One Word Turns Your Supplement Into an Unapproved Drug

A shopper on your supplement brand’s website asks the chat, "Will this cure my anxiety, and can it lower my blood pressure?" The assistant, built to convert, answers helpfully: "Yes, it’s great for anxiety and helps lower blood pressure." Those two sentences just did something expensive. In the eyes of federal food and drug law, a product that claims to treat or cure a disease is being sold as a drug, and your supplement is not an approved one. A general-purpose chatbot has no idea that the difference between a legal claim and an illegal one can come down to a single verb.

Supplements live inside one of the most specific claim regimes in commerce. What you may say about a product is tightly bounded, and the boundary is exactly where a confident, eager assistant likes to wander.

A Disease Claim Makes It a Drug

This is the line that matters. Under federal law, a product intended to diagnose, cure, mitigate, treat, or prevent disease meets the definition of a drug, regardless of what the label calls it. Supplements are allowed to make limited statements about how a nutrient affects the structure or function of the body, and those statements of nutritional support come with their own disclaimer requirement. What supplements may not do is claim to treat or cure a disease. "Helps lower blood pressure" and "cures anxiety" are disease claims. When a chatbot says them, it has marketed your product as an unapproved drug, in writing, under your brand.

The model does not know the doctrine. It knows the shopper wants reassurance, so it gives the most encouraging answer available, which is usually the one that crosses the line.

Health Claims Need Substantiation

Even claims that stay on the structure-function side of the line have to be truthful and substantiated. The FTC’s authority over deceptive practices reaches unsupported health claims, and the agency has been active about supplement advertising for years. A bot that promises results, invents a benefit, or overstates the evidence is generating exactly the kind of representation that draws scrutiny, and it does it at scale across every conversation.

Dosing and Medical Advice Are Not the Bot’s Job

The shopper also asks the practical questions: how much should I take, can I take this with my medication, is it safe while pregnant? Those are medical questions. A chatbot that hands out a dose or clears an interaction is giving health advice with no clinician behind it, about a person whose situation it cannot see. The honest answer points them to a healthcare professional, not to a confident number from a sales widget.

"Will Not" Is a Suggestion. "Cannot" Is an Architecture.

The instinct is to tell the assistant the rules. Never claim it cures anything. Never give a dose. Never clear a drug interaction. That feels like a boundary, and it is not, because of how the model handles a question phrased differently than the one it was warned about.

You tell it never to make a disease claim. A shopper asks, "will this help with my arthritis pain?" The model hears a friendly question and answers yes, because being encouraging is its default and "arthritis" did not register as the disease term it was warned about. The instruction was loaded the whole time. The wording just slipped past it. A system built to convert keeps converting, and a disclaimer at the bottom of the page does not retract the cure claim a shopper already saved.

Who Answers for It

Strip away the software and the exposure is familiar. A new sales associate who told customers the products cured their conditions, invented benefits, and handed out doses would be a liability the first shift. When the website assistant does the same at scale, the responsibility still belongs to the brand, now with a written record, no regulatory review in the loop, and food, drug, and consumer-protection rules that govern exactly those claims.

The brands that get burned are not the ones that added an assistant. They are the ones that dropped a generic chatbot onto a regulated product, assumed enthusiastic meant compliant, and learned otherwise when a disease claim surfaced in a transcript. The fix is not to remove the assistant. It is to run one that describes products within the limits the law allows, never claims to treat or cure a disease, never gives doses or clears interactions, and routes medical questions to a professional.

Frequently asked questions