Robbie Price, a Centra Health nurse practitioner based in Bedford County, has seen a lot of ways to take notes during his two decades of work in healthcare.
Simple handwriting. Microrecorders. Voice recognition software.
Now there’s something new: an artificial intelligence-powered “ambient digital scribe.” Price said it’s been “transformational on the clinical end.”
The so-called scribe is software on a medical provider’s phone that listens as the provider speaks with a patient. It then turns their conversation into a fully formed clinical note. Centra, based in Lynchburg, began using it in July.
The app allows a provider to focus more fully on the patient instead of taking notes. It automatically filters out non-clinical conversation topics, such as how someone’s dog or favorite baseball team is doing.
A few minutes after the visit ends, the app produces a complete clinical note, said Price, who also serves as Centra’s director of advanced practice provider operations.
“So really, it’s kind of turned the providers from more creators of the note to more editors,” he said.
Roanoke-based Carilion Clinic began integrating AI-powered documentation, such as digital scribes, into clinical workflows around 2023, with use expanding across care teams since then.
[Disclosure: Carilion is one of our donors, but donors have no say in news decisions; see our policy.]
Dr. Maruf Haider, associate chief medical information officer at Carilion, said that being more present during appointments can also help clinicians recall details more accurately when reviewing and finalizing notes later.
Patients can view notes during their appointments if they choose, Haider added.
“It’s a more thorough note that, in some cases, it picked up things that I would have not thought about putting in that was pertinent,” he said.
The benefits of such a tool are obvious. Many providers already spend what Price called “pajama time” — after hours, such as on evenings and weekends — working on notes and records. Automation helps providers finish their records earlier and get home sooner.
It’s one example of the rapidly expanding presence of artificial intelligence in healthcare, alongside technology that speeds up imaging when diagnosing strokes, helps providers respond to patient messages and improves efficiency for clinicians.
But with this expansion comes caution about data security, patient privacy and the changing roles of human professionals in tandem with increasingly advanced technology.
Dr. Maxim Topaz, who develops artificial intelligence technologies to improve patient-clinician interactions at the Columbia School of Nursing, said AI accuracy varies widely between clinical settings. Legal and ethical frameworks have not kept pace with adoption, either, which may discourage transparency about how the tools are used, Topaz said.
About 81% of physicians report using AI for tasks such as searching medical research, triaging and communicating with patients, generating discharge instructions and documenting visits, according to a 2026 survey by the American Medical Association.
Healthcare leaders largely view this shift as a positive development. AI can improve efficiency, reduce administrative burdens and help address clinician burnout, said Beth Bortz, president and CEO of the Virginia Center for Health Innovation.
At the same time, some patients — particularly those of younger generations — are more hesitant. Bortz said many ask questions about how their personal information is stored and used.
That concern often centers on the role of the third-party vendors that provide the AI services.
Many AI tools are developed and implemented by outside companies, introducing additional layers of data access, responsibility and risk, said Gurkan Akalin, executive director of the Institute for Applied Data Analytics at the University of Virginia’s College at Wise. Those vendors must maintain strong cybersecurity practices and handle patient data appropriately.
Data security, patient privacy among top concerns
Relying on third-party vendors can increase exposure to security breaches and data loss, making transparency between health systems and vendors critical for protecting patient information.
Hospitals are required to comply with HIPAA security rules, which govern how protected health information is stored and transmitted electronically. The law, enacted in 2003, requires administrative, physical and technical safeguards to protect patient privacy.
Federal guidance is designed to be flexible and technology-neutral, allowing health systems to adopt new tools while maintaining compliance, according to the U.S. Department of Health and Human Services.
At LewisGale Medical Center in Salem, support from its parent, HCA Healthcare, has shaped how new technology is implemented. The system’s electronic health records platform, built with MEDITECH Expanse, was developed specifically for HCA.
Carilion has taken a different approach, often building on relationships with existing vendors as it adopts new AI tools.
For example, Carilion integrates some of its AI tools through Epic, the company behind its MyChart patient portal.
Carilion receives hundreds of proposals from vendors every year pitching new AI technology, said Dr. Stephen Morgan, the health system’s senior vice president and chief medical information officer.
