Practical guidance for working with clients who use ChatGPT and other AI tools for mental health support.
A client says, “ChatGPT thinks I have ADHD.” Another brings in an AI analysis of their relationship. Someone else has been talking to a chatbot late at night when anxiety spikes.
These moments can create an understandable tension for therapists. You may want to correct inaccurate information, protect the client from overreliance, or first understand why the AI response felt important.
That third instinct is often the best place to begin.
In a GoodTherapy LinkedIn poll of 101 respondents, 81% said their first instinct when a client brings AI-generated insights into a session is to “explore what it means to them.” The poll was informal and self-selected, but it points to a useful clinical principle: before deciding what the AI got right or wrong, understand what the interaction meant to the client.
The issue is already common in practice. The American Psychological Association’s 2026 survey of 1,242 U.S. psychologists found that 77% had patients who talked about using AI, 39% had patients using it for self-diagnosis, and 35% had patients using AI as an additional mental health resource.
For therapists, the question is no longer simply whether clients are using AI. It is how to work thoughtfully with what they bring into the room.
A useful starting point is Ask, Understand, Bridge:
Ask how the client is using AI, without making the disclosure feel like a problem
Understand what needs, questions, or patterns may be underneath the use
Bridge what you learn back into assessment, the therapeutic relationship, and treatment goals
It is an emerging conceptual approach, often used in digital mental health, that helps therapists apply familiar clinical skills to safely integrate the use of artificial intelligence tools between sessions, while supporting therapy seekers in using those tools more thoughtfully.
Clients use AI for different reasons: to understand a diagnosis, organize thoughts before therapy, get reassurance, find coping ideas, rehearse conversations, or feel less alone. Those uses differ.
Start by making AI use easy to disclose:
“A lot of people are using AI to think through mental health, relationships, or things they want to bring into therapy. Has that been part of your experience?”
If yes, stay curious:
The goal is not to give AI credibility it has not earned. It is to understand the client’s need and their emotional and mental state before using the tool.
If a client arrives with pages of notes of an AI-generated diagnosis, they may have spent considerable time trying to make sense of something utterly confusing. Beginning with correction can unintentionally dismiss that effort and put the client in defense mode. Curiosity keeps the door open for a deeper conversation.
Your client’s conversation with AI can provide useful information, but it can be just as important to understand what led the client to turn to it in the first place.
They may have been looking for reassurance, trying to make sense of something confusing, putting difficult feelings into words, or simply wanting somewhere to turn when support did not feel immediately available.
Consider exploring:
What need is being met? Reassurance, certainty, validation, information, companionship, or a place to share something difficult?
When does the use increase? If a client repeatedly turns to AI after conflict, late at night, or when health anxiety spikes, that timing may point toward a trigger.
What happens afterward? Does the client feel clearer and better able to act, or does the relief fade quickly and lead to another round of reassurance-seeking?
What feels easier about AI than human connection? A client who says, “It understands me better than people do,” may be telling you something important about judgment, vulnerability, safety, or previous experiences of not being heard.
There is no need to force an interpretation. Ask instead: What does this behavior help us understand about the person sitting in front of me?
Once you understand what the AI interaction is doing for the client, you can bring that information back into your 1-1 sessions.
If a client wants to share an AI conversation, look at it together as you might review a journal entry. Ask what felt accurate, what did not fit, what assumptions the response made, and what important context about the client’s history, culture, relationships, or symptoms was missing.
This lets the client keep what felt meaningful while also practicing critical evaluation.
When a client says, “ChatGPT thinks I have ADHD,” you do not have to choose between validating the diagnosis and rejecting it. Start with the meaning of the label: What made them ask? What felt familiar? What would having an explanation mean to them?
Then return to clinical assessment. AI can produce plausible explanations, but it does not have the full history, differential assessment, context, or professional accountability involved in diagnosis. Both the APA and American Counseling Association caution against relying on general-purpose AI for mental health diagnosis.
The chatbot or LLM may have produced the label. Therapy can explore the experiences that made the label feel important.
Frequent use of AI should not automatically be assumed to be harmful. What matters is the role it is beginning to play.
If a client repeatedly asks an AI tool for reassurance, avoids decisions without AI input, withdraws from people, loses sleep to long conversations, or treats the tool as the only source they trust, explore the pattern without shame and judgment:
“It sounds like this has become one of the first places you go when you feel overwhelmed. What does it give you in that moment that no other agency can?”
Clinical judgment should take the lead when AI use intersects with significant risk or appears to intensify certain symptoms.
Pay closer attention when AI is:
In these situations, continue to use appropriate risk assessment, safety planning, consultation, documentation, referral, and crisis procedures based on your profession, setting, and jurisdiction.
A chatbot conversation may feel private, but it does not necessarily carry the same confidentiality protections as therapy.
Clients should think carefully before entering highly sensitive or identifiable health information into consumer AI tools. In the United States, HIPAA protections generally apply to covered entities and their business associates; information voluntarily shared with many consumer apps may fall outside those protections.
Therapists should also follow applicable privacy laws, licensing requirements, employer policies, informed-consent obligations, and professional ethics before using AI tools with client information.
The goal is to help clients make informed choices, not frighten them away from technology.
Sometimes. AI may help clients organize thoughts before a session, generate thoughtful questions, rehearse a conversation they were otherwise finding difficult, review a coping skill they have already learned, or consider another perspective.
The important distinction is whether AI is supporting the client’s capacity or gradually replacing it.
A useful question is: Is this helping you bring more reflection, choice, and connection into your life or is it becoming the place where those things stop?
Ask. Make AI use discussable without judgment.
Understand. Explore what the client is seeking, when they use AI, and what happens afterward.
Bridge. Bring those insights back into assessment, treatment goals, and the therapeutic relationship.
AI-generated content does not have to be treated as either a threat to therapy or a source of truth. Often, it is simply another piece of the client’s world. The therapist’s role is not to compete with the chatbot. It is to help the client understand what they are looking for, decide what is useful, recognize what may be harmful, and build insight, judgment, tolerance, and connection that extend beyond any single AI conversation.
Not automatically. Start by asking what they use it for, how often, and how it affects them. Concern increases when AI use intensifies symptoms, replaces needed care or relationships, drives risky decisions, or interferes with daily functioning.
Explore what led them to seek a diagnosis and what the label means to them before assessing whether it fits. AI-generated diagnoses should not replace a comprehensive clinical assessment.
It can be. AI may support reflection, organization, or practice of previously learned skills. Help clients also consider accuracy, privacy, overreliance, and whether the use supports their treatment goals.