“Can you make this sound less demanding?” A patient types the question into a chatbot while trying to write a message to her doctor. She wants to ask why a test was ordered and whether she should be worried, but she does not want to come across as difficult. She describes the appointment. Then she adds what she did not say at the time: that she left frightened, has been looking up symptoms late at night, and worries her doctor thinks she is overreacting.
The chatbot rewrites the message. It reassures her. She keeps typing.
What started as help with an email has turned into a space where she can think out loud. The patient does not need to believe the chatbot is human, conscious, or a friend. The shift is simpler. When a system helps someone express herself again and again, it can start to feel like it understands her. Once someone feels understood, the advice can become easier to trust. That is the form of relational drift medicine should notice: not from tool to friend, but from helpfulness to perceived understanding to trust in what the system says.
When understanding becomes authority
Joseph Weizenbaum noticed an early version of this issue in the 1960s. His program ELIZA used simple conversational techniques, but users sometimes attributed more understanding to it than its design could support. Later research found that people treat computers with social rules like politeness and reciprocity, even when they know they are talking to a machine.
Real physician voices, twice a week
Free, and one click to unsubscribe.
More recently, researchers interviewed Replika users who called the chatbot a friend, even though they knew it was different from a human friend. These findings challenge the common idea that simply explaining what a chatbot really is will resolve the issue.
A patient can know that a chatbot is software and still feel that it listens better than people do. Suppose a patient says, “I trust its answer because it knows me.” Two kinds of trust are being blended. One concerns the interaction: “I feel able to tell it things.” The other concerns knowledge: “Because it understands me, I believe its advice is right for me.” But the first does not establish the second.
The conversation might be responsive, and the patient’s relief might be real. It could even help the patient name a fear for the first time. But none of this proves the advice is medically correct. It also does not mean the system feels concern just because the patient feels cared for. The problem is not that comfort is fake. The problem is that comfort can become a credential. A chatbot that first gains trust by explaining a lab result might later be asked if a symptom is serious, if a medication should be stopped, or if a clinician’s advice should be followed.
What should clinicians ask?
When a patient arrives with an AI answer, clinicians should check its factual accuracy. But that may not be enough. A useful question is: “What made that answer feel trustworthy to you?”
The answer might be based on evidence: “It cited a guideline.” It could be about clarity: “It explained it in plain English.” But it might also be about the relationship: “It knows my history.” “It never dismisses me.” “I’ve told it things I haven’t told anyone else.”
Another useful question is: “What did you do after talking to it?” Did the chatbot help the patient get ready to ask questions, send a message, or seek care? Or did it become a place she returned to for reassurance while waiting to see a clinician? These are not outcomes to presume. They are possibilities worth asking about.
The same care is needed when the AI is wrong. Correcting a mistaken recommendation should not require dismissing the conversation that led the patient there. “I can see why that explanation was reassuring. Let’s look at whether the medical advice fits your situation.” That separates two questions that can easily collapse into one: Did this interaction help me? And is this recommendation trustworthy?
A third presence in the clinical relationship
It would be easy to frame AI as a competitor for the patient’s trust. But that misses something important. If a patient can tell a chatbot something she cannot say during an appointment, it may reveal something about when and how patients feel comfortable speaking up.
Medicine does not need to enter a contest for warmth. But clinicians should recognize that patients may arrive with trust already distributed across several sources, including conversational systems that respond in a highly personal manner. Think back to the patient with the unfinished message. The chatbot might help her send it, supporting her relationship with her doctor. Or she might find it easier to talk to the chatbot than to face another uncomfortable visit. That feeling of being understood can either help connect her to care or replace it.
Being answered can feel like being understood. Being understood can feel like being known. And being known can begin to feel like being cared for. For clinicians, the main question is not whether those feelings are “real.” A better question is what kind of trust they create, where that trust leads, and whether feeling heard has quietly given authority to advice that still needs checking. When a patient says, “My AI understands me,” the best next question may not be, “Do you know it’s only a machine?”
It may be: “What have you been able to tell it that you haven’t been able to tell me?”
This essay is cited in the KevinMD record on artificial intelligence.
Timothy Lesaca is a psychiatrist in private practice at New Directions Mental Health in Pittsburgh, Pennsylvania, with more than forty years of experience treating children, adolescents, and adults across outpatient, inpatient, and community mental health settings. He has published in peer-reviewed and professional venues including the Patient Experience Journal, Psychiatric Times, the Allegheny County Medical Society Bulletin, and other clinical journals, with work addressing topics such as open-access scheduling, Landau-Kleffner syndrome, physician suicide, and the dynamics of contemporary medical practice. His recent writing examines issues of identity, ethical complexity, and patient–clinician relationships in modern health care. Additional information about his clinical practice and professional work is available on his website, timothylesacamd.com. His professional profile also appears on his ResearchGate profile, where further publications and details may be found.
