A new report in JAMA takes a hard look at the wave of AI health tools — and the answer is more complicated than you might hope.

If you’ve ever Googled a symptom at midnight, you already know the appeal of getting instant, personalized health information. Now, AI chatbots are promising to do that better than a search engine ever could. In January, OpenAI launched ChatGPT Health, a tool that lets users upload their lab results, doctor’s notes, and Apple Watch data to get tailored health guidance. It’s just one of a growing wave of AI tools aimed squarely at patients.

But a new report published March 6, 2026 in JAMA by Rita Rubin asks the honest question: are these tools actually ready for us?

1. The Promise Is Real

About 1 in 4 of ChatGPT’s 800 million weekly users are already seeking health information through the platform. And it’s not hard to understand why — getting a doctor’s appointment is harder than ever, medical paperwork is dense and confusing, and people deserve to understand what’s happening to their own bodies.

Cardiologist and AI policy expert Dr. Haider Warraich puts it well: these tools should be a “clinician extender,” not a clinician replacer. Think of them as a knowledgeable friend who helps you prep for your appointment, decode your discharge paperwork, or understand what a diagnosis actually means — not someone who diagnoses you themselves.

Some focused AI tools are already doing this well. A urologist at USC built Pub2Post, which translates dense medical journal articles into plain language — over 6,000 people have used it. A Mayo Clinic plastic surgeon developed an AI assistant that gives patients 24/7 access to verified postoperative instructions, since many people go home from surgery while still foggy from anesthesia.

2. The Risks Are Also Real

Here’s where it gets sobering. A study previewed in Nature Medicine tested ChatGPT Health specifically on its ability to triage patient symptoms — and it struggled. It failed to correctly identify both the most serious cases (which need emergency care) and the least serious ones (which don’t). Missing an emergency could delay lifesaving treatment. Over-reacting to minor symptoms clogs up the healthcare system.

A separate Oxford Internet Institute study tested whether everyday people could use AI chatbots to identify health conditions from physician-written scenarios. When researchers fed the vignettes directly to the AI, the chatbots got the diagnosis right about 95% of the time. But when real people interacted with the same chatbots — typing in their symptoms the way you actually would — accuracy dropped to roughly one-third. The LLMs performed no better than a Google search.

The problem wasn’t just the AI. It was the communication loop: people left out important details, the AI misread what they shared, and users often ignored relevant suggestions the chatbot did make. People also tended to underestimate how serious their situation was — raising the risk that someone could feel falsely reassured and delay getting real care.

3. What About Your Privacy?

This is a big one. Neither ChatGPT Health nor Grok (Elon Musk’s AI chatbot) is HIPAA compliant — meaning they aren’t held to the same legal privacy standards as your doctor or hospital. OpenAI says it won’t use your ChatGPT Health conversations to train its models and offers options to keep chats temporary. But as one Northwestern University expert pointed out, those assurances may not hold if the company is ever sold.

For now, experts suggest caution about uploading sensitive personal health data to any of these platforms.

4. So What Should You Use AI For?

The consensus from physicians quoted in the JAMA piece is pretty clear: use AI as a low-stakes assistant, not a doctor. That means:

– Explaining medical terms you don’t understand
– Preparing questions before an appointment
– Summarizing information you’ve already been told
– Translating a confusing lab result into plain English

The Bottom Line

As radiation oncologist Dr. Danielle Bitterman advises her own patients: “Don’t take immediate action just based on what you find online. We can discuss it together.”

AI health tools are moving fast — faster, honestly, than most experts predicted. The technology has genuine potential to expand access to medical knowledge for millions of people. But right now, it works best as your prep coach, not your physician.

Source: Rubin R. “Are AI Tools Ready to Answer Patients’ Questions About Their Medical Care?” JAMA. Published online March 6, 2026. doi:10.1001/jama.2026.1122