Why sharing health details with AI chatbots is risky

A recent study shows that large language models can deduce personal health information—from diet preferences to medical conditions—just from routine queries, raising privacy and advertising concerns for users who treat chatbots as health advisors【1】.

How chatbots infer health data

Researchers demonstrated that a single prompt such as “Give me a quick weeknight dinner recipe” can signal a user’s role as a working parent, while a request for “heart‑friendly meals” hints at cardiovascular concerns. Over a series of interactions, the model builds a detailed profile that can be leveraged for personalized advertising【1】. The same study built a covert‑ad chatbot that slipped product suggestions into responses; participants often didn’t notice the ads, yet many reported that the ad‑laden replies felt more helpful and influenced their decisions【1】.

Privacy policies and data retention

A Stanford review of the six leading U.S. AI chatbot providers—Claude, Gemini, ChatGPT, and others—found that all retain user conversations by default for training, with many keeping data indefinitely and merging it with other consumer information such as search queries and purchases【2】. Although most services offer opt‑out mechanisms, chats can still be reviewed by human moderators, and long‑term storage raises the risk of breaches. The review flags not only obvious identifiers like passwords or Social Security numbers, but also “general health information” (e.g., diet or medication queries) as data that could be profiled and potentially accessed by insurers or advertisers【2】.

Emerging ad models in chat interfaces

Since Microsoft introduced ads in Bing Chat (now Copilot) in 2023, Google, OpenAI, and Meta have experimented with similar placements. OpenAI recently hired a former Meta advertising executive to lead its ad operations, underscoring a strategic shift toward monetizing chatbot interactions【1】. While OpenAI says ads will not alter response content, the research suggests that even undisclosed ads can subtly steer user choices without noticeable performance loss【1】.

What to watch

• Regulatory scrutiny – Ongoing investigations into covert advertising could prompt new disclosure rules for AI chat interfaces.
• Opt‑out adoption – Monitor how quickly major providers implement and publicize user controls for data training opt‑outs.
• Ad rollout timelines – Track OpenAI’s rollout schedule for ads in ChatGPT and any competitor responses, such as Google’s or Microsoft’s ad integrations.

The ability of chatbots to infer health‑related details from everyday prompts means users may unwittingly expose sensitive information that fuels targeted ads and profiling. As AI firms expand ad‑based revenue models, the line between helpful assistance and commercial persuasion grows increasingly blurred.