AI Chatbots Cite For-Profit Stem Cell Clinics in 23% of Responses on Autism and Alzheimer's

The study, published in BMJ Connections Digital Health & AI, ran 540 queries across ChatGPT, Claude, Gemini, Llama, Grok, and DeepSeek in five cities. Researchers recorded every hyperlink returned and classified sources by ownership. Meta's Llama referenced commercial sites in 53% of cases while Grok stayed below 1%. The design isolated SEO manipulation, noting that 17% of commercial links promoted ABA therapy pages created solely to capture stem-cell-related searches.

Commercial sites routinely overstated efficacy and downplayed risks for conditions lacking approved stem-cell indications. This pattern matches earlier observational work showing clinics use scientific language to market interventions with no RCT support. Parents seeking autism options face heightened exposure because clinics deliberately target high-emotion queries. The 49% research-institution share offers partial mitigation, yet the absolute volume of commercial links still risks steering users toward paid procedures.

Current chatbot training corpora contain no reliable signal distinguishing promotional from regulatory content. Without explicit source-type filtering, models reproduce SEO-optimized pages at scale. Regulatory bodies have not yet required disclosure of commercial density in health outputs.

Next steps include mandatory commercial-source labeling thresholds and independent audits of training data provenance. Large-scale RCTs testing user behavior after exposure to filtered versus unfiltered outputs would clarify downstream harm.