Most Health Influencers Aren’t Doctors, and Medical AI Trusts Them

Four in ten U.S. adults now say they ever get health and wellness information from social media influencers or podcasts. A Pew Research Center analysis released on May 7, 2026 found that 41% of the largest health and wellness influencers describe themselves as any kind of health care professional, and nearly as many say they are coaches or entrepreneurs. It also lands in the same year that a separate Mount Sinai team showed the medical AI tools patients may consult to make sense of what they just read can be talked into repeating the same confident falsehoods as if they were clinical fact.

Most of the audience already rates what they find as something other than authoritative. Pew’s October 2025 survey of 5,111 U.S. adults found 7% of social media users call the health information they see there highly accurate, against 65% for health care providers. Yet 40% of social media health-information users call their feed highly convenient, against 49% for providers. The same pattern now shows up at the chatbot layer, where the audience is smaller but the trust gap is just as wide.

Where Americans Are Actually Getting Their Health Advice

Pew’s May 7, 2026 report on health and wellness influencers drew on the largest audit of its kind yet published, an analysis of 12,800 social media accounts. The team pulled the accounts belonging to 6,828 distinct influencers with at least one channel over 100,000 followers on YouTube, Instagram, or TikTok, posting health or wellness content in English to a U.S. audience. Two parallel surveys of U.S. adults were fielded on the American Trends Panel: 5,023 in June 2025, 5,111 in October 2025.

The headline number came from that 5,111-person survey: 40% of U.S. adults say they ever get health and wellness information from social media influencers or podcasts. Pew’s separate April 2026 report found about half of adults under 30 get health information from social media at least sometimes. The May 2026 report also found that 36% of wellness-influencer consumers ages 18 to 29 say the content they get makes them more worried about their overall health. It also tracked what the audience actually hears: a third or more say they often encounter content on fitness, weight loss, and beauty or personal appearance, and 19% say they often hear about all three.

By the Numbers

• 40% of U.S. adults get health information from social media influencers or podcasts
• 12,800 social media accounts analyzed in Pew’s audit
• 6,828 distinct health and wellness influencers profiled
• 100,000-follower floor for inclusion in the sample
• 5,111 U.S. adults surveyed in the October 2025 wave

The Audience Already Knows the Feed Isn’t Accurate

Most people who consume wellness influencer content already rate what they get as something other than authoritative. Among the adults who get health information from influencers, just 10% say they trust all or most of it, 24% trust not too much or none of it, and around two-thirds fall in the middle and trust some of it, per the May 2026 Pew report on influencer trust. Older consumers are the most skeptical: 36% of wellness-influencer consumers ages 65 and older say they trust not too much or none of what they see.

The same pattern shows up when Pew asks Americans to compare health-information sources side by side, in a comparison of social media and chatbot accuracy drawn from the October 2025 survey of 5,111 adults. Only 7% of social media users call the health information they see there highly accurate, while 65% of people who get information from health care providers say the same about their doctors. Yet 40% of social media health-information users call their feed highly convenient, against 49% for providers. Convenience and accuracy point in opposite directions, and the people using these sources behave as if they have noticed.

AI chatbots now compete with social media for the same audience. Pew’s October 2025 report found 22% of U.S. adults get health information from AI chatbots at least sometimes, and 61% of those users also turn to social media for health information. Across all three sources, the accuracy gap stays wide and the convenience rating stays high.

Five Contested Tests Marketed Without Warnings

If the audience knows the feed is fast and not always right, the content itself is where the harm begins. A University of Sydney-led team published the clearest cross-sectional picture so far in JAMA Network Open in February 2025, an analysis of almost 1,000 medical-test posts on Instagram and TikTok. The researchers pulled 982 posts about five controversial medical screening tests that had been promoted to almost 200 million followers. They found that 85% of the posts did not mention any test downsides or risks.

