[Company name changed. Based on a composite of enforcement actions in our corpus.]
On January 9, 2026, the FDA's Office of Regulatory Affairs mailed a Warning Letter to a company we are calling Sunshine Botanicals LLC, a contract packer operating out of a 4,000-square-foot facility in Clearwater, Florida. The company made elderberry and zinc products: gummies, capsules, and a liquid concentrate sold under three private-label brands, none of which most consumers would recognize.
The letter was four pages. It cited six claims. And within four weeks, it quietly rewrote the rules that AI engines apply to the word "immunity" in supplement marketing.
What the letter said
Sunshine Botanicals' products carried standard immune-support language: "boosts your immune system," "strengthens immune defense," "provides immune support during cold and flu season." These phrases are common across the supplement industry. Variations appear on thousands of product pages. The first two have historically been treated as permissible structure-function claims by many brands, though "defense" language has drawn scrutiny in certain contexts.
The third phrase is the one the FDA targeted.
The agency's reasoning was precise. By linking immune support to "cold and flu season," the claim implied that the product could prevent or mitigate specific diseases (cold, influenza). The temporal framing transformed a permissible structure-function claim into an implied disease claim. The letter cited 21 CFR 101.93(g)(2)(iv), which addresses claims that are implied disease claims because of the context in which they are made, and the 2002 FDA Guidance for Industry on structure-function claims, which states that a claim becomes a disease claim when it "implies an effect on a disease through the use of descriptions of the product's mechanism of action."[1]
The key distinction: linking immune support to a named disease season ("cold and flu season") crosses the line because it implies protection against specific diseases. Generic temporal references ("during seasonal changes") without naming diseases sit in different territory, though Post 9 in this series explores why even that softer phrasing now carries more risk than it did before 2026.
The remaining five cited claims were more obviously over the line: "fights viruses," "kills bacteria naturally," "proven to reduce cold duration." Those would have drawn a letter in any year. The seasonal-framing citation was the new ground.
What happened next, in the engines
We run a weekly panel of 120 immune-support queries across ChatGPT, Perplexity, Claude, and Gemini. The panel includes both direct product queries ("best elderberry supplement") and informational queries ("does zinc help with colds"). For each answer, we record the presence or absence of regulatory language, disclaimers, cautions, and source citations.
In the four weeks before January 9, an average of 31% of immune-support answers across the four engines included any form of FDA disclaimer or regulatory caution. That number was consistent with our six-month baseline.
In the four weeks after January 9, the rate jumped to 58%.
The increase was not uniform. ChatGPT moved first, with its disclaimer rate rising from 29% to 61% by the week of January 20. Perplexity followed a week later. Claude's rate climbed more gradually, reaching 54% by February 6. Gemini moved last but moved furthest, from 33% to 67%.
Here is the part that matters: the language the engines used in their new disclaimers did not match the language in the Warning Letter. The FDA's letter cited "cold and flu season" as the trigger for an implied disease claim. The engines began warning users about seasonal immune claims more broadly, using phrases like "claims about seasonal immune support have been questioned by regulators" and "the FDA has raised concerns about immune-boosting language tied to specific seasons or illness periods."
Those phrases do not appear in the Warning Letter. They do not appear in any FDA guidance document we have been able to locate. They appear to be the engines' own synthesis, generated by models that absorbed the enforcement signal and extrapolated a broader rule.
The engines did not quote the FDA. They interpreted the FDA, then presented their interpretation as though it were the FDA's position. That is a new kind of regulatory propagation, and the supplement industry is not tracking it. EvidenceSignal Research
The propagation mechanism
How does a Warning Letter mailed to a small Florida packer reach the training data or retrieval index of four AI engines within a month? Three channels, working in sequence.
Channel 1: Direct indexing. All FDA Warning Letters are published on fda.gov within days of issuance. AI engines with web-crawling capabilities (Perplexity, Gemini with Google Search, ChatGPT with browsing) index these pages. The Sunshine Botanicals letter appeared in Google's index within 72 hours of publication.
Channel 2: Trade press amplification. NutraIngredients-USA and Natural Products Insider both ran brief items on the letter within a week, focusing on the seasonal-framing precedent. These articles were indexed by all four engines and appeared in Perplexity's cited sources by January 18.
Channel 3: Model generalization. This is the channel that creates the gap between what the FDA said and what the engines say. Once the enforcement signal enters the retrieval window, the model generalizes. It does not limit its caution to the specific phrase "cold and flu season." It extends the principle to all seasonal immune-support claims, and sometimes to all immune-support claims regardless of seasonal framing.
The third channel is where the distortion occurs. And it is invisible to the brand whose product page still reads "supports healthy immune function," a claim the FDA has not challenged, but which the engines may now treat with suspicion because it sits in the same semantic neighborhood as the cited language.
What this means for immune-support brands
Our index currently tracks 1,305 FDA-cited claims extracted from 223 confirmed supplement Warning Letters.[2] The Sunshine Botanicals letter added six claims to the index. But its real impact is in the precedent it set for how engines process enforcement signals.
Three practical implications:
First, seasonal framing is now a risk factor. Any product copy that ties a supplement benefit to a named disease season ("cold and flu season," "flu season") should be reviewed against this precedent. The FDA drew the line at disease-naming seasonal framing. The engines drew it wider. Note that softer temporal language ("during seasonal changes") without naming diseases occupies different ground, but the engines may not always make that distinction. See Post 9 for more on this.
Second, the engines' interpretation does not match the agency's statement. This creates a compliance gap that brands cannot close by reading the Warning Letter alone. You need to see what the engines are actually saying. That requires testing your queries, weekly, across all four engines.
Third, the lag is four weeks. From letter issuance to engine behavior change, we measured roughly 28 days. That is the window a brand has to audit and adjust its copy before the engines begin applying the new standard to its products.
In our corpus of FDA Warning Letters, seasonal immune framing had not previously been cited as the basis for an implied disease claim in the supplement category. The Sunshine Botanicals letter is, to our knowledge, the first.[4] It will not be the last.
Footnotes
- FDA, "Guidance for Industry: Structure/Function Claims, Small Entity Compliance Guide," January 2002. See Section IV.B, "Claims That Suggest Disease Prevention or Treatment." The final rule on structure-function claims was published January 6, 2000 (65 FR 1000); the 2002 guidance implements it.
- EvidenceSignal corpus statistics as of May 23, 2026: 6,717 total Warning Letters indexed; 223 confirmed supplement-related; 1,305 individual cited claims extracted and classified under DSHEA taxonomy. Full methodology at /methodology.
- EvidenceSignal's weekly AI engine monitoring tracks how each engine responds to a brand's core product queries across ChatGPT, Perplexity, Claude, and Gemini.
- We searched the full corpus for prior Warning Letters citing seasonal or temporal framing as the basis for an implied disease claim in the supplement category. No prior instance was found. Adjacent precedent exists in the homeopathic category (see 2019 Warning Letter to Standard Homeopathic Company, re: "flu season" claims for Oscillococcinum), but the regulatory framework differs.