“Buyer beware in these cases,” Morgan said. “Some of them have very good data. I’ve worked with some that are very good and talk with their clients. Some do not have peer review data or data that we would really trust.”
Hospital leaders say that any new technology, whether it’s considered AI or not, needs to be vetted for security and patient privacy concerns before it’s adopted.
That might mean that the technology comes on board more slowly than some AI enthusiasts within Centra might like, said Jen Halikman, Centra’s vice president of applications information technology.
“I am proud of that process, as much as it can be viewed as sort of onerous, or a bottleneck at times. … I’m sort of the IT governance grim reaper,” she said.
For those concerned about privacy, Carilion’s Haider noted that many people already share significant personal information through everyday technology.
“The question about the general public, their concern, I think that concern probably should be across the board,” Haider said. “We all use smartphones and get on the internet and things like that.”
Centra obtains patients’ consent to use AI and renews that permission annually. Providers are instructed to tell patients when they’re being recorded by the digital scribe, and patients can revoke their consent at any time.
If a patient doesn’t consent to the use of AI, the provider can still conduct the visit the traditional way, said Price, the Centra nurse practitioner.
“There is still a fair amount of hesitation around AI, particularly when it comes to being in the room with a provider. That’s somewhat of a safer space for a lot of people,” Price said.
Not every health system is as transparent about its use of AI tools, according to Topaz.
There is no federal mandate requiring disclosure, and health system policies vary widely, he said. Patients should have the right to know, and in many cases, to opt out.
“The analogy I use: We require informed consent when a medical student participates in a patient’s care. AI is at least as consequential,” Topaz said.
AI accuracy improving, but human oversight remains critical
With any software, there can be errors. When it comes to using tools such as the digital scribe in a healthcare space, it’s up to the provider to catch those errors, Price said.
“The provider, at the end of the day, is still responsible for the note they signed because their name was on it,” he said. “But, you know, if you take having to create a note from scratch versus having to go through and edit a few paragraphs, it really does decrease the burden on the provider.”
Halikman said that providers have reported few errors to Centra’s information technology department about the AI digital scribe.
As Centra’s providers use it, the AI model improves based on the data it gets from those providers.
And while AI is a useful tool, human providers still retain the responsibility of making important medical decisions, hospital leaders say.
Patient advocacy organizations are also examining how AI intersects with care, said Gwen Darien, executive vice president of advocacy and education with the Patient Advocate Foundation, a nonprofit based in Hampton that focuses on health equity issues in the U.S.
As health systems adopt more AI tools, both patients and clinicians have widely varying levels of understanding about how the technology works, Darien said.
Some of that oversight may also fall to patients. Darien said patients should feel empowered to ask providers how they use generative AI — technology that creates new content, including text — and how they verify the accuracy of AI-generated medical notes.
Darien said reducing administrative burdens for healthcare workers could ultimately improve the patient experience, but she emphasized that AI should be viewed as a tool — not a solution.
Since 2023, Centra has been using an AI-powered platform to speed up stroke diagnosis, said Mandi Zemaiduk, director of practice operations at Centra Medical Group Neurology.
In stroke care, there’s a popular mantra: “Time is brain.” It means that the longer a patient waits for treatment, the more brain tissue can be lost.
The platform cuts the time from “door to read” — from when the patient enters the hospital to when their brain scan images are read — in half by analyzing the images and sending them directly to a provider’s phone or computer.
Nonetheless, the final word on the patient’s situation — “bleed or no bleed,” or whether the patient’s brain has a hemorrhage — comes from a human radiologist.
“We still need humans to be able to look at these images and really give us the right call. It’s just only partnering with us to make the efficiency of how we treat stroke even better,” Zemaiduk said.
As these tools evolve, hospital leaders are beginning to consider a key question: How autonomous should these systems become?
Haider said that shift could change how future clinicians are trained.
“Soon, we’ll have medical students who grew up with this. And so maybe knowledge recollection is not as important as the decision-making and verifying,” Haider said.
Dr. Zachary Williams, the emergency department medical director at LewisGale, said medical educators now face a balancing act.
Young doctors need to understand how AI tools function, but they also must develop strong independent note-taking and clinical reasoning skills.
Williams said providers are already discussing broader questions about the future role of medical documentation and what matters most in a patient note.
