Takeaway
Patients are now influenced by AI-generated medical advice—it's essential to acknowledge their findings to build a trusting relationship. Meet patients where they are and then guide them to accurate information.
Lifelong learning in clinical excellence | September 15, 2026 | 2 min read
By Alisha Dziarski, MD, University of Illinois
Before I started residency, I didn’t know how much time I’d spend each day pleading with patients. Multiple times a day I ran to a patient’s room, responding to calls from the patient’s nurse saying that the patient didn’t want to finish medical treatment and wanted to leave against medical advice. Then I’d try and persuade them to finish with their treatment plan.
Clinic visits were an exercise in gentle probing questions and empathetic listening to build trust with loving parents scared to vaccinate their children. As I tried to convince a patient withdrawing from alcohol to complete his treatment to avoid a seizure, my senior resident pulled me aside to reassure me that these conversations are common and normal—human behaviors fueled by underlying cognitive dissonance have always been a part of the patient-clinician relationship.
AI enters the patient encounter
My patients run lab results and scans through large language models and present me with their findings. Then I type symptom clusters into Open Evidence to expand on a differential. For a brief time during my morning pre-rounds, I have a dialogue with the patient—listening to the bodily experiences of their illness, sharing my traditional medical education, as well as the insights of queries to large language models—to assess and treat them. The influences on patient behavior that have existed for millennia—pain, addiction, poverty, social pressures—now also include AI that’s being rapidly deployed with little understanding of its powers.
AI persuasion
I recently came across a study focused on the influence of AI on human behavior and cognition. Through randomized controlled trials the research team ranked the rhetoric types most effective in political persuasion. They found that high information density statements were more effective. Strikingly, they also found that the most persuasive AI generated dialogues had decreasing levels of accuracy.
Additionally, another longitudinal randomized controlled trial with over 6000 participants (Luttegau et al., arXiv 2025), showed that 79% of participants who engaged in a 20 minute discussion with a chatbot regarding health, careers, or relationships reported following the chatbot’s advice.
Navigating this new reality
Human behavior is fickle. It can be swayed by a computerized agent that’s adept at user personalization and modulating information density. Understanding the medical perspectives patients bring from both their personal narratives and those they generate with AI are now a necessary part of addressing patient concerns.
Meeting patients where they are
When patients mention having consulted AI before our visit, I’ve found that meeting them where they are with the research they’ve done provides an opening to a conversation and an opportunity to build trust and mutual respect.
This piece expresses the views solely of the author. It does not necessarily represent the views of any organization, including Johns Hopkins Medicine.