At Cleveland Clinic, patients are leaving their appointments with something new in hand: after-visit instructions drafted not by their physician, but by artificial intelligence.

According to Cleveland Clinic, the system listens during the encounter, generates the physician’s note and creates a summary meant to help patients remember what was said. For Eric Boose, MD, associate chief medical information officer at the health system, the appeal is obvious.

“We were looking at ways to be able to bring the joy of medicine back in the sense of not being on the computer the whole time while you’re seeing a patient,” he told Becker’s. “Patients, meanwhile, benefit from having personalized patient instructions and a nice summary of what was talked about during the visit and what their next task would be.”

Those summaries never are issued without a physician’s review. According to Dr. Boose, that oversight helps ensure accuracy, while patients value the clarity of the written instructions and the ability to revisit them after the visit.

“Maybe there’s been some heavy information that day that’s hard for them to remember,” Dr. Boose said. “Now they have it all written out and they really, really appreciate that.”

Clinicians have welcomed the change. Dr. Boose said many described less after-hours charting and a stronger ability to focus on the patient. For Cleveland Clinic, that combination — easing burnout while improving patient comprehension — is what convinced leaders the technology was worth adopting more broadly.

Cleveland Clinic is one of a handful of health systems experimenting with AI-generated patient communication. At New York City-based NYU Langone Health, researchers have been testing whether large language models can simplify discharge instructions so they are easier to read.

“We give patients access to our notes as physicians, but how can you expect a patient to really understand something that is truly a physician-to-physician communication?” Jonah Feldman, MD, the system’s medical director of transformation and informatics, told Becker’s. “Large language models could have a really great role in simplifying our communication so that patients can understand.”

The health system has put the idea to the test in a randomized controlled trial. Drafts are produced by AI, then routed back into the EHR for review.

“Nothing we produce goes directly to the patient,” Paul Testa, MD, chief health informatics officer, told Becker’s. “Yet I easily foresee a day that that will change. But in the current times, there’s always a clinician in the middle.”

NYU Langone also created a rubric to evaluate every patient-facing AI output: readability, understandability, patient-centeredness, accuracy and safety. Researchers say the next step is tailoring those measures to the individual. Instead of offering the same simplified text to everyone, the team is exploring whether AI could adjust instructions to match a patient’s literacy level or preference.

“Some patients want it very simple, some patients want more detail,” Dr. Feldman said. “We think these tools can actually help us get to that kind of personalization.”

Weill Cornell Medicine, based in New York City, has taken a different path, starting with documentation that patients never see: the handoff notes that emergency medicine physicians write for inpatient teams. In a recent study, the hospital tested AI-generated versions of those notes against physician-written ones, using a World Health Organization-inspired framework for safety.

“Our study found that LLM-generated notes were actually more detailed and closer to the original emergency department documents than the manual notes,” Rahul Sharma, MD, chief of emergency medicine at Weill Cornell Medicine and NewYork-Presbyterian Hospital, told Becker’s. “But they did have slightly more errors, although without any life-threatening patient safety risk.”

For Dr. Sharma, the lesson was not about replacing doctors but reimagining their role.

“I see AI-generated notes as the draft, not a finished product,” he said. “Physicians stay in the loop, reviewing and editing because oversight isn’t just a checkpoint, it’s a partnership.”

Though patients have not yet received AI-generated documents, Dr. Sharma said he believes that day will come.

“For patient-facing documents, you need to be even more careful with language and readability,” he said. “Oversight might involve both physicians and perhaps even patient communication specialists.”

Across all three institutions, the experiments point in the same direction: AI may draft, but humans remain responsible.

“The real opportunity lies in making documentation smarter, not just faster,” Dr. Sharma said. “The future isn’t just about automating notes, it’s about elevating them into strategic assets for better medicine.”

The post How 3 systems use AI notes to help patients better understand their care appeared first on Becker’s Hospital Review | Healthcare News & Analysis.