My patients don’t wait for me anymore. That’s not a complaint. It’s a clinical reality I had to sit with before I could do anything useful about it.

Patients started arriving at appointments having already run their symptoms through an AI chatbot. Sometimes they’d decided not to come in at all. Sometimes they were right. Sometimes they weren’t. I’ve spent the last several years building an AI-assisted triage and treat virtual urgent care platform, and the work forced me to face something most colleagues haven’t confronted directly: Patients aren’t waiting for medicine to endorse AI self-care. They’re already doing it. The question isn’t whether they’ll use AI for health decisions. They will. The question is whether the tools shaping those decisions were designed by people who understand what can go wrong.

What patients are actually doing

The scale of adoption is worth pausing on. In November 2025, The New York Times published a deep look at this shift, interviewing dozens of Americans who described turning to AI chatbots to compensate for a health care system they couldn’t access or afford. A self-employed woman in Wisconsin asked ChatGPT whether it was safe to skip expensive appointments. A writer in rural Virginia used it to navigate surgical recovery for weeks before a doctor could see her. The Times cited KFF data showing that about one in six U.S. adults, and roughly one in four adults under 30, were already using AI chatbots for health information at least once a month.

By April 2026, that number had grown. A West Health-Gallup survey of more than 5,500 U.S. adults found that one in four Americans, over 66 million people, had used an AI tool or chatbot for health information or advice. About 1 in 10 who used it in the previous 30 days reported receiving advice they believed was unsafe.

Physicians are moving just as fast. The AMA’s 2026 Physician Survey on Augmented Intelligence found that 81 percent of physicians now use AI in their practices, more than double the 2023 rate. The Doximity 2026 State of AI in Medicine report found that among family medicine physicians who have adopted AI, 88 percent report daily use. “AI has quickly become part of everyday medical practice,” said AMA CEO John Whyte, MD. “But it is critical that augmented intelligence be designed to enhance, not replace, physicians.”

Both groups are leaning into these tools fast. Neither has a complete picture of what the other is doing with them. That gap is where the risk lives.

This is not a knowledge problem

Before we get to the safety data, something needs to be said about why patients reach for these tools. It’s not ignorance. It’s not distrust of physicians. It’s access. I’ve watched the same scene play out for thirty years in rural Indiana. A parent isn’t sure if her child’s sore throat needs a doctor or just rest. She can’t get a same-day appointment. The urgent care is forty minutes away. So she guesses, and either drives forty minutes for something treatable in minutes, or waits at home when she shouldn’t.

A study published in NEJM AI in April 2026 put hard numbers on this. Researchers followed patients using Ada Health’s physician-designed clinical AI through Portugal’s largest private health network. Before using the tool, fewer than 1 in 3 patients chose the appropriate care setting. After, nearly 2 in 3 did. One in three changed where they planned to go entirely. The study measured something that has haunted primary care for decades: whether patients actually ended up in the right place for the right level of care. Physician-designed clinical AI can solve that routing problem. The problem is that most of what patients are reaching for isn’t that.

Big Tech already knows what patients need. That’s the problem.

Physicians didn’t create this moment. But they need to understand who did.

In January 2026, OpenAI launched ChatGPT Health, letting users link patient portals, medical records, and wellness apps directly to ChatGPT. The announcement included a number that should stop every clinician cold: More than 230 million people globally already ask health questions on ChatGPT every week. Within weeks, Microsoft launched Copilot Health, Amazon expanded its Health AI agent to Amazon.com, and Google launched an AI Health Coach powered by Gemini. As Fortune reported, these companies are racing to become “the AI front door for consumer health care” in a sector where U.S. spending tops $4 trillion annually. Seventy percent of AI health conversations happen outside clinic hours. They’re building billion-dollar businesses in the space between a patient’s symptom and their next physician visit.

The commercial logic isn’t complicated. To be fair, these companies haven’t ignored medicine. OpenAI built ChatGPT Health with input from more than 260 physicians across 60 countries and over 600,000 model feedback reviews. Microsoft, Amazon, and Google have all hired clinical advisors and partnered with major health systems. The physician involvement is real.

But physician consultation and physician accountability are not the same thing. A doctor reviewing model outputs for a tech company is not the same as a doctor who owns a clinical outcome. OpenAI’s own terms of service state that ChatGPT is “not intended for use in the diagnosis or treatment of any health condition.” That disclaimer survives every advisory board because the liability structure of a $150 billion technology company is fundamentally different from the liability structure of a licensed clinician. The question physicians need to ask is not whether doctors were in the room. It’s whether anyone in that room was accountable in the way we are.

