Wall Street Journal: Doctors Dislike Patients Using AI to Interpret Lab Results, But Can't Stop It

Doctors already had to contend with patients who sought medical advice online, but now they face a more powerful new version of 'Dr. Google.' Research shows that most American adults now view their lab results online before speaking with their doctors, thanks in part to a 2021 regulation that allows patients near-instant access to personal health information, including lab reports, imaging results, and clinical notes. More patients are now uploading their test results directly to AI chatbots, raising concerns among healthcare professionals about AI 'hallucinations,' misinterpretations, and privacy breaches. 'Patients come in convinced they have a certain condition based on their online searches, and we have to try to counter that,' said Dr. Jason Goldman, an internist in Coral Springs, Florida. 'You get conclusions ranging from “you just have a cold” to “you’re going to die from cancer.”' Goldman noted that some patients have used AI conversations to justify stopping life-saving cholesterol-lowering medications. Others have questioned doctors’ vaccine recommendations because AI validated their concerns. 'I give a recommendation, and they say, “I’ll ask AI,”' Goldman said. 'And I think, “Then why are you here?”' A 2026 KFF public opinion poll found that about one-third of adults consult AI chatbots for health advice, with 19% specifically using AI to interpret lab data or medical reports. OpenAI reports that over 300 million people ask health-related questions on ChatGPT each week. Numerous companies are eager to capitalize on this trend. Anthropic’s Claude can connect to some users’ medical records, and electronic health record giant Epic Systems has built an AI assistant into its software. In July, OpenAI broadly rolled out a health feature allowing users to link their medical records and wearable device data, offering insights into health trends over time. Ashley Alexander, OpenAI’s vice president of health products, said the model’s 'reasoning ability surpasses that of clinicians.' The company has stated the model is not a substitute for a doctor. Many clinicians remain skeptical. Dr. Mary K. Mulcahey, director of sports medicine at Loyola University Medical Center and a specialist in shoulder and knee injuries, said AI might struggle to distinguish between types of meniscus tears. Her treatment recommendations vary based on tear type, patient activity level, and arthritis severity—context that AI may miss when pulling information from large language models. An increasing number of Mulcahey’s patients are inputting imaging results into AI. 'It’s dangerous when patients come in believing they already know what treatment they need, without considering the physician’s or surgeon’s clinical judgment,' she said. Mulcahey and other doctors acknowledge AI can be helpful in translating complex medical jargon into more understandable terms for patients. A 2024 study examined how AI chatbots simplified pathology reports. It found ChatGPT reduced the average reading level to a seventh-grade level with nearly 98% medical accuracy. However, ChatGPT falsely reported no cancer in lymph nodes when none had been tested—a common error also seen in other chatbots, according to the study’s authors. In 2024, 32-year-old software engineer Megg Gawat received blood test results and was confused by medical terms and acronyms. Instead of waiting for a call from her clinician, she turned to ChatGPT for interpretation. She appreciated the freedom to ask questions anytime without fear of judgment or embarrassment. Initially, AI gave a positive assessment: 'You’re doing great,' it wrote. But Gawat immediately spotted errors, including a misreading of her low vitamin D levels. After she corrected the AI, its tone became urgent, telling her to 'immediately' start taking vitamin D supplements, complete with a red alert emoji. Gawat recalled that when a nurse later called with the results, they also recommended vitamin D supplementation—but in a far less urgent tone. Another case, documented in a spring 2024 case study published in the American Journal of Geriatric Psychiatry, involved a 65-year-old veteran. He input results from two prior memory tests into ChatGPT. While his doctor had diagnosed mild cognitive impairment, the chatbot told him his condition had progressed to dementia. In reality, according to the case study, his cognitive function had not declined further. An OpenAI spokesperson said individual case studies do not represent broader user experiences; newer OpenAI models have enhanced capabilities and safeguards, including additional encryption layers for integrated health data. Brad Malin, a biomedical informatics professor at Vanderbilt University who studies health data privacy, said patients should understand that uploading health data to chatbots means giving up federal protections under HIPAA. Companies set their own privacy policies and can change them at will, he added. 'In this context, they’re not patients—they’re consumers,' Malin said. 'It’s largely buyer beware.'