Do Patients Know When AI Is Assisting in Health Care?

Artificial intelligence may already have participated in your health care without introducing itself.

It might have helped a radiologist flag an abnormal scan, listened during an office visit to draft a clinical note, suggested a reply to your patient portal message, identified a possible medication risk, or helped an insurer process a request for coverage. In many cases, the patient sees only the final human-facing result. The algorithm remains backstage, quietly moving props while the doctor gets top billing.

So, do patients know when AI is assisting in health care? Sometimesbut not consistently. In the United States, disclosure depends on what the AI does, where it is used, whether a human reviews its output, the policies of the health system, and, increasingly, state law. There is no single nationwide rule requiring a pop-up that says, “Hello, an algorithm helped with this.”

That matters because AI is becoming a routine part of medical work. The American Medical Association reported in 2026 that more than 80% of surveyed physicians were using AI in some aspect of their professional work. At the same time, the AMA emphasizes that AI use in health care should be transparent to physicians and patients.

Where Is AI Already Assisting in Health Care?

The phrase AI in health care can sound as though a robot is about to roll into the examination room wearing a white coat and asking where it hurts. The reality is less cinematic and far more widespread.

Medical imaging and diagnostic support

AI can analyze medical images, highlight suspicious regions, prioritize urgent cases, estimate risks, and support specialists reviewing X-rays, mammograms, CT scans, MRIs, and other studies. The U.S. Food and Drug Administration maintains a public list of AI-enabled medical devices authorized for marketing, although the agency notes that its list is not necessarily comprehensive.

A patient may therefore receive a radiology report without realizing that software helped the radiologist identify, measure, prioritize, or evaluate a finding. The clinician remains responsible for the medical interpretation, but AI may have influenced the workflow before the report ever appeared in the patient portal.

Clinical decision support

Hospitals use predictive tools to estimate risks such as deterioration, readmission, sepsis, medication problems, or other complications. AI can also help organize complex medical information so clinicians can make decisions more efficiently.

Federal health IT rules have introduced transparency requirements for certain predictive algorithms included in certified health technology. The federal HTI-1 rule was designed to make information about these tools more available to the organizations and professionals using them. However, algorithm transparency within a health IT system is not the same thing as automatically notifying every patient each time a model contributes to care.

AI medical scribes

One of the most visible new uses of AI is ambient documentation. With permission under a health system’s chosen consent process, a tool can capture a clinical conversation and use generative AI to create a draft note for the clinician to review.

Research on ambient documentation shows why disclosure is not as simple as adding one sentence to a form. In one study involving clinicians and patients, consent commonly occurred through a verbal conversation before the visit. Patient comfort varied according to trust in the clinician, understanding of the technology, and perceptions of its risks and benefits. About three-quarters of the participating patients reported being comfortable or very comfortable with the physician using the technology.

In this setting, patients are more likely to know AI is present because a microphone is effectively entering the conversation. But even then, the quality of the explanation can vary. “This helps me write notes” is not quite the same as explaining what is recorded, where the data goes, what the AI produces, who checks the output, and whether the patient can decline.

Patient portal messages

Generative AI can draft responses to messages about symptoms, medication questions, follow-up instructions, and other concerns. A clinician may then review, edit, and send the message.

This creates a surprisingly philosophical question for something that arrives next to a pharmacy refill request: Who wrote the message?

Recent patient research found support for AI-assisted portal communication when it improved efficiency, but patients strongly preferred clinician review and wanted disclosure of AI involvement to help preserve trust.

Administrative and insurance decisions

AI and automated systems can also operate far from the examination room. They may help with scheduling, billing, coding, claims processing, prior authorization, fraud detection, and coverage review.

This may be the least visible form of health care AI to patients, even though it can affect how quickly care is approved or whether a coverage decision needs to be challenged. KFF has documented the expanding role of automation and AI throughout prior authorization and claims review, along with an evolving mix of federal and state consumer protections.

Why Patients Often Do Not Know AI Was Involved

The biggest reason is simple: AI is usually a tool inside a workflow, not a separate person standing beside the doctor.

A clinician does not necessarily explain every technology used during a visit. Patients are rarely given a tour of the electronic health record software, laboratory information system, imaging workstation, scheduling algorithm, or medication interaction database. As AI becomes embedded in those same systems, organizations must decide when its involvement becomes important enough to disclose.

That boundary is still being debated.

An AI tool that automatically fixes punctuation in a draft clinical note poses a very different level of risk from a system that recommends a cancer diagnosis or influences whether an insurer authorizes surgery. Treating both uses as ethically identical would make little sense. On the other hand, saying that no disclosure is needed because “the doctor made the final decision” may also be inadequate when an AI recommendation substantially shaped that decision.

The emerging practical approach is often risk-based transparency: the greater the AI’s influence on a patient’s diagnosis, treatment, communication, privacy, or access to care, the stronger the argument for meaningful disclosure.

Does U.S. Law Require Doctors to Disclose AI Use?

