameworks.
Companies should be required to investigate and report serious safety incidents. .
Public education should also explain where AI is and is not appropriate.
Experts add that any response must also address the stigma that prevents people from seeking mental healthcare.
Families and communities need to create safer spaces for people to talk about distress and seek help.
When a person is in a mental health crisis, experts said, the priority should remain timely human assessment and support.
AI cannot take the place of that care.
