JAMA Data Brief
AI in Medicine
March 21, 2025
Poll: Trust in AI for Accurate Health Information Is Low
Avery Orrall; Andy Rekito
JAMA. Published online March 21, 2025. doi:10.1001/jama.2024.22951
As more artificial intelligence (AI) tools become available online, people are struggling to differentiate truth from fiction, including when it comes to health information. As part of its ongoing survey on health information and trust, KFF, a nonprofit organization focused on health policy, conducted a poll last year to investigate adults’ trust in AI to inform their health decisions.
Researchers surveyed a nationally representative sample of more than 2400 Black, Hispanic, and White adults in the US about the frequency with which they used AI, their trust in AI sources, and if they thought AI was doing more to help or hurt people trying to find accurate health information online.
About two-thirds of participants reported that they had used or interacted with AI. One in 10 said they use it several times a day. Younger adults were more likely to use AI than older age groups. The researchers noted, however, that many people may be unknowingly interacting with AI as it is increasingly integrated into search engines and social media sites.
The poll also found that on average 17% of adults reported using AI chatbots such as ChatGPT, Microsoft Copilot, or Google Gemini at least once a month specifically to find health information and advice. Among adults younger than 30 years, this increased to 25%. Only 10% of participants aged 65 years or older used AI for health information.
Most adults, including the majority of those who used AI, weren’t confident that the health information provided by AI chatbots was accurate. Six in 10 adults reported being “not too confident” or “not at all confident” in the accuracy of AI-generated health information. Only 5% of total adults said they were “very confident” in it. Although confidence in AI for health information did not vary widely by age, younger adults were more confident in the health advice given by AI chatbots than their older counterparts.
“Younger adults are more likely to say that they are using and interacting with artificial intelligence. So it follows that familiarity with AI platforms, and AI in general, is building confidence among younger adults,” Lunna Lopes, MSc, the lead author of the report and a senior survey analyst at KFF, told JAMA Medical News in an interview.
Although only about 1 in 3 respondents reported trusting AI-derived health information, a larger percentage said they trusted chatbots to provide accurate information about other topics. More than half of total adults said they had a “great deal” or “fair amount” of trust in AI for practical tasks, like cooking and home maintenance. A little less than half of total respondents said they trusted AI chatbots to provide them with technology information.
Participants were most skeptical about AI and politics. Only 2 in 10 said they had faith in AI to provide accurate political information. A quarter of AI users trusted the tools for political information, but among nonusers, the confidence dipped to just 9%. Both politics and health “carry more weight than advice on cooking or home maintenance,” Lopes said.
More than half of adults weren’t certain as to whether AI was helping or hurting people seeking accurate health information. The remainder were about evenly split between thinking AI was having a positive or negative effect. Even among those who used AI, almost half were unsure, 30% said AI was doing more to help, and 21% said it was doing more to hurt people trying to find health information.
The bottom line? “People aren’t quite ready to trust AI blindly when it comes to health information,” Lopes said. She added that she’s uncertain about what the future will bring, as the technology—and its usage—grows. “As people use AI more frequently, are they going to become more confident in their own skills of being able to differentiate what’s true vs what’s false? Or will it become an even greater challenge for people to determine what’s true and false from these AI models?”
Article Information
Additional Information: Percentages in figures may not add up to 100% due to rounding and because the share of survey respondents who skipped the question are not shown.
Published Online: March 21, 2025. doi:10.1001/jama.2024.22951
Conflict of Interest Disclosures: None reported.
