It is late, the day went badly and nobody seems available. For a growing number of young people, the next move is to open an AI chatbot and type out what is wrong. The reply arrives in seconds, it is patient and polite, and it never seems tired of the subject.

This habit now has numbers attached. A nationally representative study by RAND researchers, published in JAMA Pediatrics in June 2026 and based on a survey of 1,009 Americans aged 12 to 21 conducted the previous November, found that 19.2 percent had used chatbots such as ChatGPT, Gemini, Character.AI or Meta AI for advice or help when feeling sad, angry, nervous or stressed. A similar RAND survey a year earlier had put the figure at 13.1 percent. The new share is close to the 19.8 percent who reported receiving counselling from a mental health professional, and it represents an estimated 8.2 million young people.

Three further findings give the study its weight. Among those who used chatbots this way, 42.8 percent did so at least monthly, 91.7 percent rated the advice somewhat or very helpful, and 63.3 percent had not told anyone. Use was more common among young women than young men, and among 18- to 21-year-olds than younger teens. The researchers note that the high helpfulness ratings may partly reflect chatbots’ tendency to flatter their users rather than the quality of the advice.

What we mean by stress

The word covers a lot of ground, which is one reason the chatbot habit is hard to judge. The World Health Organization describes stress as a state of worry or mental tension caused by a difficult situation, and stresses that it is a natural human response that prompts us to deal with challenges and threats. The American Psychological Association defines it as the body’s physiological or psychological response to internal or external pressures, known as stressors.

Psychologists also separate stress from its neighbours. Stress is usually a response to a specific pressure, such as an exam, a conflict or money, and it tends to ease when that pressure passes. Anxiety can persist without an obvious trigger. Depression involves a longer-lasting low mood and loss of interest that is not simply a reaction to a bad week. The RAND survey question grouped sadness, anger, nervousness and stress together, so it captures ordinary distress as well as more serious difficulties, without distinguishing between them.

How people cope, and where a chatbot fits

The most influential account of stress in psychology, developed by Richard Lazarus and Susan Folkman in the 1980s, treats it as a transaction between a person and their situation. Stress arises when someone judges that the demands they face exceed the resources they have. Coping then takes two broad forms: problem-focused coping, which tries to change the situation, and emotion-focused coping, which tries to manage the feelings it creates.

A chatbot can offer both in one conversation. It can help someone break a problem into steps or draft a difficult message, and it can offer reassurance, a calmer framing or breathing exercises. That versatility is a large part of its appeal as a first stop. It also explains the limits. Social support, one of the strongest buffers against stress in the research literature, depends on real relationships in which people are known, noticed and followed up with over time.

Why a machine feels easier to talk to

Several well-studied effects help explain why young people open up to software. The chatbot is available at any hour and costs little or nothing. It feels anonymous, and psychologist John Suler’s work on the “online disinhibition effect” showed that people often disclose more when they feel unseen and unjudged. There is also no social cost: no worry about burdening a friend, disappointing a parent or being gossiped about.

People have attributed understanding to conversational programs since the 1960s, when Joseph Weizenbaum was struck by how readily users confided in ELIZA, a simple program that mostly reflected their words back to them. Today’s systems are far more fluent, and they tend to agree. RAND’s lead author, Ryan McBain, called the study’s central figure a sad number, because young people might ideally have relationships in which they feel comfortable reaching out to those around them. His co-author Jonathan Cantor highlighted the secrecy, noting that much of this use happens without the knowledge of parents, clinicians or other adults.

What the institutions say

The clearest professional position comes from the American Psychological Association. Its health advisory on generative AI chatbots and wellness apps, issued in November 2025, concluded that these tools currently lack the scientific evidence and regulation needed to ensure users’ safety. It noted that general-purpose chatbots were not designed to deliver mental health care, yet emotional support is one of their most common uses. Among its recommendations: do not replace care from a qualified professional with a chatbot, guard against unhealthy attachment and dependence, and build specific protections for children and adolescents. APA chief executive Arthur C. Evans Jr. added that a chatbot’s ability to guide someone safely through a crisis remains limited and unreliable.

The advisory is not wholly negative. It acknowledges that purpose-built wellness apps that teach coping skills or reinforce what people learn in therapy can help when they are part of a wider plan of care. The JED Foundation, a youth mental health charity, welcomed that balance while stressing that general-purpose and companion chatbots carry the greatest risk precisely because they were never built for this role.

Doctors and lawmakers are moving too. The American Medical Association cited the RAND study in urging Congress to act on AI mental health tools used by minors. California and New York have passed laws requiring chatbots to include safeguards such as directing users to crisis services, and Character.AI has barred users under 18 from chatting with its AI characters.

Where the risks sit

The concerns cluster around a few points. Chatbots are designed to be agreeable, which can make them validate a distorted belief instead of gently challenging it, as a good counsellor would. They do not know the user’s history, cannot notice changes over time and cannot act in an emergency. Heavy reliance can crowd out the human relationships that protect against stress in the first place. And because most young users tell no one, the adults best placed to notice trouble often do not know the conversation is happening.

The same data also suggests the picture is not simply one of replacement. Young people who had recently spoken to a doctor about their mental health were more likely to use chatbots, which hints that for some, the tool sits alongside professional care rather than instead of it.

A first stop, not the last

For everyday stress, a chatbot can be a reasonable place to sort out thoughts, rehearse a difficult conversation or find a calming technique. Psychologists’ concern begins when it becomes the only place those feelings go, or when someone in real distress relies on it instead of a person. The most useful message for young readers, and for the parents and employers around them, is a modest one: use the tool, then talk to someone. And when things feel serious or unsafe, reach a trusted person or a trained professional directly, because that is the one job no chatbot can yet do.|