There is no universal federal requirement that every patient be notified whenever AI contributes in any way to health care. Instead, the regulatory landscape is a patchwork involving medical device regulation, health information rules, health IT standards, professional ethics, state laws, and institutional policies.

FDA oversight does not automatically equal bedside disclosure

The FDA evaluates many AI-enabled medical devices and provides public information intended to help clinicians and patients identify AI-enabled technologies. The agency and international regulatory partners have also promoted transparency principles emphasizing clear, essential information and the performance of the human-AI team.

Still, an FDA-authorized AI-enabled device does not mean that every patient will receive a separate notification every time that device contributes to care.

HIPAA is mainly about health information privacy and security

The federal HIPAA framework establishes protections for identifiable health information and electronic protected health information. Its core purpose is privacy and security, not the creation of a broad nationwide rule requiring an “AI was used” notice for every clinical interaction. Health organizations nevertheless must consider how patient data is handled when AI vendors or systems process protected information.

Some states are moving toward specific disclosure rules

California provides a useful example of how targeted AI transparency laws can work. Its law on generative AI in certain health care communications requires covered health facilities and practices to disclose when generative AI produces patient communications involving clinical information and to provide a way to contact a human. However, the requirement does not apply when a licensed or certified health professional reads and reviews the communication.

That exception illustrates the complexity of the issue. A message could be substantially drafted by AI, reviewed by a clinician, and sent without the same mandatory disclosure that would apply if it were not reviewed. From the patient’s perspective, the words may look nearly identical.

What Do Patients Actually Want to Know?

Patients generally do not need a 47-page technical manual explaining neural-network architecture while sitting in a paper gown. They usually want answers to practical questions:

• Was AI used in my care?
• What did it actually do?
• Did a qualified human review the result?
• Could the AI’s recommendation affect my diagnosis or treatment?
• What information about me was processed?
• Can I ask for human review or decline the tool?
• Who is responsible if the AI is wrong?

Research suggests that these details can affect trust. A 2026 study involving 3,000 U.S. adults found that people were more likely to trust and choose AI-assisted medical scenarios when the AI performed well, a clinician remained involved, representative data had been used, and credible oversight or certification was present. The presence of a clinician was particularly important.

In other words, many patients are not demanding that hospitals unplug every algorithm. They are asking for accountability. “The computer said so” is not a reassuring medical philosophy.

Patient Attitudes Toward AI Are More Complicated Than Simple Fear

Americans have expressed caution about medical AI for years. A large Pew Research Center survey found that 60% of U.S. adults would have felt uncomfortable with a health care provider relying on AI for activities such as diagnosis and treatment recommendations. Fifty-seven percent thought greater AI use could worsen the patient-provider relationship.

Yet public behavior is evolving. By 2025, according to a Pew survey published in 2026, 22% of U.S. adults said they obtained health information from AI chatbots at least sometimes. Users tended to value the convenience and understandability of chatbot responses while remaining much less confident about accuracy.

This is not really a contradiction. People can be curious about AI and cautious about it at the same time. Most of us have managed this emotional balancing act with plenty of technology. We happily use navigation apps while also yelling, “That cannot possibly be the fastest route.”

The same pattern may emerge in medicine. Patients may welcome faster responses, earlier detection, less paperwork, and more time with clinicians while still wanting to know when an algorithm influenced something important.

Why Transparency Can Improve Health Care AI

Disclosure is sometimes treated as a burdena warning label that may frighten patients. Done badly, it can. A vague notice saying, “Artificial intelligence may be used somewhere in our operations,” tells patients almost nothing except that lawyers were involved.

Useful transparency is specific and proportionate.

A better explanation might say that an AI tool is listening to create a draft visit note, that the clinician will review and correct the note, that the tool is not making the diagnosis, and that the patient can ask questions or decline according to the organization’s policy.

For a diagnostic system, patients may need to understand that AI assisted the clinician rather than independently replacing medical judgment. For an automated portal message, they may want to know whether a licensed professional reviewed it. For a coverage decision, they may want access to meaningful human reconsideration.

Transparency can also improve the behavior of health systems. When organizations know they may need to explain an AI tool to patients, they have a stronger reason to understand how it works, what evidence supports it, how its performance is monitored, and who is accountable. That is healthy pressure. Medicine has enough mysterious boxes already; several of them are called “billing statements.”

What Patients Can Ask When They Suspect AI Is Being Used

Patients do not need to become AI auditors. A few direct questions can clarify a great deal:

“Was any AI system used to help produce this result, message, note, or recommendation?”

Then ask what role the system played and whether a clinician independently reviewed the output. When the AI directly affects an important medical decision, patients can also ask what happens when the clinician disagrees with the algorithm and whether an alternative form of review is available.

For tools that record or process conversations, privacy questions are reasonable: What data is captured? Is it retained? Who can access it? Is a third-party vendor involved?

Not every staff member will immediately know the answers. That itself can reveal whether an organization has mature AI governance. A health system should not need to conduct an archaeological expedition to discover which algorithm is helping make clinical decisions.

