A third of Americans are asking chatbots for health advice. The real question is what they do next.

Americans have a new place to take their health worries: the chatbot window.

A sore throat at midnight. A rash that does not look serious enough for urgent care. A lab result written in medical language. A panic attack that feels easier to describe to a screen than to a person. For many people, AI now sits somewhere between Google, a nurse hotline and a friend who never sleeps.

That shift is no longer small. A KFF poll reported by People found that roughly one-third of adults surveyed had used AI chatbots for advice about physical or mental health. Among those who used AI for physical health advice, 42 percent did not follow up with a doctor or health professional. Among those who used it for mental health advice, 58 percent did not follow up with a mental health professional. (People.com)

That is the part that should make clinicians pay attention.

The issue is not that people are curious. Curiosity is normal. Patients have always searched symptoms online, asked relatives, read forums, watched videos and compared notes before seeing a doctor. AI simply makes that behavior faster, cleaner and more convincing.

The risk is that the answer sounds too good.

Why people are turning to AI first

The appeal is obvious. Chatbots are fast. They do not charge a co-pay. They do not make users wait three weeks for an appointment. They do not sound rushed. They can explain a medical term in plain language and help someone prepare better questions before a visit.

AP reported that many Americans using AI for health information are looking for quick answers, extra context before or after seeing a doctor, or help when care feels too costly or inconvenient. Some users also cite embarrassment, lack of time, prior dismissal by a provider or difficulty accessing care. (AP News)

That should not surprise anyone who has dealt with the modern health system. In many places, getting care is expensive, slow and emotionally tiring. If a chatbot responds instantly, people will use it.

The more difficult question is whether they know when to stop using it.

AI can be helpful for low-stakes explanation. It can translate medical jargon, summarize what a test might measure, help draft a message to a doctor or suggest questions to ask at an appointment. Used that way, it can make a patient more prepared.

But a chatbot is not examining the patient. It cannot feel an abdomen, listen to lungs, review a full medical history, order the right test or notice the small details that change a diagnosis. It also has no legal or ethical responsibility for what happens after the answer is given.

The danger is confidence without responsibility

A good doctor often pauses. They ask follow-up questions. They say, "I need to examine you." They order another test. They tell a patient when symptoms require urgent care.

A chatbot can sound certain even when it is wrong.

That problem is not theoretical. A physician-led red-teaming study of publicly available chatbots tested 888 responses to 222 patient-style medical questions. The study found that problematic responses varied widely across models, and unsafe responses appeared in a meaningful share of answers. The authors warned that chatbot responses had the potential to lead to serious patient harm. (arXiv)

That does not mean every AI health answer is dangerous. Many answers are sensible. Some are careful. Some tell users to seek professional care. The problem is inconsistency.

For a patient, inconsistency is hard to detect. A wrong answer can arrive in the same calm tone as a right one.

That is especially risky when someone is trying to decide whether chest pain is anxiety or something worse, whether a child's fever can wait until morning, whether a medication interaction matters, or whether a mental health crisis is serious enough to call for help.

Mental health makes the problem more delicate

Mental health is where the chatbot trend becomes most complicated.

People may turn to AI because they feel ashamed, alone, uninsured or afraid of being judged. In that situation, a responsive chatbot can feel comforting. It may help someone name emotions, organize thoughts or practice how to talk to another person.

But mental health care is not only information. It is relationship, judgment and safety.

The Washington Post reported that KFF polling found about 3 in 10 adults ages 18 to 29 had turned to AI chatbots for mental or emotional health advice in the past year, and nearly 60 percent of adults who used a chatbot for mental health did not follow up with a human professional. (The Washington Post)

That matters because a chatbot may affirm harmful thinking instead of challenging it. It may miss the seriousness of a crisis. It may respond warmly while failing to push the person toward emergency support, a clinician, a trusted family member or a crisis line.

There is also a privacy issue. People may type deeply personal information into tools without understanding where that information goes, how it may be stored, or whether it receives the same protections as a conversation with a licensed clinician.

Doctors should not just scold patients for using AI

It would be easy for doctors and health systems to tell people not to use chatbots. That will not work.

People are using them because the system leaves gaps. Appointments are hard to get. Insurance is confusing. Medical language is intimidating. A 10-minute visit may not give patients enough time to ask every question.

AI is filling a demand that health care has not met.

The better response is practical. Doctors should assume some patients are already using AI and ask about it without judgment. A simple question could help: "Did you look this up or ask an AI tool before coming in?"

That opens the door to correction. It also gives clinicians a chance to show patients how to use AI more safely.

A chatbot can help prepare for care. It should not replace care.

A safer way to use chatbots for health questions

For ordinary users, the rule should be simple: use AI as a note-taking and question-preparation tool, not as a doctor.

A safer prompt sounds like this:

"What questions should I ask my doctor about these symptoms?"

Or:

"Can you explain this lab test in plain English, and what should I confirm with my clinician?"

A riskier prompt sounds like this:

"Tell me what disease I have."

Or:

"Should I stop this medication?"

Or:

"Do I need emergency care?"

Those decisions belong with qualified professionals, especially when symptoms are severe, sudden, worsening or hard to explain.

AP reported that only about one-third of adults who had recently used AI for health information said they trusted the accuracy of AI-generated health advice, while most adults still relied on health professionals as their main source of health information. (AP News)

That skepticism is healthy. AI can support a patient. It should not become the final authority.

The future will need rules, not just warnings

The use of chatbots for health advice is not going away. The tools will become more common, more polished and more integrated into websites, apps and health systems.

That means the question is no longer whether people will ask AI about health. They already do.

The real questions are harder. Which tools are safe enough? Who tests them? What should they be allowed to say? What happens when they are wrong? How should privacy be protected? When should a chatbot be required to tell a user to seek urgent care?

Until those answers are clearer, patients should treat AI health advice as a starting point, not a diagnosis.

A chatbot can explain. It can summarize. It can help someone feel less lost.

But it cannot replace the person who takes responsibility for care.

This article is for general information only. It is not medical advice, diagnosis or treatment. For urgent symptoms, medication decisions, mental health crises, pregnancy concerns, severe pain or worsening conditions, contact a qualified health professional or emergency service.