# How Should Therapists Respond When Clients Use ChatGPT for Mental Health Advice? By Cody Saunders, LMSW • October 4, 2026 Start by making room for the conversation. A client who brings an AI reply into therapy is showing you part of how they seek help. Listen to what they hoped to find and what they did with the reply. You can use that moment to build trust, check the advice, and make a clear plan together. The aim is not to win an argument with a chatbot. It is to help the person use sound information and stay connected to care. A warm response does not mean agreeing with every AI answer. ## Welcome the topic without shame A person may use ChatGPT to sort thoughts, find words for a feeling, or get a quick answer between visits. Those are possible reasons, not claims that the tool treats a condition. Ask about the purpose and the effect on the person’s daily life. A simple opening might be: **“I’m glad you brought this in. Let’s look at what felt helpful and what we need to check.”** This is a suggested conversation, not a tested treatment method. Avoid making the client defend their choice before you understand it. Focus on what they read, how they understood it, and whether it changed their actions. A client can feel heard while you correct a false or unsuitable claim. ## Explain the difference between a reply and care ChatGPT is a general AI tool that creates replies from patterns in language. It can sound warm and sure without having assessed the person. A clear answer is not proof of a correct diagnosis or a good care plan. The American Psychiatric Association’s [2026 position statement on AI in mental health services](https://www.psychiatry.org/getattachment/4f7b33ca-40f2-4f23-b29a-f2204cc17d4e/Position-Use-of-AI-Mental-Health.pdf) sets out professional concerns and recommendations for these uses. It is a professional position, not a law or proof that all tools behave the same way. Keep general chatbots distinct from tools designed and tested for a specific clinical purpose. Evidence about one tested tool does not establish that a general chatbot is a substitute for therapy. You might say: **“It can help you put thoughts into words. It has not done a clinical assessment. We’ll check any advice that affects your health or care.”** ## Review the advice in context With the client’s permission, look at the relevant part of the reply. You do not need their whole account or chat history. Follow your practice’s rules for handling any messages or screenshots. Do not copy them into another unapproved AI tool. Separate three things: what the client told the tool, what the tool replied, and what the client understood. A missing detail can change the answer. An added claim can also change its meaning. Use trusted sources to check factual claims. For care advice, use your trained judgment within your role and the client’s full context. If a reply suggests a medication change, direct the person to the prescriber for review. Do not treat the AI reply as an order. A useful next step is specific: clarify a fact, bring a concern to the right care professional, or choose a coping step already agreed in treatment. Keep the plan linked to the client’s goals. ## Turn a useful draft into a care conversation Consider a made-up case. A client uses ChatGPT to write a list of feelings before a visit. The list helps them explain what has been hard. In session, the therapist invites them to change the words so they fit their own experience. The helpful part is the client’s voice and the discussion it supports. The therapist does not need to accept the chatbot’s labels. A draft can open the door while the person and therapist decide what it means. In another made-up case, a reply says that one argument proves a lasting mental health condition. The therapist checks what happened and explains why one event does not establish that conclusion. Together, they return to the assessment and care plan. These examples are teaching ideas, not real cases or evidence of better outcomes. They show how curiosity and clear review can work together. ## Pay attention to changes in use Look at the role the tool is taking in the person’s life. A short writing task differs from spending many hours seeking repeated reassurance. Notice whether use is helping the person act on goals or pulling them away from sleep, daily tasks, supportive people, or care. Do not diagnose a new condition simply because someone uses AI often. Assess the person’s symptoms, functioning, and context through your usual clinical process. If the tool seems to be reinforcing a false belief, respond to the person’s experience without affirming the belief as fact. Plan follow-up around the concern you find. A person may need a clearer boundary for use, more human support, or a review of their current care. The right step comes from assessment, not from the product name. ## Build simple boundaries together A client may choose to use a general AI tool. You can discuss ways to keep that use within a clear purpose. These are suggested teaching points, not a claim that they make every use safe: - Use a draft as a starting point to discuss, not a diagnosis or final care decision. - Check health claims with trusted sources or a trained care professional. - Bring advice that