AI health advice harms Canadians, survey finds
A Canadian Medical Association survey found that roughly half of Canadians use AI chatbots like ChatGPT and Google AI for health advice, and those who do are five times more likely to experience adverse health effects. The AI systems frequently provide medically inaccurate answers and lack the critical reasoning needed for safe health guidance.
What happened
A Canadian Medical Association survey of 5,000 Canadians conducted from November 3 to 13, 2025, found that roughly half of Canadians turn to artificial intelligence platforms such as ChatGPT and Google AI summaries for health information.
Canadians who consulted AI for medical analysis and treatment were five times more likely to report experiencing adverse effects on their health compared to those who did not use AI for health advice.
University of Waterloo researchers found that GPT-4 answered open-ended medical questions incorrectly roughly two-thirds of the time. Harvard researchers in a 2025 study found that chatbots rarely pushed back on nonsensical requests, such as asking why acetaminophen is safer than Tylenol (the same drug).
Reported impact
Canadians who followed AI counsel for medical analysis and treatment experienced adverse health effects at five times the rate of non-users.
Despite 73 per cent of Canadians expressing limited or no trust in AI for accurate health information, roughly half still use AI platforms for health advice.
Index analysis
The reliance on AI for health information appears driven by barriers to accessing traditional medical care: 57 per cent of survey respondents said they only searched the internet for health information when they could not access a family doctor or other health care professional.
AI systems providing health advice suffer from overconfidence and a tendency to prioritize helpfulness over critical reasoning, producing medically inaccurate guidance.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.