What Better AI Disclosure Could Look Like

The future of AI transparency in medicine will probably not be a constant stream of frightening pop-ups. Patients would quickly develop “disclosure fatigue,” the health care equivalent of clicking “accept all cookies” while understanding none of them.

A better system would match disclosure to risk and context. Low-impact administrative automation might require little or no individual notification. Patient-facing generated content could carry a clear label. Recording and ambient documentation could involve a meaningful explanation and an easy choice. High-impact diagnostic or treatment tools could require stronger disclosure, clinician discussion, and pathways for human review.

Standardized “AI facts” could eventually help patients and professionals understand what a system does, what data supports it, how well it performs, who monitors it, and what role humans retain. The goal should not be to make every patient an engineer. The goal should be to prevent invisible automation from becoming invisible authority.

Patient Experiences: What AI Assistance Can Feel Like in Real Life

The following are composite, illustrative experiences based on common health care AI workflows and published patient concerns. They are not presented as individual case reports.

Experience 1: The doctor who suddenly makes more eye contact

A patient arrives for a follow-up visit expecting the familiar routine: the doctor asks a question, turns toward the computer, types for 30 seconds, apologizes, asks another question, and returns to the keyboard.

This time, the doctor barely types. A small device or application is helping create a draft note from the conversation. The visit feels warmer and less interrupted. The patient may appreciate the change immediately.

But comfort can change if the patient discovers afterward that an AI system processed the conversation without a clear explanation. The same technology that improved human connection can undermine trust if its presence feels hidden. The lesson is almost comically simple: people often accept useful technology more readily when nobody acts as though it is a secret.

Experience 2: The scan that had an invisible second reader

A patient receives a message saying a scan showed no urgent abnormality. The report was signed by a radiologist, and that is the only name the patient sees.

Behind the scenes, however, AI may have helped prioritize the image, highlight a suspicious region, calculate a measurement, or flag a possible emergency. The radiologist reviewed the case and made the final interpretation.

Would the patient want to know? Some would say yes, especially if the AI directly influenced the diagnostic process. Others may care only that a qualified specialist remained responsible. This is why a one-size-fits-all disclosure policy is difficult. The significance of AI involvement depends on what it actually did.

Experience 3: The portal message that sounds unusually polished

A patient sends a long message about a new medication side effect. Twenty minutes later, a clear and empathetic response arrives with organized instructions and perfect punctuationalways a slightly suspicious event on the internet.

An AI system may have drafted the response before a nurse or doctor reviewed and edited it. The patient benefits from faster communication, and the clinician saves time.

But the patient may still want to know whether the named clinician personally wrote every word, merely approved the message, or never saw it at all. Those are different levels of human involvement, and disclosure can prevent the efficiency of AI from being mistaken for personal authorship.

Experience 4: The unexplained insurance obstacle

A patient is told that a treatment needs additional review. The notice contains formal language but offers little insight into whether an automated system screened the request, ranked it, compared it with criteria, or influenced the decision.

For patients, this can be the most frustrating form of invisible AI because it affects access rather than convenience. A person may reasonably care less about whether AI scheduled an appointment and much more about whether an algorithm contributed to delaying a procedure.

Here, meaningful transparency should include more than the phrase “technology was used.” Patients need understandable reasons, applicable criteria, and access to genuine human review when a consequential decision is disputed.

Experience 5: The patient brings AI into the examination room

AI transparency is not only a question for hospitals. Patients increasingly use chatbots themselves. A person may arrive with a list of possible diagnoses generated after a midnight conversation with an AI assistant.

The doctor now faces a new version of an old problem. Years ago, patients arrived with printed search results. Today, they may arrive with a confident five-paragraph AI explanation that has already diagnosed three rare diseases and recommended worrying immediately.

The best response is not necessarily ridicule or blind acceptance. Clinicians can ask what tool the patient used, what information was entered, and what advice was given. Patients, in turn, should understand that a general-purpose chatbot may not have access to their complete history, examination findings, laboratory results, or the clinical accountability of a licensed professional.

This shared transparency may become one of the most important features of future medical visits: doctors explaining when they use AI, and patients feeling comfortable explaining when they use it too.

Conclusion: Patients Deserve More Than Invisible Intelligence

AI is no longer a futuristic guest waiting outside the hospital door. It is already helping analyze information, draft documentation, support imaging, answer messages, manage workflows, and influence administrative decisions.

Yet patients do not always know when it is involved.

The best path forward is neither mandatory panic nor silent automation. It is meaningful, risk-based transparency. Patients should receive clearer information when AI directly shapes important clinical communication, diagnosis, treatment, privacy, or access to care. They should also know when a qualified human remains responsible for checking the result.

Health care has always relied on tools. AI is another toolbut one with an unusual ability to generate language, predictions, and recommendations that can look remarkably human. That makes transparency more important, not less.

Patients do not need a computer science lecture at every appointment. They do deserve an honest answer to a reasonable question: “Whoor whathelped make this decision about my health?”

Note: This article provides general educational information and reflects the evolving U.S. health care AI landscape. It is not medical or legal advice, and specific disclosure requirements may vary by jurisdiction, technology, and clinical setting.