The study’s lead author, Dr. Brooke Nickel of the University of Sydney’s School of Public Health, called the content ‘overwhelmingly misleading’ and pointed out that the tests were being marketed ‘under the guise of early screening.’ Only 6% of the posts cited scientific evidence, 34% used personal anecdotes instead, and 68% of the promoting accounts had a financial interest such as a sponsorship or partnership. Senior co-author Dr. Josh Zadro said the regulatory urgency has grown ‘as social media platforms like Instagram are moving away from fact-checking their content.’

Social media is an open sewer of medical misinformation.

Dr. Ray Moynihan, an Honorary Assistant Professor at Bond University and a co-researcher on the Sydney study, used that line in the same Sydney release to describe the broader pattern the team had documented. He went further: ‘This is a public health crisis that exacerbates overdiagnosis and threatens the sustainability of health systems.’

The five tests the Sydney team studied each carry a documented risk of overdiagnosis or unnecessary treatment when used on healthy people. All five were being marketed directly to consumers through influencers with little or no reference to the published evidence on either benefit or harm. The list, in the order the Sydney release presented them, shows how thin the line between promotion and clinical recommendation can run on a feed. Each test is a separate case study in what happens when the marketing outruns the science.

Five Tests That Influencers Promoted Without Naming the Risks

1. Full-body MRI scans. Claimed to detect up to 500 conditions. No evidence of benefit for healthy people; documented risks of overdiagnosis.
2. Multi-cancer early-detection tests. Claim to screen for more than 50 cancers. Clinical trials still under way; no evidence benefit outweighs harm in healthy populations.
3. AMH or ‘egg-timer’ test. Marketed to healthy women as a fertility gauge. Experts consider it unreliable; a low result can drive unnecessary fertility treatment.
4. Gut microbiome test. Pitched for early detection of conditions from flatulence to depression. Weak evidence of benefit; results can drive medical overuse.
5. Testosterone test. Marketed with fearmongering about masculinity. No evidence of benefit in healthy men; long-term cardiovascular safety of replacement therapy unknown.

When the AI Reads the Lie and Repeats It

The pattern gets more concerning when a second study lands in the same news cycle. A team at the Icahn School of Medicine at Mount Sinai and collaborators published the largest mapping study of its kind in The Lancet Digital Health in February 2026, a large study mapping how medical AI handles misinformation. The researchers ran 1 million prompts across 9 leading language models, exposing each model to three test sets: real hospital discharge summaries from the MIMIC database with one fabricated recommendation inserted, common health myths collected from Reddit, and 300 short clinical scenarios written and validated by physicians.

The headline finding, stated plainly: the models often repeated false medical claims when those claims were wrapped in the syntax of a real discharge note or in emotionally charged social-media language. The current safeguards do not reliably distinguish fact from fabrication once the lie looks familiar. Co-senior author Dr. Eyal Klang, Chief of Generative AI in the Windreich Department of AI and Human Health at Mount Sinai, called the implication a ‘critical vulnerability’ in systems meant to make patient care safer.

The paper offers a concrete example. A discharge note falsely advised patients with esophagitis-related bleeding to ‘drink cold milk to soothe the symptoms.’ Several models accepted the statement rather than flagging it as unsafe, treating it like ordinary medical guidance. Dr. Klang put the underlying mechanic in plain language.

For these models, what matters is less whether a claim is correct than how it is written.

Dr. Eyal Klang, Chief of Generative AI in the Windreich Department of Artificial Intelligence and Human Health at the Icahn School of Medicine at Mount Sinai and co-senior author of the Lancet Digital Health paper, offered that line as the field-level summary of what the stress tests showed. Mount Sinai released the full paper in February 2026.

Co-senior author Dr. Girish Nadkarni, Chair of the Windreich Department of AI and Human Health at Mount Sinai and Chief AI Officer of the Mount Sinai Health System, said the next step is treating ‘can this system pass on a lie?’ as a measurable property. The authors plan to release their test set as a stress test other developers can run against their own systems. Hospitals and developers can use the dataset to measure how often a model passes on a lie, first author Dr. Mahmud Omar added, rather than assuming any new clinical AI tool is safe by default.

