Can AI Give Better Health Advice Than Your Doctor?
By Jon Scaccia
0 views

Can AI Give Better Health Advice Than Your Doctor?

New research suggests artificial intelligence can sometimes sound more empathetic than physicians, explain complex conditions more clearly, and even outperform doctors on certain medical tests. But does that make it a better source of health advice? The answer is more complicated—and more interesting—than you might think.

Imagine waking up at 2 a.m. with a strange symptom. Maybe your heart is racing. Maybe you have a persistent headache, or you just received blood test results you don’t understand.

Your doctor’s office is closed. An internet search gives you everything from harmless explanations to frightening possibilities.

So you open an AI chatbot.

Within seconds, you receive a thoughtful response. It explains what might be happening, acknowledges your concerns, describes possible causes, and suggests questions to ask your doctor. It doesn’t seem rushed. It doesn’t interrupt. You can ask the same question five different ways without feeling embarrassed.

You might even leave thinking the chatbot understands you better than your physician.

But here’s the uncomfortable question: What if the AI sounds more caring, yet gives you worse medical advice?

That possibility sits at the center of an emerging body of research on AI and healthcare. And the findings challenge some of our assumptions about what makes medical advice trustworthy.

Free weekly newsletter

Get the latest public health insights—
every Thursday morning.

We scan 70+ journals so you do not have to.
One email. Zero jargon. Unsubscribe anytime.

No spam. 1-click opt-out. Privacy-first.

When AI Sounds More Empathetic Than a Doctor

One of the more surprising findings comes from a study published online in December 2025 in the Journal of General Internal Medicine.

Researchers Mollie Ruben, Danielle Blanch-Hartigan, and Judith Hall wanted to understand how people judge medical responses written by physicians compared with those generated by ChatGPT.

The researchers examined responses to questions originally posted on Reddit’s AskDocs forum. They recruited 1,454 participants in the United States and randomly presented them with medical answers written either by a physician or by AI.

But the experiment included an important twist.

Sometimes participants were correctly told who wrote the response. Other times, they were told an AI response came from a physician—or that a physician’s response came from AI.

This allowed researchers to separate two things: the actual quality of the communication and people’s expectations about its source.

Ratings reported in Ruben and colleagues’ study, on a five-point scale. The comparison involved written answers to online patient questions, not actual medical consultations.

Participants rated AI-generated answers as significantly more empathetic than physician-written answers. The AI responses also tended to include more reassurance, validation of patients’ feelings, nonjudgmental language, and structured explanations.

In other words, AI was particularly good at several behaviors we associate with a caring healthcare professional.

But there was another finding.

People rated responses more favorably when they believed a physician had written them—even when the actual author was AI. That tells us something important about trust. We don’t judge medical advice solely by its content. We also judge it by who we believe is providing it.

A doctor’s professional identity still carries considerable weight.

However, an important limitation remains. Participants evaluated written responses to online questions, not medical care itself. The study did not test whether AI made better diagnoses, produced safer treatments, or improved health outcomes.

Still, its implications are significant.

If patients consistently find AI-generated messages clearer and more compassionate, healthcare professionals may have something to learn from the technology.

The goal doesn’t have to be replacing physicians. It could be helping them communicate better.

But Patients Still Trust Human Doctors More

If AI can produce more empathetic answers, why don’t people simply prefer it?

A second study, published in the Journal of Medical Internet Research in 2025, provides an interesting explanation.

Researchers Weiqi Guo and Yang Chen conducted an experiment with 249 participants. They examined how people evaluated three medical consultation arrangements: advice from AI alone, advice from AI with a human physician involved, and advice from a human physician.

They also explored whether explanations mattered and whether patients responded differently when the medical issue involved stigma.

The researchers distinguished three aspects of the consultation experience: confidence in the diagnosis and treatment, relationship quality, and emotional responses to the consultation.

The results revealed an intriguing contrast.

Participants trusted diagnoses attributed to human physicians more than those attributed to AI or human-involved AI systems.

