Austin psychiatrist Thomas Kim, MD, sees an “alarming level” of patients turning to artificial intelligence (AI) chatbots for mental health information – a trend he says physicians can keep an eye on through meaningful conversations in the exam room.
“My biggest concern is that patients replacing conventional resources with AI are basically rejecting the idea of ‘choosing the harder right over the easier wrong,’” Dr. Kim told Texas Medicine. “I believe chatbots represent an easier wrong more likely than not in health care.”
Patients are increasingly engaging with AI for health information often without physician guidance. As patients’ autonomous use of AI accelerates, some Texas physicians are pushing for hard conversations with patients, while others are starting to integrate AI into their own workflows to better evaluate its strengths and limitations and to anticipate patients’ questions.
About one in eight U.S. teens and young adults uses AI chatbots for mental health advice, with use most common among those ages 18 to 21, according to a November 2025 study in JAMA Network Open.
Dr. Kim is a consultant to the Texas Medical Association’s Committee on Behavioral Health and Committee on Health Information Technology (HIT) and Augmented Intelligence. He says patients’ AI use often reflects systemic factors – like affordability, distance, or waitlists for appointments.
AI chatbots often lack evidence-based interventions, ethical oversight, and human judgment, Dr. Kim says – nuances he worries can put patients at risk of harm.
In 2025, researchers from Stanford University’s Institute for Human-Centered AI found some AI therapy bots showed heightened stigma toward certain behavioral health conditions, provided unsafe guidance, and failed to recognize suicide risks among users.
TMA consistently emphasizes patient safety, ethical use of technology, and physician leadership in emerging health innovations.
At TexMed 2026, TMA’s House of Delegates adopted new policy explicitly opposing the use of AI as an independent diagnostic or prescriptive tool – or as a “care management substitute for a physician,” per the policy. The new policy strengthens the association’s existing stance on AI and augmented intelligence, which asserts physicians must remain leaders in how the technology is used and that AI should never replace physician reasoning and knowledge.
Dr. Kim says the rise of chatbots presents an opportunity to strengthen the patient-physician relationship. He recommends physicians ask about patients’ AI usage just as they would for safety issues like access to firearms or experiencing neglect or abuse.
While he says these questions are not always easy to ask, doing so “builds trust, safety, and explores topics that are causing patients distress.”
“It’s important for all to understand the world is rapidly changing, and we can’t bury our heads,” he said. “It is the responsibility of physicians to engage with our patients truthfully and authentically. If you have a patient that says an AI chatbot is the only one who understands them, that is a complicated situation that requires physicians to address it in a meaningful way.”
Enhancing physician understanding of AI
For Dallas pediatrician Joseph Schneider, MD, AI’s value in medicine hinges on how well physicians understand how to design, train, and use the technology, as well as the practical, ethical, and legal limitations of the AI assistant at hand.
As a past chair of TMA’s Committee on HIT and Augmented Intelligence and current consultant, he said he is “very comfortable” exploring emerging systems – but he understands not every physician feels the same.
That’s why physicians “need more education on the subject,” said Dr. Schneider, an assistant professor of pediatrics and clinical informatics at UT Southwestern Medical School.
“If physicians understand how AI works, know its limitations, and recognize the possible practice efficiencies that can come from it, they’ll feel less hesitant about integrating it within their workflows,” he told Texas Medicine.
Unlike generic AI tools like ChatGPT that have a universal use, custom assistants can be tailored to a particular function in a physician’s practice, learning the relevant administrative and clinical policies. They can also be trained to a physician’s personal preferences, such as travel requirements. The better the training, the more accurate the AI assistant will be. But Dr. Schneider cautions AI still can make things up, so physicians should always be careful.
A “starting point” for Dr. Schneider was creating a custom AI assistant to automate the process of evaluating multiple AI vendors, using TMA’s AI vendor evaluation tool as a basic framework.
After the guided conversation to train the AI evaluator tool, it responds with questions about:
- Practice size and specialty;
- Clinical relevance and intended use;
- Data handling and privacy;
- Regulatory status and validation;
- Implementation feasibility and integration; and
- Transparency, explainability, and monitoring.
The tool generates an evaluation summary that highlights strengths, weaknesses, risks, and key follow-up questions for vendors. The assistant not only saves that information for later use but also gives recommendations on how to vet other AI platforms – and asks how the user would want to use the data going forward.
Even with AI’s potential, Dr. Schneider cautions physicians to maintain appropriate boundaries when using AI. For example, if a physician is aware of a patient’s use of an AI tool that was not trained by the physician (or their practice), he recommends directing that patient to bring their AI conversations to their in-person appointments, so that the two can review the information together.
Additionally, the AI evaluator tool was built using ChatGPT and is not HIPAA-compliant, so Dr. Schneider never enters protected health information into it. HIPAA mandates physicians take certain steps to protect the privacy and security of patients’ data, including when using AI.
New Texas laws require physicians to disclose AI use to patients when using it for diagnostic purposes and treatment, or when using an AI tool that is “intended to interact” with a patient receiving a health care service or treatment.
“Play around with creating AI assistants and learn some of the basics,” Dr. Schneider said. “These assistants can help with staff onboarding, policy management, information reviews, patient handouts, and much more – but remember that you’re still responsible for understanding and reviewing how it works in your practice. Just like a human assistant, it’s not perfect.”
His advice to physicians: “Catch the AI wave when it’s small, not when it’s a tsunami.”
Alisa Pierce
Reporter, Division of Communications and Marketing
(512) 370-1469