Max Parikh, MD, founding partner of the Advanced Ophthalmology Institute and an award-winning surgeon, joined professional co-editor Rehan Ahmed, MD, to discuss the rapid rise of large language models (LLMs) as everyday clinical companions for patients. They explore how consolidation of device and implant data can create patient insights, and combined with conversational AI, bring new opportunities and responsibilities for clinicians.
Impact of LLMs on Consultation: More Questions, Less Time
Dr. Parikh and Dr. Ahmed break down an important trend across the eye care space: the sharp increase in patients consulting large language models (LLMs) – like ChatGPT, Gemini, Claude, etc. – before their appointments.
Dr. Parikh shares a striking data point: there are 50 million health inquiries per day on Bing and Copilot. He characterizes the information patients bring to consults as abundant but uneven, as AI-generated responses can sound authoritative, yet include errors or lack clinical nuance.
In turn, these AI-generated results then change the clinical encounter. Some patients arrive well prepared and ready to proceed after a few clarifying answers, while others bring layered, AI-generated question lists that strain standard consult time. Dr. Parikh says, “It’s really difficult to stay on a timeline and answer their questions.”
Are There Solutions?
Dr. Parikh and Dr. Ahmed both agree that the practical solution is to develop dedicated, clinically-vetted tools that deliver accurate information. Dr. Ahmed suggests domain‑specific LLMs with medical guardrails that allow deep, persistent patient conversations. This would provide information outside the clinic, while preserving clinician time during the visit. “If we can direct them to a platform that we can all agree upon that’s excellent, that would be great,” says Dr. Ahmed. He notes that such direction could raise the quality of information patients receive.
Dr. Parikh concurs, “When patients do their query on a large LLM that doesn’t have those guardrails, the confidence interval of getting the right information is much lower . . . If we can direct them to a platform that we as clinicians can all agree is excellent, that would be great. Then, you can at least feel like the data is maybe 97%, 98% accurate, as opposed to 70% accurate. It’s the 30% that’s wrong that really creates the problem.”
Both clinicians emphasize that the net effect of AI is positive if the profession converges on validated, transparent tools that improve outcomes without eroding the patient and physician relationship.
LLMs are already empowering patients in ways clinicians cannot ignore. The urgent task is to shape AI capability so it amplifies clinical judgment, protects patients from unvetted information and does not affect consultation efficiency. When implemented properly, these tools will enable more timely, evidence‑based care for patients.
For more on this conversation, listen to this episode of Real Talk.
Author
