Almost two-thirds of Americans who have asked artificial intelligence tools for medical advice acted on the guidance without consulting a doctor, an eHealth survey of more than 1,000 people found.
Half of respondents had turned to AI for medical advice. More than four-fifths of AI users said they trust the medical advice the tools provide, with 29% of Americans saying they have complete faith in the outputs. EHealth found 17% of AI users mostly distrust the medical advice the technology provides, while 1% completely distrust the outputs.
While a minority of people using AI for medical advice are skeptical of the tools’ outputs, most users are confident enough in the information to use the technology to shape their care decisions. More than 70% of AI users changed their decision to seek medical care based on a tool’s medical advice. The figure includes 36% of AI users who opted against seeking medical care based on the advice.
Chatbots based on large language models such as ChatGPT, Claude and Gemini have performed well on medical exams. Yet research suggests that basing care decisions on the advice of such AI tools may lead to poor choices.
A recent study found that people struggled to use the tools effectively, providing incomplete information and submitting queries that were misinterpreted by the AI models. Slight variations in user questions led to very different answers, and tools provided a mix of good and bad information that users struggled to filter.
One caveat is that the study was performed in 2024, and AI models have improved since then. A preprint paper focused on AI performance in women’s health found newer models show “significant improvements in avoiding inappropriate recommendations,” although the authors still concluded the tools were unable to provide reliable advice.
The survey by eHealth, which helps Americans find healthcare coverage, also covered the use of AI to answer questions about insurance. About one-third of insured Americans have used AI to help them understand how health insurance works. More than 90% of those people said they trust the information they received.
Three-quarters of respondents said they are confident in their ability to understand key health insurance terms. Yet eHealth found 61% of people were unable to accurately define “coinsurance,” and 56% were unable to accurately define “deductible.” Between 33% and 45% of people were unable to accurately define “out-of-pocket maximum,” “premium” and “copayment.”