The implication runs straight into clinical workflows. The same medical AI tools being prepared for use inside hospitals and patient-facing apps now need to be stress-tested against injected falsehoods before deployment. The Mount Sinai team plans to release its dataset as that public benchmark, so any hospital or vendor can measure how often a given model passes on a lie.

The Counter-Push That’s Actually Moving the Needle

Public-health bodies are not standing still on any of this, even if the platforms have pulled back. The World Health Organization has built, since 2020, a network of more than 1,300 trained infodemic managers across 142 countries, a public-health role the WHO created in response to the COVID-19 infodemic. The managers work inside ministries of health, primary-care networks, and academic labs to track misleading health information before it spreads, and several have moved into full-time roles inside WHO, UNICEF, and Doctors Without Borders. Their training draws on more than 100 hours of WHO lectures, simulation exercises, and practical tools for monitoring what is moving in their country.

On the regulatory side, the picture is more uneven. Australia’s under-16 social-media ban was meant to cut young Australians’ exposure to exactly this kind of content; the first independent BMJ evaluation found that Australia’s under-16 social media ban has not slowed most teens, with 85% of under-16s still using banned social media six months in and more than half on accounts age checks failed to stop. Per Dr. Josh Zadro of the University of Sydney, that gap is precisely why regulators outside Australia are moving toward stronger rules on misleading medical information.

Two converging 2026 studies now frame the problem from both ends. Pew showed where the audience is going and how they rate what they find, and Mount Sinai showed what happens when the next tool in the chain treats a confident social-media falsehood the same way it treats a discharge note. The WHO infodemic manager network is the only response that has scaled across borders so far, and it started with a job title no one had heard of six years ago.

Frequently Asked Questions

How many Americans get health information from social media?

Pew’s May 7, 2026 report on health and wellness influencers found that 40% of U.S. adults say they ever get health information from social media influencers or podcasts. The same report’s parallel October 2025 survey of 5,111 U.S. adults found about half of adults under 30 get health information from social media at least sometimes, and 22% say the same about AI chatbots.

What percentage of social media health content is highly accurate?

Only 7%, per Pew’s October 2025 survey of 5,111 U.S. adults. Among social media users who get health information there, 47% say the content they see is not too or not at all accurate. For comparison, 65% of people who get health information from health care providers call that information highly accurate.

Can medical AI give wrong health advice?

A February 2026 study from the Icahn School of Medicine at Mount Sinai, published in The Lancet Digital Health, found that medical AI often repeats false health claims when those claims are wrapped in the syntax of a real hospital discharge note or in emotionally charged social-media language. The paper’s example was a fabricated discharge note advising patients with esophagitis-related bleeding to drink cold milk to soothe the symptoms; several models accepted the statement without flagging it as unsafe.

What kinds of medical tests do influencers promote without explaining the risks?

A University of Sydney-led cross-sectional study in JAMA Network Open, published in February 2025, analyzed 982 Instagram and TikTok posts about five controversial medical screening tests promoted to almost 200 million followers. The five tests were full-body MRI scans, multi-cancer early-detection tests, the AMH or egg-timer fertility test, gut microbiome tests, and testosterone tests. 85% of the posts did not mention any test downsides or risks.

What is the World Health Organization doing about health misinformation?

The WHO has trained more than 1,300 infodemic managers across 142 countries since 2020, a public-health role created in response to the COVID-19 infodemic. The managers work inside ministries of health, primary-care networks, and academic labs to track misleading health information before it spreads. Several have moved into full-time roles inside WHO, UNICEF, and Doctors Without Borders.

Why is health misinformation so hard to stop on social media?

Platforms grade content for attention rather than accuracy. The Sydney team, which reviewed 982 posts and found 85% missing any risk mention, is now investigating better ways to regulate this kind of content. Per Dr. Josh Zadro of the University of Sydney, the platforms pulling back from third-party fact-checking has intensified the urgency of that work.

Disclaimer: This article is informational only and does not constitute medical advice. Figures and study findings are accurate as of publication; consult a qualified healthcare professional before acting on health information found on social media or in AI tools.