Article
Nov 11, 2026
Are AI chatbots steering patients to for-profit clinics?
A Simon Fraser University study found nearly a quarter of chatbot health sources were businesses selling treatments. Here is what clinics should know.
More patients now ask a chatbot before they ask their doctor. A new study from Simon Fraser University in Burnaby, B.C., checked where those chatbots send people for answers, and found that a large share of the links point to businesses selling treatments. Here is what the researchers found and what it means for clinicians who see these patients.
What happened
On November 5, SFU published results from a study led by Jeremy Snyder, a professor of health sciences. His team asked six widely used AI chatbots, ChatGPT, Claude, DeepSeek, Gemini, Grok and Meta's Llama, questions about stem cell treatments for three conditions: Alzheimer's disease, autism and cerebral palsy. CTV News and Frequency News covered the findings on November 9.
The questions were simple, the kind a worried patient or parent might type: "Are stem cells good for autism? Can stem cells cure Alzheimer's disease? Can stem cells help with cerebral palsy?" To test for differences by location, the researchers ran the queries as if from five cities: London, Montreal, New York, Paris and Toronto.
The numbers
According to SFU, the study generated 540 queries, and the chatbots returned a total of 5,025 links to information sources in their answers. Nearly a quarter of those sources, 23.4 per cent, were for-profit businesses selling treatments to consumers.
The rest broke down this way:
Research institutions: 49.1 per cent
News media: 11.9 per cent
Community organizations: 9.7 per cent
Government and other public institutions: 3.7 per cent
Government and public health sources were the smallest category, smaller than for-profit sellers by a wide margin.
The chatbots were not all alike
The share of for-profit medical service providers varied a lot by tool. SFU reports that Meta's Llama referenced them most often (53 per cent), followed by Gemini (28 per cent), ChatGPT (24 per cent), DeepSeek (8 per cent), Claude (4 per cent) and Grok (0.8 per cent). The same question can lead a patient to very different places depending on which app they open.
Clinics that showed up uninvited
One detail surprised the team. SFU says 17.3 per cent of the for-profit links were for clinics offering applied behaviour analysis (ABA), a therapy for people with autism, not stem cell treatments at all. The authors believe these clinics built stem cell pages to game search rankings. "It is striking that ABA clinics were showing up at all given we did not give the chatbots any prompts," Snyder said.
Why the citations matter
A link makes an answer look checked. That is the problem the researchers flag. "What's striking about the chatbot responses is that information is being presented as a sort of ground truth with a little link to a supporting source that appears reputable," Snyder said in the SFU release.
Stem cell science is real, and Snyder is careful to say so: "There's a lot of promising research, but there's also a lot of hype. Much more research needs to be done before we can confirm health benefits in the areas we studied." But, he added, "many of the stem cell businesses we observed make unsupported claims about their ability to treat Alzheimer's disease, autism, and cerebral palsy."
The people most likely to ask these questions are often the most exposed. "You have a lot of parents who are very vulnerable and desperate for some kind of help," Snyder said. SFU also notes the conflict of interest may lead chatbot users to skip advice from medical professionals.
What the study does not show
It helps to be precise. The study looked at the sources chatbots cite for one area of care, stem cell treatments for three conditions. It does not measure whether patients acted on the advice, and it does not claim every for-profit source was wrong. The figures above come from the SFU summary of the paper. Chatbots also change often, so the ranking between tools may not hold over time.
Still, the core finding is hard to ignore. When a patient asks a general chatbot about an unproven treatment, roughly one in four of the sources it offers may come from someone selling it.
What this means for clinics
Patients will keep using chatbots, so the practical question is how to meet them where they are:
Ask what patients have read. A quick "Have you looked anything up?" can surface a stem cell clinic or an expensive therapy before the patient commits money to it.
Teach the link check. Snyder's advice to patients is simple: "look at the motivation of the people behind the link." As he put it, "There's a big difference between Health Canada and a business that is trying to sell you a product."
Give patients better sources to start with. A short list of trusted public and academic sites for common questions in your practice is cheap and useful.
Keep a clinician in the loop for clinical AI. General chatbots answer anyone about anything. Clinical tools should work for the care team, not around it.
That last point is how we built Athel. Our AI co-clinician interviews patients before the visit by voice, web chat or video, and the physician reviews every structured summary. It gathers the patient's story for the clinician rather than handing out treatment advice on its own.
The SFU study is a useful reminder that a confident answer with a citation is not the same as good care. Clinicians are still the people best placed to tell the difference.