conflicts with the care plan to the right clinician before acting on it. - Keep human support available. Do not rely on a chatbot as the only source of help. - Limit private details. Review what the service stores, uses, and shares. For a client who wants help preparing for therapy, a paper list or a note on their own device may meet the same need. Offer options that fit their comfort and skills. Do not make AI use a requirement for taking part in care. ## Explain privacy in plain words A personal app may not have the same privacy duties as a health care provider. The FTC explains that many health apps may fall outside HIPAA and that other consumer protections can apply. [FTC guidance on health apps and privacy](https://consumer.ftc.gov/features/protecting-your-privacy-online). Encourage the client to review the exact service and settings. Ask them to think before sharing names, records, or details about other people. Deleting a chat does not by itself prove that all copies are gone. A therapist recommending a tool or using it in care needs a separate practice review. A client’s own use does not approve the tool for the organization. For that review, read [Can I Put Client Information Into an AI Tool?](/resources/client-information-ai-tools/) and use the [data-handling worksheet](/resources/ai-data-flow-map/). ## Use human crisis support when needed If the conversation reveals immediate safety concerns, move into your established assessment and crisis process. Do not wait for the chatbot to decide the next step. This article is not a crisis assessment guide. In the United States, people can call or text **988** for crisis support. The 988 Lifeline advises calling **911** when someone is in immediate harm or danger. Outside the United States, use local crisis and emergency services. [988 Lifeline contact guidance](https://988lifeline.org/contact-us/). Make the client’s support plan clear, including who to contact and when. A general chatbot is not a replacement for that plan or for emergency help. ## Document what matters for care Follow your usual record standards. When relevant, note the client’s reported use, the advice that affected care, your assessment, and the agreed next steps. Keep the amount of detail suited to the care purpose. Do not save a full chat history by default. A sample structure is: “Client reports using an AI tool for [purpose]. Discussed [care-relevant point]. Reviewed [finding or concern]. Agreed on [next step and follow-up].” This is a suggested structure, not a required form or a completed clinical note. Keep client reports and your assessment distinct. Do not place an AI diagnosis into the record as an established fact. If a message needs review or storage, use the approved process. ## Make the conversation part of good care A calm response makes it easier for clients to bring outside advice into the room. Listen first, check the content, explain the limits, and agree on a useful next step. Return to the plan at later visits when needed. AI may help a person prepare words or questions. Therapy adds a trained human relationship, context, and care over time. Keep those strengths at the center. The goal is a better connection with the client and a clearer path to helpful care. ## Explore agreement and a sense of connection Some models may agree with a user even when the answer has a weak basis. This is called **sycophancy**. [Vendor research from Anthropic](https://www.anthropic.com/research/towards-understanding-sycophancy-in-language-models) found it in tested models; it does not show how every product affects therapy. Ask what the tool said and how the client used it. Explore the advice together without treating its agreement as a clinical finding. A client may also feel close to an **AI companion**. A one-sided sense of connection is sometimes called **parasocial attachment**. The term is not a diagnosis. Ask about daily use, sleep, real relationships, and whether it supports or gets in the way of care. A warm tone does not establish a product’s clinical value. Some users try **jailbreaking**, or bypassing a model’s safeguards. Avoid assuming that built-in limits will hold in every chat. In the discussion, focus on the actual output, the person’s needs, and your usual care process. AI-literacy additions reviewed October 11, 2026. See [AI Terms for Care Teams](/resources/ai-terms-for-care-teams/) for the related terms and examples. ## Sources and scope Sources checked October 4, 2026. This article is professional education, not individual treatment, legal advice, or a crisis protocol. Sample language and cases are teaching ideas, not tested interventions. No benefit is promised for ChatGPT or any other product. - [American Psychiatric Association: Position Statement on the Use of Artificial Intelligence for Mental Health Services and Treatment](https://www.psychiatry.org/getattachment/4f7b33ca-40f2-4f23-b29a-f2204cc17d4e/Position-Use-of-AI-Mental-Health.pdf), 2026. Professional recommendations; not legislation. - [FTC: Protecting Your Privacy Online](https://consumer.ftc.gov/features/protecting-your-privacy-online). Consumer guidance, including health-app privacy. - [988 Lifeline: Contact Us](https://988lifeline.org/contact-us/). U.S. crisis contact and emergency guidance.