How Is AI Reshaping Trust and Communication in the Patient-Physician Relationship?

Artificial intelligence (AI) is increasingly embedded in oncology practice, from ambient documentation tools and prior authorization support to clinical literature searches and guideline navigation. As adoption grows, the conversation has shifted from whether AI belongs in the clinic to how physicians and their patients should navigate it together.

CancerNetwork® spoke with Margarita Racsa, MD, MPH, board-certified radiation oncologist in Daytona Beach, Florida, and co-chair of the American College of Radiation Oncology (ACRO) Artificial Intelligence Subcommittee about the current state of AI use among oncologists, how clinicians can communicate transparently with patients as AI expands from documentation to clinical decision support, and what this technology means for the patient-physician relationship.

Racsa began by surveying the administrative and clinical landscape of AI use in oncology, drawing on recent survey data showing that more than half of oncologists are already incorporating AI into their practice in some form. She then detailed a practical, 4-layered communication framework for introducing AI into patient encounters, built on transparency, written and verbal consent, and a clear opt-out pathway. Finally, she offered a candid assessment of AI’s impact on the patient-physician relationship: the genuine opportunity to empower more informed patients alongside a real risk of trust erosion when AI-generated information conflicts with a physician’s clinical judgment.

CancerNetwork: How are oncologists currently using AI in clinical practice?

Racsa: In terms of oncology use, there are 2 broad categories: administrative and clinical. Far and away, most physicians are more comfortable using AI in an administrative capacity. Some of the most recent data comes from a survey conducted by Doximity, in which approximately 63% of physicians reported “currently using AI” in their practice.¹ Specifically among oncologists, approximately 57% reported “currently using AI” in their practice.

When looking more closely at specific use cases, the most common administrative application is the ambient scribe. Many patients who have visited a physician’s office recently may have noticed their physician saying, “Before we begin, I want to let you know we will be using AI during this consult to help me write my note.” Approximately 29% of physicians are currently using ambient scribes.Other common administrative applications include writing prior authorizations, drafting patient letters, developing chart summaries before a patient’s visit, and assisting with billing codes.

On the clinical side, AI can be used as an aid for conducting literature searches or looking up standards of care, such as NCCN [National Comprehensive Cancer Network] guidelines or information from RTOG [Radiation Therapy Oncology Group] studies. Approximately 35% of physicians are using AI to conduct literature searches.While AI use cases for advanced clinical decision support (CDS) certainly exist, the level of data on physician adoption is currently limited.

What framework should clinicians use to communicate with patients about AI as its role in oncology care expands?

In some ways, this is not very different from what we already do as physicians. With respect to any clinical decision or treatment we propose, the framework is essentially: what is the treatment, and what are the risks, benefits, and alternatives? The same framework applies to introducing AI in the clinical setting; being transparent and upfront with patients about what we are using and why.

A useful model, particularly for ambient AI, is what I refer to as a layered approach. The first layer is having signage in the waiting room indicating that a physician may use AI during a visit; a visual alert so patients are aware before they enter the room. The second is providing written information, paper or digital, giving a more detailed explanation patients can read at their own pace before seeing the physician. The third, and perhaps most important, is explicit verbal permission from the physician. Something as direct as: “To help me better take care of you so that I can focus fully on our visit today, I will be using AI. Is that okay with you?” Patients may have signed a stack of paperwork upon arrival without fully registering this, so that verbal moment of acknowledgment matters. The fourth element is providing a very easy opt-out option, verbal or written, at any point in the process. If a patient says they are not comfortable, the AI use stops. This is one of example of a comprehensive approach and an informed discussion with a patient that gives them genuine agency over how their care proceeds.

What impact is AI having on the patient-physician relationship, particularly with respect to trust and the quality of direct patient interaction?

At the moment, we do not have very robust data on this. What we know comes from publications including those from OpenAI, released earlier this year, indicating significant patient use of foundation models. For example, 1 in 4 weekly active users have used ChatGPT to answer a health-related question, which is remarkable. AI use among patients is only going to become more common, and that is reflected in the clinic. As a community-based radiation oncologist, and from conversations with medical (and radiation) oncology colleagues, we are increasingly seeing patients arrive with printed information from an AI interaction and very specific questions in hand.

The impact on the patient-physician relationship is, honestly, divided. Having patients come in with information is excellent when it helps them make a more informed decision. The challenge is that the quality of that information is variable. AI is generative, and the output depends on what the patient entered about their diagnosis and on the training data. For common conditions, the quality will be higher; for edge cases or clinical gray areas, it may be quite different. Compounding this is the fact that AI tends to deliver very polished, confident-sounding recommendations, and our ability as human beings to distinguish accurate information from inaccurate information is influenced by how confidently something is presented. That is where the real danger lies.

When a physician’s recommendation does not align with what AI told the patient, it can cause some patients to have less confidence in their physician; even after a careful explanation of why a particular recommendation does or does not apply to their specific case. There is real potential for AI to erode the patient-physician relationship, and these discussions also take additional time, adding pressure on physicians who are already operating under significant throughput demands.

On balance, I believe AI has the potential to help, but I want to caution patients to keep an open mind when consulting with their physician, because there are real limitations to using AI for individual healthcare information. Ultimately, patients benefit from an informed, shared decision-making approach: incorporating the patient’s values, preferences, and goals into a unified decision about next steps in their care.

References

1. State of AI in medicine. Doximity. Accessed August 28, 2026. https://tinyurl.com/yyfjcnee
2. AI as a Healthcare Ally: how Americans are navigating the system with ChatGPT. Open AI. January 2026. Accessed September 1, 2026. https://tinyurl.com/3xnv94ye
3. Raynes S, Maese E. Americans turning to AI to supplement healthcare visits. Gallup. April 14, 2026. Accessed August 28, 2026. https://tinyurl.com/yc5nhzyu