While traveling to Portland, Peggy Dahl developed a cough so bad it left her gasping for breath after each bout. Far from her doctor, and with a sneaking suspicion of bronchitis, the Minneapolis resident turned to ChatGPT.

This wasn’t the first time she had used it as a medical sounding board.

During a visit to her daughter in Georgia, the chatbot had correctly flagged symptoms of a urinary tract infection. On the Portland trip, ChatGPT ruled out bronchitis and urged her to find a professional. An X-ray at a nearby urgent care soon confirmed a case of pneumonia. Without ChatGPT, Dahl said, she probably would have delayed getting treatment.

“It was really good,” Dahl said. “It just talks to you like your best friend.”

Like Dahl, 29% of American adults say they use AI tools or chatbots for health information each month — nearly double the number two years ago, according to June survey data from the health research firm KFF.

Medical experts see helpful and harmful outcomes from AI in consumer health. A health chatbot could provide clarity and more individualized guidance to patients. However, early research suggests an overreliance on AI could pose privacy risks or prevent people from seeking care when needed.

“There’ve been many technological innovations in health care in the last few decades, but this is by far the fastest adoption of anything I’ve seen,” said Dr. Andrew Olson, a physician and professor at the University of Minnesota who has studied artificial intelligence in health care.

A decade ago, Olson said, patients were searching for symptoms on Google. Now, he sees AI tools loaded on their personal devices in the exam room. Some of the systems are designed to explain diagnoses, evaluate symptoms and inform patients, so they come to appointments with questions.

Tech companies are looking to cash in. Earlier this year, OpenAI and Amazon both launched health-focused chatbot features for uploading medical records, test results and fitness tracker data. In June, Microsoft announced a partnership with the Mayo Clinic to develop its own health-focused AI product.

Those AI models can be “empowering” when they distill dense medical data to provide accurate personalized information, triage new injuries and clarify obscure medical terms, said Martin Michalowski, an AI-focused researcher at the University of Minnesota’s School of Nursing.

But he’s wary of the limitations: models like ChatGPT work by predicting the most likely next words in a sentence based on the texts used to train them, not by holistically generating knowledge about a topic. And he advises patients to be cautious about providing data to a chatbot they might not be comfortable sharing publicly.

Besides privacy concerns, the medical results so far have been mixed. One study testing OpenAI’s ChatGPT Health model found it underdiagnosed emergency cases about half the time. Last month, a Florida pastor sued OpenAI, claiming ChatGPT led him to delay care for a critical blood clot in his lungs.

Dr. Olson shares the concern that patients might blindly trust tools not designed for health care. He’s wary of bias in chatbots that agree with whatever a user suggests, even if it’s not medically accurate.

But he’s also optimistic about new uses for AI in the clinic, especially when it can help patients better understand their own health or decide whether to make a long drive to a hospital.

“I think we should encourage people to advocate for their own health,” Olson said. “You should look stuff up. And then you should talk about it with your health care provider.”

When lawyer Marc Berris started feeling brain fog, he thought his clinical depression was getting worse. Swapping antidepressants didn’t change his symptoms, and his primary care doctor was stumped.

But an AI tool flagged an important caveat. The 57-year-old from New Hope had recently seen other specialists who’d prescribed the drug topiramate for weight loss, and they hadn’t warned him that brain fog was a possible side effect.

He brought this AI-powered theory to his primary care physician, who helped him swap topiramate for a GLP-1 medication. Soon after, Berris’s brain fog lifted. Since then, he’s used ChatGPT to build workout plans and find recipes to manage the nausea that can accompany the drugs.

“What I think [AI] is doing is creating more informed patients,” Berris said. “Even if the information that I get through an AI tool is not spot-on, it’s at least getting me to ask the right questions of my providers.”

Berris doesn’t think ChatGPT will replace his doctor anytime soon. He still trusts providers with a traditional medical school education over a machine learning model, he said.

But other patients are embracing AI to make health care decisions. St. Paul, Minnesota, resident Bryan Hayes, 62, has accumulated a lot of health care data over nearly a decade of dabbling in the DIY longevity movement known as biohacking.

Before, each test result was locked behind its own patient portal; now he has AI pore over his blood work, MRI results and lung scans together.

Hayes used Anthropic’s Claude tool to build his own website, themed after the science fiction novel “The Hitchhiker’s Guide to the Galaxy,” where he could upload and store his results. Hayes feeds in new data, and the website uses a Claude-powered agent named “Eddie” to analyze it and provide recommendations.

When he brought his AI-generated health report to a cardiologist to obtain a prescription for an uncommon cholesterol medication, he said the physician was intrigued.

“In 30 years, he had not ever seen anything quite like that,” Hayes said.

After Henry Dorgan slipped on a spot of water in his bathroom and hyperextended his knee, he thought about making a doctor’s appointment.

During his visit the month prior for a painful jaw disorder, the 31-year-old from Minneapolis felt frustrated to receive a prescription for a muscle relaxer, but no rehab options and a four-month wait to see a provider. For a knee injury, he guessed the costs of imaging and prescription drugs would add up fast.

So instead, he wrote a detailed message to Microsoft’s Copilot AI chatbot, which he’d used at work. He offered his height, weight, injury description and pain levels, and Copilot shot back a possible diagnosis: a grade 1 patellar tendon strain.

As he updated the chat with symptoms, it adjusted his physical therapy plan. Dorgan was an avid runner, so Copilot suggested he might try a low-impact spin bike while he recovered. And if he experienced swelling, the chatbot said, he should see a doctor to check if the tendon was torn.

“It was extraordinarily descriptive,” Dorgan said. “I probably saved hundreds of dollars in medical costs.”

If he realized something was seriously wrong, Dorgan said he’d go straight to a doctor. But with rising health care costs and long wait times to see humans, he’s among a new cadre of patients who plan on asking AI to triage symptoms and find savings.

Woodbury, Minnesota, resident Leigh Ann Ahmad, 51, turned to AI when she felt her optometrist dismissed symptoms of a growing blind spot in her eye. ChatGPT’s triaging encouraged her to return with recommended follow-up questions. Her optometrist referred her to a specialist, who eventually diagnosed her with a long-hidden nerve disorder in her eye.

But then, she faced a new problem: a $5,000 bill for follow-up imaging her doctor recommended, but her insurance refused to cover.

ChatGPT helped her find the right words, she said. Based on its instructions, Ahmad asked the insurance agent about initiating an “appeal process.” Something clicked: the insurance agent called her doctor’s office to seek authorization, and the bill disappeared.

“When I used that (language) in the call with my insurance company, it was like night and day,” Ahmad said. “I never had to fill out another form, or had to get another bill outstanding, or make another call.”

©2026 The Minnesota Star Tribune. Visit at startribune.com. Distributed by Tribune Content Agency, LLC.