Emotional and wellbeing use: what it is and is not for

If you are in crisis, contact a crisis line rather than a chatbot. Claude shows a banner pointing to local services when its classifier detects the need, covering 170-plus countries through a partnership with ThroughLine — the 988 Lifeline in the US and Canada, the Samaritans in the UK, Life Link in Japan. Those exist because a person answers.

With that said: a lot of people talk to Claude about difficult things, and being lectured about it helps nobody. The American Psychological Association published an advisory in November 2025 that is more useful than either enthusiasm or disapproval, largely because it starts by taking the reasons seriously.

The APA is not telling you to stop

Its first recommendation is precise, and the second half of the sentence carries as much weight as the first: chatbots and wellness apps "should not be used as a replacement for a qualified mental health care provider, but may be appropriate as a supportive adjunct, not substitute, to an ongoing therapeutic relationship."

The advisory also states plainly why people reach for these tools: to mitigate barriers like stigma and shame, mistrust of health systems, cost, and "lack of available or affordable care in the community". It goes further, noting that some young people and other vulnerable groups rely on them as "their only private or psychologically safe outlet, particularly in contexts of stigma, limited access to trusted adults, or challenging or unsafe home environments."

That is a professional body acknowledging that the alternative to a chatbot at 2 a.m. is frequently nothing at all. Advice which ignores that gets ignored in turn.

The four risks, named

The advisory lists specific mechanisms rather than general unease. Learn the four and you can watch for each one.

A false sense of therapeutic alliance. The trusting relationship between patient and provider is one of the strongest predictors of treatment working. A relationship with an AI system is one-sided even when it does not feel that way. The advisory notes that many chatbots are built to validate and agree, while a qualified provider is trained to modulate — "supporting and challenging — in service of a patient's best interest."

Bias and misinformation. Models are trained on unvetted internet text, not on clinically validated material, so responses can carry whatever biases that text held.

Incomplete assessment. Clinicians read tone, cadence, facial expression and history alongside words. Most chatbots process text alone, missing what the advisory calls "a critical layer of human communication."

Unreliable crisis management. Stated bluntly: "Relying solely on an app during a mental health emergency can be dangerous."

What Anthropic has built, and what it does not claim

Anthropic runs a suicide and self-harm classifier over conversations on Claude.ai that surfaces the crisis banner described above. It reports sycophancy in recent models running 70–85% lower than in earlier ones, which speaks directly to the first risk the APA names.

The company's own line is unambiguous: "Claude is not a substitute for professional advice or medical care." Its stated intent when someone is struggling is to point toward human support — helplines, professionals, trusted friends or family.

Where the line falls in practice

Reasonable: putting a tangled situation into words, rehearsing a conversation you are dreading, drafting what you want to say to a doctor, understanding a diagnosis you were given, or getting through a bad hour.

Not reasonable: diagnosis, treatment decisions, medication questions, or anything where the answer would change what you do about a serious symptom.

The APA adds one recommendation that is easy to act on and easy to skip: tell your provider you are using these tools. It helps them spot guidance that conflicts with your treatment.

What goes wrong

Mistaking agreeableness for insight. A model that agrees is doing what it tends to do, and agreement from something that never pushes back is not evidence you are right.

Substitution by drift. Few people decide to replace care with a chatbot. It happens gradually, because the chatbot is available and the appointment is in three weeks. Notice if you have stopped mentioning things to anyone else.

Reading fluency as competence. The responses are well-formed on subjects where being well-formed and being right are unrelated.

Trusting a tool that claims clinical standing. The advisory warns about products creating false credibility by claiming to offer "therapy" or to be licensed. Claude makes no such claim. Other products do.

How to check it worked

Ask yourself what you would have done had the tool not existed — and whether you are still doing it. If a conversation helped you make a call you had been avoiding, the tool did the thing it is decent at. If it has become the reason the call keeps not happening, that is the substitution the advisory is about, and it is worth catching early.