A patient tells a therapist that they use an AI chatbot to journal between sessions. Another says:

“The chatbot diagnosed me with bipolar disorder and told me I should stop taking my medication.”

Those situations present very different concerns. Patients are increasingly using AI to explore symptoms, understand diagnoses, and seek advice between appointments. In many cases, that use may have little significance to treatment. But when a patient relies on AI for diagnosis, medication advice, or other treatment decisions, the issue can become clinically and legally significant.

The 2026 American Psychological Association survey found that 77% of psychologists reported that patients had discussed their AI use, while more than one-third said patients were using AI as an additional mental health professional. The APA has encouraged psychologists to ask patients about AI use and discuss information patients receive from these tools.

A patient mentioning ChatGPT does not automatically require documentation. Providers cannot reasonably be expected to document every website, app, or online resource a patient encounters.

The situation changes when AI begins to affect treatment. Consider a patient who tells a psychiatrist that an AI chatbot believes the psychiatrist made the wrong diagnosis, or that an AI tool advised them to stop taking a prescribed medication. The provider now knows that the patient has received information that could affect their understanding of the diagnosis or willingness to follow the treatment plan.

The provider is not responsible for what the chatbot said. But the provider may need to address what the patient has been told. That distinction could matter later. If the patient stops taking the medication and suffers a setback, the provider may eventually have to establish what they knew and how they responded. Without documentation, that issue may come down to competing recollections.

A provider does not need to copy an entire chatbot conversation into the medical record. The relevant information is what has a bearing on the patient’s care and the provider’s response. For example, if a patient reports that an AI chatbot told them to discontinue a medication, the record could note that the patient raised the issue, that the provider discussed the risks of stopping the medication, and what treatment or follow-up plan was agreed upon.

The purpose is not to document the chatbot for its own sake. It is to document the clinical encounter. That can become important in a malpractice claim, licensing-board investigation, payor review, or other professional proceeding. If a patient later alleges that the provider knew they intended to stop medication and failed to respond, a contemporaneous note may provide important evidence of what actually occurred.

The APA has encouraged psychologists to ask patients about AI use, invite patients to bring AI-generated material to sessions, and consider addressing AI use in intake materials. Practices do not necessarily need a lengthy policy covering every possible use of ChatGPT or another AI platform. Instead, clinicians should have some guidance about when AI-related information becomes relevant to patient care.

For example, documentation and implementation of appropriate protocols may be warranted in situations where:

• a patient presents an AI-generated diagnosis that conflicts with the clinician’s assessment;

• a patient reports receiving medication or treatment advice from an AI tool;

• the patient intends to act on that advice; or

• the clinician discusses or corrects information the patient received from an AI system.

The point is not to document every reference to AI. It is to recognize when the patient’s use of AI becomes part of the clinical encounter.

Patients will continue to use AI alongside traditional mental health care. Providers do not need to police that use, and consulting an AI chatbot does not, by itself, create a compliance problem. The issue arises when AI begins influencing the patient’s treatment, medication decisions, or understanding of their condition.

When that happens, the provider should consider whether the issue needs to be addressed and documented. The AI-generated advice may not be the provider’s responsibility. But once the provider knows about it, the provider’s response may become part of the record, and eventually part of the evidence.