Photo: Alexander Spatari/Getty Images
More than 52% of Americans turn to ChatGPT when experiencing concerning medical symptoms, according to a new report from The Mesothelioma Center.
The organization surveyed 750 U.S. adults and more than 250 clinicians to gauge how AI tools are influencing patient behavior, trust and clinical decision-making.
The findings suggest a widening gap between how patients rely on AI and how healthcare providers assess the risks.
Nearly one in three Americans says they would skip or delay seeing a doctor if an AI tool characterizes their symptoms as low risk. Half of the respondents who used ChatGPT for symptom checking said the tool "led to a diagnosis," underscoring concerns about overconfidence in AI-generated assessments.
Dr. Daniel Landau, oncologist and hematologist, told MobiHealthNews that when patients say AI "led to a diagnosis," they typically mean it gave them a name to cling to, whether or not it is accurate.
"That’s where it gets dangerous; it creates a false endpoint in what should be the beginning of a diagnostic process," Landau said.
He cautioned that this kind of premature closure can create bias or form an opinion ahead of time, making it harder to accept new information and delaying further evaluation, which in turn makes it harder to preserve safety and clinical accuracy.
He points out that many serious conditions, especially cancers, do not always present themselves with textbook symptoms.
"When AI incorrectly flags these as conditions as low-risk, patients may miss critical windows for intervention," Landau said.
Delayed care does not just raise health risks; it also increases the financial burden of long-term costs and the complexity of treatment when patients eventually seek help.
The report indicates that clinicians are seeing the effects at the exam room level, with 58% of healthcare professionals saying AI is making it harder to treat patients.
One in three professionals reports encounters with patients convinced they had a serious illness based solely on an AI tool’s output.
While some providers see limited promise in early screening, only about one in seven believe AI should be allowed to diagnose cancer, and only in tightly controlled, early-detection scenarios.
"Patients frequently walk into their provider's offices expecting validation, not evaluation,” Landau said. "They’ve already Googled their symptoms or consulted with ChatGPT, and now they want a doctor to reaffirm their diagnosis and co-sign it."
He said this shift puts clinicians on the defensive side of things, needing to "disprove" their patients’ findings rather than diagnose them.
"It slows down appointments, damages rapport and turns medical visits into debates rather than collaborative care," Landau said.
He argues AI should not enter clinical screening without the same rigor and testing that patients expect from new medications or medical devices.
This includes putting in the work through peer-reviewed validation across diverse populations, independent testing, and clear documentation of false positives and negatives.
"Without strong regulatory oversight, we risk embedding bias or error into systems that were meant to improve access and accuracy," Landau warned.
THE LARGER TREND
Despite the risks posed by AI use in diagnosis and preliminary care, adoption of the technology in the healthcare industry is accelerating. An OpenAI report found healthcare is among the fastest‑growing sectors of AI adoption.
In professional settings, AI continues to advance as a means of aiding diagnosis. Korean researchers at Seoul National University Hospital recently developed an AI tool to identify osteoporosis risk from a standard chest X-ray.
Pangaea Data and AstraZeneca are partnering on AI tools to detect rare and hard-to-diagnose diseases, aiming to identify undiagnosed patients, support clinical trial enrollment and advance more precise treatment strategies.
