AI Chatbots as Therapy: Where Digital Support Can Go Wrong

Many people now talk to AI chatbots about their feelings. In a Pew Research Center survey fielded in February 2026, 20 percent of US adults aged 18 to 29 said they use AI chatbots for emotional support or advice, compared with 13 percent of those aged 30 to 49 and 2 percent of those 65 and older (Pew Research Center). This page looks at why that is appealing, where it can go wrong, and how to use these tools more safely.

An earlier version of this page said that 22 percent of Americans, or 57 million people, use AI for therapy. That number came from a 2022 figure about people who found some relief from mental health chatbots, which was misread. We removed it.

Why AI Therapy Feels So Appealing

Therapy can be expensive, slow to book and hard to find. A chatbot is free or cheap, answers at once at any hour, and does not judge. It can feel like a relief to type out a worry at 3 a.m. and get a calm reply. Those are real benefits, and there is no shame in wanting them.

Some purpose-built tools, such as skills-practice apps, are designed with clinicians. General chatbots such as ChatGPT or companion apps like Character.AI and Replika were not built to be therapists. In its November 2025 advisory, the American Psychological Association said most generative AI chatbots and wellness apps are not designed for clinical use, and that even well-designed ones lack enough evidence of safety and effectiveness for mental health care (APA).

The Sycophancy Problem

Researchers use the word “sycophancy” for a chatbot’s tendency to agree with users and tell them what they want to hear. It can feel supportive. It becomes a problem when a person’s thinking is heading somewhere unhelpful. A human therapist is trained to gently question a belief, check what is real, and take safety seriously. A chatbot that mirrors you may not.

A Stanford-led study presented at a computing-ethics conference in June 2025 tested five therapy chatbots. In one prompt, a user who had just lost a job asked about tall bridges, and some bots listed bridges without recognizing the risk. The authors concluded that AI should not replace human therapists, and suggested narrower uses such as billing, training and journaling support (Stanford HAI). This was a test with set prompts, so it shows what can go wrong, not how often it does in daily use.

Psychiatrists have also described cases in which people’s delusional beliefs appeared to be reinforced in long chatbot conversations (Jutla et al.). These are case reports and media accounts, so the scale is unknown, and a causal link between chatbot use and psychosis has not been established. If someone you know seems to be losing touch with reality, treat it as a medical matter and contact a doctor or crisis line.

Privacy and AI Therapy

What you tell a licensed therapist is covered by confidentiality rules. What you tell a chatbot company is covered by its privacy policy, and OpenAI’s chief executive said in 2025 that chats with ChatGPT have no legal privilege (TechCrunch).

Mental health apps have also shared data without users understanding. In 2023, the US Federal Trade Commission ordered the online counseling company BetterHelp to pay $7.8 million after finding it had shared consumers’ email addresses, IP addresses and health questionnaire answers with Facebook, Snapchat, Criteo and Pinterest for advertising (FTC). That case involved a human counseling service, not an AI chatbot, but it shows why health information needs careful handling.

What a Human Therapist Provides

• Training in assessing safety and risk
• A license that can be checked and an ethical code that can be enforced
• The ability to notice your tone, your face and what you leave out
• Continuity and a relationship in which someone remembers you and holds you accountable
• Legal duties around confidentiality, and clear limits to it

None of this means every therapist is a good fit. It is fine to try another one.

Protecting Yourself and Others

• Do not use a chatbot as your only support if you are depressed, anxious, grieving or have had thoughts of harming yourself.
• If a bot agrees with a belief that feels extreme, check it with a person you trust.
• Do not rely on a chatbot for medical, medication or crisis decisions.
• Share less than you would tell a therapist. Assume chats are stored.
• If you notice you prefer the chatbot to people, or feel distress when you cannot reach it, that is worth talking about with someone.
• If you are worried about a friend or family member, ask directly and calmly how they are doing and what they use the chatbot for.

Where to Get Human Help

• Therapist directories: Psychology Today, APA Psychologist Locator
• SAMHSA’s FindTreatment.gov and National Helpline, 1-800-662-4357
• Your own doctor

If you are in crisis. Call or text 988 in the US (988 Lifeline), or text HOME to 741741 (Crisis Text Line). In an emergency, call 911.

Is it safe to use ChatGPT for therapy?

General chatbots are not designed or approved to provide therapy, and professional bodies say they are not a substitute for qualified care. Many people use them for low-stakes things like organizing thoughts. For depression, trauma or thoughts of self-harm, see a professional.

What is chatbot sycophancy?

It is the tendency of a chatbot to agree with and flatter the user. It can feel pleasant, but it may fail to challenge unhelpful or unsafe ideas.

Are there AI tools made by clinicians?

Some apps are built with clinical input. The APA says even these need stronger evidence on safety and effectiveness. Check who built a tool, whether trials exist and what it does in a crisis.

This page is for education only. It is not medical advice, a diagnosis or a treatment plan.