Yet the study found no statistically significant differences between human and AI consultations in relational and emotional evaluations.

This means participants didn’t necessarily find AI less supportive or emotionally satisfying. They were simply more confident in the physician’s diagnostic authority.

Even more interestingly, explaining the diagnosis improved evaluations across all three consultation types.

This suggests that patients want more than an answer. They want to know why an answer makes sense.

Consider the difference between these two responses:

“Your symptoms suggest a migraine.”

And:

“Your symptoms could be consistent with a migraine because of the headache pattern and sensitivity to light. However, other conditions can cause similar symptoms, and certain warning signs would require urgent medical assessment.”

The second response offers context, acknowledges uncertainty, and gives the patient a framework for understanding the recommendation.

That is valuable whether the explanation comes from a physician or a computer.

The study’s simulated consultation design does not establish which source actually produced more accurate diagnoses. But it reinforces a central lesson: good healthcare communication involves explanation, not merely information.

The Bigger Question: Can AI Actually Make Better Diagnoses?

So far, the evidence suggests AI can communicate effectively and that patients value the explanations it provides.

But communication is only part of healthcare.

A chatbot can write a wonderfully reassuring paragraph about chest pain. That doesn’t mean it can reliably determine whether someone is experiencing a heart attack.

To understand whether AI provides better health advice, we need to distinguish between three different abilities:

  • Medical knowledge: Can the system identify diseases, interpret medical information, and answer clinical questions?
  • Clinical reasoning: Can it gather the right information, recognize uncertainty, and decide what to do next?
  • Real-world patient care: Can it help an actual person make a safer, more informed decision about their health?

These abilities are related, but they are not interchangeable.

When people actually use AI, performance can fall apart

A randomized study published in Nature Medicine examined what happened when ordinary people used large language models to work through medical scenarios.

The researchers focused on something laboratory tests often miss: the interaction between a person and the AI.

A chatbot might recognize a condition when researchers provide a complete, carefully written medical case. But real people rarely describe their symptoms that way.

They leave out details. They use everyday language. They may not recognize which symptoms are important. And they sometimes misunderstand the responses they receive.

The findings were sobering.

Participants using AI identified relevant medical conditions in fewer than 34.5% of cases and correctly determined the appropriate level of care in fewer than 44.2%.

Reported upper bounds, not exact success rates. Participants using AI did not perform better than the control group using more traditional information sources.

This does not mean AI systems were incapable of recognizing the medical conditions themselves. The researchers found that models that performed well on their own did not necessarily help people reach accurate conclusions.

The problem was partly in how they interacted.

A knowledgeable AI and a confused patient don’t automatically produce a well-informed patient.

That is an important lesson for public health. Providing access to medical information is not the same as helping people use it safely.

Some AI-generated health advice is genuinely unsafe

Another 2026 study in npj Digital Medicine examined 888 responses from four publicly available AI chatbots to 222 patient questions.

The researchers used physicians to evaluate the clinical safety of the advice.

Depending on the chatbot, between 21.6% and 43.2% of responses were classified as problematic. The proportion judged unsafe ranged from 5% to 13%.

Those numbers deserve attention.

They do not mean that every AI health interaction carries those exact risks, or that today’s models perform identically to the older systems tested. But they show why fluent answers aren’t a reliable indicator of safe advice.

AI can make mistakes that patients may not recognize, including recommending inappropriate actions, omitting important warnings, or failing to appreciate the seriousness of a medical situation.

The most concerning failures may not look obviously wrong. They may appear reasonable, confident, and helpful.

But AI Is Improving—and Some New Results Are Remarkable

It would be a mistake to look at those safety concerns and conclude that AI has little future in medicine.

This field is changing extraordinarily quickly.

In October 2026, researchers at Microsoft reported results from an AI system designed to approach complicated diagnoses through a sequence of questions, examinations, and tests.

Rather than receiving all the information upfront, the system worked through medical problems step by step, updating its conclusions as new information became available.

In an evaluation of more than 300 unusually complex medical cases, the system was reported to be more than twice as accurate as general physicians using reference materials and the internet.

