Virtually all the 645 practicing Canadian physicians who participated in a recent national survey believed that misinformation from AI is putting their patients’ health at risk.

In a survey conducted by the Canadian Medical Association (CMA) and Abacus Data, published in Physician Pulse, 97% of physicians reported having had to intervene to prevent harm or address consequences after a patient followed false or misleading health information found online, including advice from AI.

This finding supplements the results of the CMA’s 2026 Health and Media Tracking Survey, which found that people who followed health advice from AI were five times more likely to experience harms than those who did not.

“There is a lot of information out there in general, and with AI in particular, it’s a time of rapid change in the way patients are getting information about their health,” Bolu Ogunyemi, MD, CMA president-elect and clinical associate professor of medicine at Memorial University of Newfoundland in St. John’s, Newfoundland and Labrador, Canada, told Medscape News Canada.

“Some of the things we are seeing are causes for concern. We found that relying on AI to make health decisions can be harmful to patients,” Ogunyemi said.

Patients turn to AI for health advice because it’s fast and convenient, he noted. Another important reason is reduced access to primary care providers.

“Only about 27% of patients actually trust the information about their health that AI is giving them, but because a lot of patients do not have access to a family doctor or emergency care in a timely manner, even though they don’t necessarily trust the data, they figure that some information may be better than no information, so they are still going online,” Ogunyemi said.

The Government’s Role

Recently, the CMA, through its new Physician Advocacy Network, met with government officials in Ottawa to discuss issues such as the federal government’s role in curbing harm caused by false and misleading health information online.

“There is a regulation that the federal government can enact to reduce these harms and to keep social media outlets and other sources of health information accountable. There is Bill C-63, which is generally called the Online Harms Act, and we think this should be reintroduced. This bill addresses harmful material online, so because folks are being harmed by false health information found online, this can give some teeth to the federal government in terms of holding the purveyors of this false information to account,” Ogunyemi said.

“We also have the Controlled Drugs and Substances Act, so there is power to ensure that we can penalize someone who is selling information or medication that doesn’t work or has side effects, even though they are not selling it in person but promoting it online,” he added.

Ogunyemi also stressed the importance of increasing access to family physicians in Canada, as well as alleviating challenges that physicians currently face.

“Our recent data showed that for 85% of respondents, a physician is their most trusted source of health information. But it is important for us to understand that 1 in 6 Canadians does not have access to a family doctor, so we can’t blame them for wanting to get information about their health. Part of the overall solution is to increase capacity by having team-based care and by making the conditions that family doctors work in in Canada better so that more people are choosing the specialty,” Ogunyemi said.

Physicians can also do their part to make healthcare information more accessible to patients, he said.

“The CMA now has a program called Healthcare for Real. It’s an initiative you can find on social media, including Instagram, where we have doctors give information about health system navigation and how to find credible health information sources in a way that is fun, entertaining, and digestible for patients. It’s important for us to adapt our messages and our medium to reach patients in that way, to make sure there are accessible, trusted sources of health information out there for our patients,” Ogunyemi said.

Helping Patients Discriminate

AI has changed the scale and tone of health misinformation, Ma’n H. Zawati, PhD, associate professor of human genetics and research director at the Centre of Genomics and Policy at McGill University in Montreal, told Medscape News Canada.

“It can produce [seemingly] competent, authoritative answers that are wrong or oversimplified. For example, a chatbot may confidently state that a symptom is harmless when it could actually be serious. It may suggest a medication dosage that is approximate or not suitable for the person. AI may fabricate studies, recommendations, or scientific links that do not exist. Part of the issue is that we’ve created a system that sounds like a doctor and does not know the patient. It often lacks context: it doesn’t know your medical history or your risk profile, which are essential for proper medical care,” Zawati said.

Misinformation has become part of routine clinical practice, so doctors aren’t just diagnosing and treating, they’re now spending more time correcting beliefs that patients arrive with. It’s not just a fringe issue anymore, it is becoming systemic,” he added.

“The risk isn’t just misinformation, it’s misinformation that sounds credible and personalized. This is really at the heart of the problem, which is exacerbated by the disconnection of the health systems we have at the moment. We have a time where expertise doesn’t matter anymore. We have this false equivalence between validated data and something that is just spewed up from an influencer,” Zawati said.

Responses from AI can often be biased, depending on the sources. “For example, AI training datasets often overrepresent lighter skin tones, so AI systems may be less accurate in defining melanoma on darker skin.”

Genetic research is skewed toward European ancestry, Zawati said. “A large proportion, more than 75%-80% of genomic data comes from people of European descent, so AI models may generate less reliable predictions for other populations.”

Relying on AI could mean that cardiovascular disease goes underdiagnosed in women, he added. “Many datasets are built around classic symptoms of chest pain, which are more common in men. Women often present differently (with fatigue and nausea), so AI may miss or downplay their risk.”

Even mental health data can be biased, Zawati pointed out. “AI trained on specific linguistic patterns may misinterpret how distress is expressed in different cultures, so it may underestimate or misclassify symptoms in many individuals.”

He suggested the following steps for helping patients discern valid information from misinformation:

Check the source. “Prioritize information that comes from recognized health institutions such as hospitals and licensed professionals. Nothing anonymous, nothing that comes from influencer content.”

Be skeptical. “People chase certainty in health and medicine. Anything that sounds absolute or too definitive, especially online, is often a red flag. Good medical advice usually includes nuance, which is not something you will see if you ask ChatGPT or any other tool.”

Use online tools as a starting point. “People…shouldn’t replace a conversation with a healthcare professional who knows their situation. If something you read online changes your mind about a treatment, that’s exactly the moment to discuss it with your doctor, not act on it alone.”

Ogunyemi and Zawati reported having no relevant financial relationships.