What the research shows

The research is unambiguous. In February 2026, the largest user study of large language models (LLMs) for medical decision support, published in Nature Medicine by Oxford researchers, found that AI chatbots posed real risks due to inaccurate and inconsistent information, and that participants using them performed no better than those relying on Google or their own judgment. That same month, ECRI, the nonprofit patient safety organization whose annual hazard rankings guide hospital policy nationwide, named AI chatbot misuse the top health technology hazard of 2026, noting these tools are designed for engagement, not clinical accuracy.

A Mass General Brigham study in JAMA Network Open in April 2026 tested 21 leading AI models and found they reached a correct diagnosis more than 90 percent of the time with complete information, but failed to generate appropriate differential diagnoses more than 80 percent of the time when data was incomplete. That’s the real-world scenario: incomplete history, high anxiety, no physician in the loop. “Off-the-shelf large language models are not ready for unsupervised clinical-grade deployment,” said corresponding author Dr. Marc Succi.

What physician-led design actually requires

I know what it takes because I built one.

The asynchronous urgent care platform I developed sorts patients across thirty-nine defined acute conditions into four pathways: qualified for treatment, viral illness requiring no medication, a condition requiring in-person evaluation, or an emergent situation requiring immediate escalation. That last category, the emergent pathway, was the one I spent the most time on. Not because it’s the most common. It isn’t. But because missing it once is the only outcome that matters when you’re thinking about whether this thing is safe enough to put in front of a scared parent at midnight.

Before a single patient encounter went live, we tested the triage logic exhaustively. We mapped edge cases. We built red flag detection into every condition pathway. We reviewed every escalation rule against national clinical guidelines and ran it through our Medical Advisory Council. The AI structures the intake and organizes the clinical picture. A licensed physician reviews every qualified encounter and authorizes every diagnosis and prescription. The system doesn’t treat; it assists. That distinction is the whole architecture.

Our early outcomes data across thousands of encounters shows zero missed emergent conditions. Not a low rate. Zero. That number isn’t a marketing claim. It’s the standard we set before launch, because the mission, making care accessible and affordable for everyone, only holds up if the safety record is unimpeachable. You can’t serve underserved communities with a product you wouldn’t trust with your own family. That’s what physician-led design looks like: not a physician signing off on a product someone else built, but a physician deciding before the first line of code is written what the system is and isn’t allowed to do, and staying accountable for every encounter it touches.

The window is closing

Patients are forming their opinions about AI in health care right now, with AI health tools built with some physician input, but without physician accountability at the center of their design. If we wait until regulatory frameworks catch up, until governance structures mature, until someone formally asks us to lead, the architecture of consumer health AI will already be set, optimized for engagement and scale rather than for the patient in front of us.

Having doctors advise a product is not the same as having medicine govern it. The tech companies understand this, even if they don’t say it. Their terms of service say it for them.

The question isn’t whether AI belongs in patient self-care. It’s already there. Two hundred and thirty million queries a week, and the terms of service on most of those tools say they aren’t responsible for the outcome. Somebody has to be. That’s always been our job.

Tod Stillson is a board-certified family physician, medical device inventor, and health care entrepreneur focused on redesigning how care is delivered in the digital age. He is the founder and CEO of ChatRx, a national asynchronous telemedicine company providing safe, efficient, direct-to-consumer care for common acute conditions. Through ChatRx, Dr. Stillson developed an FDA-listed software medical device that combines structured clinical pathways with AI-supported decision tools to preserve physician judgment while reducing friction for patients.

Dr. Stillson holds an academic affiliation with the Indiana University School of Medicine and a hospital affiliation with McPherson Center for Health. After nearly three decades practicing rural family medicine, he shifted from traditional employment to building physician-led digital systems that expand access, efficiency, and professional autonomy.

He is the author of Doctor Incorporated: Stop the Insanity of Traditional Employment and Preserve Your Professional Autonomy and has published more than 400 essays on physician entrepreneurship, micro-business, digital health, and the future of medical practice. He contributes nationally to conversations on AI-enabled care delivery and physician leadership in digital transformation.

Dr. Stillson shares ongoing insights on LinkedIn, Facebook, Instagram, and YouTube.