That is an impressive result, but it also requires context.

The evaluation involved challenging diagnostic cases rather than routine, unsupervised medical care delivered to patients at home. The comparison does not establish that a general-purpose chatbot is twice as good as a person’s own physician or that it can independently manage treatment.

Indeed, the researchers planned to test the approach in emergency department settings, where real-world conditions introduce complications simulations cannot fully capture.

Nevertheless, the findings illustrate how rapidly the frontier is moving.

An AI system that was unreliable for a particular task in 2024 might perform very differently in 2026. Newer systems can also use tools, retrieve external information, and organize their reasoning in ways earlier chatbots could not.

That makes simplistic statements about AI’s medical abilities increasingly difficult to defend.

The meaningful question is not whether AI is good at medicine.

It’s which AI system, performing which medical task, with what safeguards, and supported by what evidence?

The Most Promising Approach May Be AI Working With Doctors

Another possibility deserves more attention than the question of replacing physicians.

What if AI makes healthcare professionals better at their jobs?

A 2026 randomized trial published in Nature Health examined whether physicians in Pakistan could improve their diagnostic reasoning with access to an AI chatbot.

The study included 60 licensed physicians, 58 of whom completed the trial. Those using AI alongside conventional resources achieved an average diagnostic reasoning score of 71.4%, compared with 42.6% among physicians using conventional resources alone.

This was a substantial improvement in a controlled evaluation.

It does not prove that AI assistance improves every clinical encounter or health outcome, particularly across different settings and levels of training. But it illustrates AI’s potential as a professional support tool.

Consider what this could mean in practice.

A physician could use AI to generate a list of diagnoses worth considering, identify potential medication interactions, summarize complex research, or translate technical explanations into language a patient understands.

The physician would still be responsible for evaluating those suggestions, interpreting clinical findings, and making decisions with the patient.

That arrangement potentially combines AI’s ability to process information quickly with a clinician’s physical examination, medical judgment, and responsibility for patient care.

There are still risks. Physicians can become overly reliant on automated suggestions, and incorrect recommendations may influence their decisions. Human oversight must therefore be meaningful, not just a signature attached to an AI-generated answer.

But research increasingly points to a productive question: How should healthcare systems design collaboration between people and machines?

What Happens When AI Learns How You Want to Be Treated?

Medical decisions are not always just technical questions.

Imagine two patients who receive the same cancer diagnosis.

One wants a clear recommendation from their physician. They find complex medical choices overwhelming and would prefer an expert to explain the most appropriate course of treatment.

The other wants to understand every option, compare the risks and benefits, and discuss how treatment might affect work, family life, and long-term quality of life.

Neither preference is inherently wrong.

But the same communication approach might not serve both patients equally well.

This is the premise of a paper by Hazem Zohny and colleagues in the Journal of Medical Ethics, published online in 2025 and appearing in the journal’s 2026 volume.

The researchers explored whether GPT-4 could emulate four established models of physician-patient communication.

These models ranged from a more directive approach, where the clinician recommends a course of action, to approaches that emphasize information sharing, exploring patient values, and collaborative decision-making.

Their proof-of-concept demonstration showed that AI could generate conversations reflecting these different styles. AI might eventually help patients prepare for medical decisions by adjusting explanations to their communication preferences.

Someone who wants straightforward facts could receive a concise explanation. Someone who wants to discuss uncertainty could explore competing options. Someone who struggles with medical terminology could ask for simpler language.

This flexibility could be especially valuable for patients with limited health literacy or those who feel intimidated during medical appointments.

But the researchers also highlighted an ethical danger. An AI that can adapt to your preferences may also become unusually good at persuading you. A chatbot might reinforce mistaken beliefs, subtly steer decisions, or present one treatment option more favorably than another. Because its responses can feel personal and supportive, that influence may be difficult to recognize.

The paper was a conceptual and proof-of-concept exploration, not a clinical trial demonstrating better patient outcomes.

Even so, it raises a fundamental question about the future of medical AI: Should a health chatbot simply provide information, or should it help shape our decisions?