AI tools serve a variety of functions
Another area where AI is saving time at Carilion is within its patient portal, MyChart, where patients can send questions to their care teams.
These messages have traditionally been monitored by nurses, but AI now helps triage them, Morgan said.
When a patient submits a message, the system can generate a draft response for the provider to review and edit. Morgan said the AI-generated suggestions are designed to reflect a clear and empathetic tone.
AI can conduct an initial review of incoming messages and flag them based on urgency, according to Bortz.
As in other aspects of AI-assisted healthcare, Bortz emphasized that clinicians remain involved in the process and that AI does not make final decisions.
“They [clinicians] were getting just dozens and dozens of messages through the portal and feeling like, my God, how do we keep up with this?” Bortz said. “And now they use AI to help with kind of reviewing those messages and triaging the messages.”
Hospitals are also using AI tools to search clinical trials and analyze recent research, Bortz said.
These tools can help smaller practices identify treatment options or refer patients for more specialized care.
LewisGale, which rolled out a new electronic health records system with AI capabilities in March, uses similar research tools.
As part of HCA Healthcare, one of the largest health systems in the country, LewisGale has access to large datasets that help inform care and guide patient management, said Sean Pressman, CEO of LewisGale Hospital Pulaski.
“AI has the ability to pull those important elements from all that research that’s been done, and to help make sure that we don’t miss something,” Pressman said.
What will the future bring?
Some experts say AI could eventually take on more direct interactions with patients.
Xuan Wang is an assistant professor in the Virginia Tech Department of Computer Science and faculty member of the university’s Sanghani Center for Artificial Intelligence and Data Analytics.
For several years, Wang has been working with hospitals in Washington, D.C., and Seattle to study whether AI can improve how patients are triaged, or prioritized based on their symptoms when they come to an emergency department.
For example, patients with life-threatening injuries such as gunshot wounds are given higher priority than patients with minor injuries whose care can wait.
The triage process typically begins with a conversation between the patient and a nurse. Wang believes AI can improve it, and one day, the patient might speak with the AI agent for triage, reducing workload for nurses and doctors.
“It’s very time-consuming for the nurse and sometimes they may not accurately prioritize the patient. But this is something language model agents can do well. It’s mostly about conversations — understanding how severe the case is,” she said.
For the time being, though, AI agents’ abilities are limited in this area.
“They can provide knowledge but they do not act like doctors. Doctors usually ask follow-up questions. The models are not really that doctor-like,” she said.
As the technology continues to advance, research will focus on what Wang called “multimodal” artificial intelligence — when an AI system pulls together not just text documents and conversations with patients but also images, blood tests or even DNA samples.
Different tools could analyze those different types of data and send them to an AI “brain” for final analysis, Wang said.
Wang is also researching whether AI can help reduce bias in patient care.
Well before the introduction of AI, researchers have been examining the extent of bias — conscious or unconscious — that healthcare providers display toward women, people of color and people with disabilities.
“My own hypothesis is they probably help reduce the bias because the models only see the conversation, basically. They don’t have the personal information, they don’t know who they are talking about,” Wang said.
Meanwhile, regional hospital systems are preparing to expand their use of AI technologies.
Centra is looking at AI-generated patient summaries beyond the clinical notes produced by digital scribes.
For example, a provider could ask an app for the last two years of a patient’s complete blood count test results.
Another technology focuses on putting in patient orders.
Another would work similarly to the system that Carilion uses in MyChart, summarizing messages that patients send to providers and helping providers prepare responses.
“In the clinical space, there are a lot of AI agents that we plan to roll out starting in 2026,” Halikman said.
Carilion plans to introduce similar AI agents into the nursing workflow. As the largest segment of the clinical workforce, nurses form the backbone of the hospital operations, and integrating AI into their day-to-day tasks could significantly improve efficiency, Morgan said.
AI could also help streamline scheduling, Morgan said. By analyzing a patient’s medical history and the reason for the visit, an AI agent could recommend a 15-minute appointment or allocate an hour, depending on the patient’s needs.
In the future, patients can expect to see more autonomous AI systems that act independently to identify and carry out tasks without direct human oversight. While that shift may still be some time away, Morgan said these tools could support decision-making.
“It’s really about how to improve that efficiency,” Morgan said.