And if it does shape our decisions, whose values should guide it?

How Should You Actually Use AI for Health Questions?

Given these findings, it’s understandable that people might feel conflicted.

AI can explain medical information quickly. It can offer reassuring language, help patients prepare questions, and make complex subjects easier to understand.

But it can also confidently provide incorrect or incomplete information.

The safest approach is to treat AI as a tool to become a more informed patient, not as an independent medical authority.

Here are five ways to use it more effectively.

  • Ask AI to explain, not diagnose. Instead of asking, “What disease do I have?” consider asking, “What are some possible explanations for these symptoms, what information would help distinguish them, and what warning signs require medical attention?”
  • Use AI to prepare for medical appointments. Ask it to organize your questions, explain unfamiliar terminology, or identify topics worth discussing with your clinician.
  • Ask about uncertainty. A useful question is, “What information is missing, and what would change this assessment?” This can reveal limitations that a confident first answer might hide.
  • Verify important claims. Check medical information against authoritative clinical sources and discuss treatment recommendations with a qualified healthcare professional. AI-generated references can be inaccurate or fabricated.
  • Protect your privacy. Avoid sharing identifiable medical records, insurance information, or sensitive personal details with general-purpose chatbots without understanding how that information will be stored, used, and protected.

And critically, don’t use a chatbot to decide whether to delay emergency care. Severe or rapidly worsening symptoms, including possible signs of a heart attack or stroke, require immediate medical assessment.

Why This Matters for Public Health

The AI healthcare debate extends far beyond individual patients and physicians.

For public health, one of the greatest potential benefits is broader access to understandable health information.

Millions of people face barriers to healthcare, including cost, transportation, provider shortages, language differences, and limited appointment availability.

AI could help reduce some of those information barriers. A well-designed system might help people understand preventive health recommendations, prepare for care, navigate health services, or make sense of medical instructions.

But there is a serious equity question.

People with greater health literacy may be better equipped to recognize when an AI response is questionable. They may know how to verify information or have a trusted physician available for clarification.

People with fewer resources or limited access to healthcare may be more likely to rely on AI without that support.

If AI systems provide incorrect advice, the consequences could fall disproportionately on people who already experience barriers to medical care.

That is why evaluation matters.

Health systems should not assess medical AI solely through impressive demonstrations, exam scores, or user satisfaction ratings. They need to examine real-world outcomes, clinical safety, accessibility, and whether the technology works equally well across different populations.

They also need to distinguish systems designed and validated for specific healthcare purposes from general-purpose chatbots anyone can access with an internet connection.

Those are not necessarily interchangeable technologies.

So, Can AI Give Better Health Advice Than Your Doctor?

The answer is: sometimes, on certain tasks, but the evidence does not support treating AI as a generally superior replacement for a physician.

AI can generate explanations that people find more empathetic. It can process large amounts of medical information. And increasingly sophisticated systems have demonstrated remarkable diagnostic capabilities in controlled settings.

Yet important gaps remain between answering medical questions, understanding a real patient’s circumstances, and delivering safe, effective care.

Perhaps the most useful lesson from this research is that we have been asking the wrong question.

The future of healthcare may not come down to choosing between a human doctor and an AI system.

It may depend on developing healthcare in which physicians have better tools, patients receive clearer information, and medical decisions reflect both scientific evidence and individual needs.

The research will keep evolving, probably faster than many healthcare institutions can adapt.

For now, the most promising vision is not a chatbot replacing your doctor.

It’s a healthcare system where better technology helps your doctor provide better care—and helps you become a more informed participant in your own health.

Build Your Public Health Issue

Tell us a little about your work, and we’ll build a personalized issue with five things worth knowing.

We’ll use your answers to personalize your sample issue. No spam.

Research Sources and Further Reading

Discussion

No comments yet

Share your thoughts and engage with the community

No comments yet

Be the first to share your thoughts!

Join the conversation

Sign in to share your thoughts and engage with the community.

New here? Create an account to get started