The State of Supplement Claims
No regulator approves a supplement claim before it goes live. The market runs on self-policing, so the only question that decides anything is who actually reads the copy. For most brands, the first careful reader is a platform’s machine, after the copy is live. We read the best-selling listings in eight major Amazon US categories and ruled every phrase against all 109 standards, in a full September 2026 re-scan and a 28 Sep supplementary pass: 2,243 listings, 793 brands, more than 244,000 phrase checks. 76% received an outright FAIL verdict, and 42% of the brands behind them carry at least one Critical finding, a named-condition or disease claim that fails 21 U.S.C. §343(r)(6) outright. One in four listings carries a disease-type claim. Most of the rest can be fixed with a sentence or a citation.
Method · every phrase ruled against FDA structure/function, FTC substantiation and marketplace policy · pharmacist-signed methodology · brands never named in public data · free to cite with attribution
Executive summary
Three in four top-selling supplement listings fail outright.
76% of the listings we ruled received a FAIL verdict against the published standards. These are not fringe products. They are the bestsellers in sleep, gut, stress, immune, cognitive, joint, sports and energy.
Two in five supplement brands carry a claim that can take a listing down.
42% of scanned brands, 330 of 793, carry at least one Critical finding: a named-condition or disease claim that fails 21 U.S.C. §343(r)(6) outright and can draw enforcement without further notice.
The single most common failure is also the cheapest to fix.
The number one finding in our scan data is not an exotic claim. It is a sentence that costs nothing to add: the FDA disclaimer, missing at 455 of the 793 brands. Close behind are an unsupported superlative (“best”, “most effective”) at 404 brands and a Non-GMO claim with no document behind it at 398. The industry’s most widespread compliance failures are a sentence that costs nothing, a word that needs support and a claim that needs a document.
These are not fringe products. They are the listings with the most revenue at stake, the most resources to get it right, and the most to lose when a marketplace switches them off. When three in four of those fail, the explanation is not that the industry is dishonest. It is that nobody was checking.
Method and data
Listings enter the database through the same engine that runs our paid audits. Each listing is ruled, phrase by phrase, against all 109 standards (98 in the September 2026 re-scan, 11 added on 28 Sep 2026 in a supplementary pass), which is built on FDA structure/function rules, FTC substantiation guidance and Amazon’s supplement claims policy. These are rulings, not scraped copy: every phrase gets a verdict and a severity of Critical, High, Medium or Low. Every finding must quote the page word for word, and any quote the software cannot find on the page is discarded before it enters published figures. Paid audit verdicts are signed by a licensed pharmacist. Since 28 Sep 2026 each category also holds every ruled listing from Amazon’s own Best Sellers lists for that category. Aggregate figures update as listings are added, and individual brands are never named in public data. The full method and the standards behind it are documented on the Standards page.
The nearest prior work measured what is inside the bottle rather than what the listing says. A 2023 JAMA analysis found 88% of melatonin products mislabeled. This report measures the other half of the package: the claims.
The verdict split
A FAIL verdict means the listing carries at least one Critical or High finding, a phrase that breaks a published standard seriously enough that the listing as a whole does not pass. FIX THEN PASS means the only findings were Medium or Low. PASS means no findings at all.
Every ruled listing by verdict · 2,243 listings
A FAIL is not a takedown, and it says nothing about whether the product is safe or works. Every listing here was live and selling when we read it. This report measures marketing language, not chemistry, and the language is what a platform, a payment processor or a regulator can act on when it looks.
Percentages rounded to the nearest point. Counts are exact.
How verdicts work
A listing fails on its worst line. Most fails are not disease claims. They are proof a listing does not show, or a sentence it forgot to add. Every one can be fixed without losing the sale.
534 listings. The class a platform can remove a listing for, and the FDA can write a warning letter about.
755 listings. "Clinically proven" with no study named, or a promise the rules bar. The kind that can get an ad disapproved.
411 listings. Mostly the missing FDA disclaimer. The cheapest fix in the file, one sentence.
The rest either pass clean or need only Medium or Low fixes to pass.
| Severity | What it means | A real example | What it can cost |
|---|---|---|---|
| Critical | Names or implies a disease, or compares the product to a drug. | → “Promotes a sense of calm” | The listing can be removed and the brand can draw a warning letter. Of the 196 FDA warning letters to supplement companies we read in full (2017 to 2026), 106 cite exactly this: marketing that makes the product a drug. |
| High | Breaks a published rule outright: proof not shown, or a required line missing. | → name the study, or say what it supports | An ad can be disapproved and a listing can be suppressed on review. |
| Medium | Cannot be backed as written. | → hold the document, or drop the badge | A competitor or a reviewer can challenge it, and you have nothing to hand over. |
| Low | Wording to tighten. No rule broken on its own. | Rare: fewer than 1 in 100 findings in the database rule Low. | Little on its own. It adds up across a page a reviewer reads. |
FAIL means at least one Critical or High line. FIX THEN PASS means Medium or Low only. PASS means nothing found.
A FAIL is not a takedown. It means the page carries a line that Amazon, an ad network, a payment processor or a regulator can act on the day it looks. Every listing here was live and selling when we read it. Most simply have not been looked at yet. Until then the line keeps selling. After, it can stop the sales it was making. Fixing it first is how you keep them.
Fail rate by category
Compliance is not evenly distributed. Stress and gut and digestion are the lowest-compliance categories we scan, at 82% and 81% fail rates against 76% across all eight. Both invite condition-shaped language: stress and cortisol sit next to named anxiety disorders, digestion, bloating and regularity next to named GI conditions in a copywriter’s mind, and the listings show it.
The right-hand column is the share of that category’s brands carrying at least one Critical finding. Gut and digestion leads there, at 54% against 42% overall. Category rows count every listing ruled in that category, so they total 2,359 against 2,243 distinct listings: 109 listings sell into more than one category and are ruled in each.
Fail means at least one Critical or High finding on the live listing.
The phrase table
This is the centre of the report. If the failures were the work of a few bad actors inventing wild claims, the table below would be a long tail of oddities. It is not. It is ten phrases, repeated across hundreds of unconnected brands, ranked by how many carry each one. Struck phrases are quoted from live listings. The brands are not named. Their copy names them.
Counts are distinct brands, among the 793 ruled, that carry the phrase or a direct variant. Bars are scaled to the top row.
Two patterns here are worth naming. First, three of the top five findings are paperwork, not science: a required disclaimer that is missing, a Non-GMO claim with no document behind it, and a certification badge that cannot be verified from the listing. Second, the rest are the ordinary vocabulary of supplement marketing. “Best”, “boosts immunity” and “balances hormones” are not written by rogue copywriters. They are the default language of the category, and the default language fails.
Inside the findings
Across the 2,243 listings we logged 11,446 individual findings. Critical findings are the most dangerous: 977 of them, clustered in 330 brands rather than spread evenly across the 793.
Every finding logged, by severity · 11,446 findings
The dated record
The numbers above would matter less if enforcement were theoretical. It is not, and it has accelerated on several fronts.
Four institutions with four different incentives now read supplement listings the same way, and the platforms read with machines: Amazon’s 2025 Trustworthy Shopping report describes AI scanning billions of attempted changes to product detail pages daily. The window where a risky phrase was somebody else’s problem has closed.
Sources: FTC penalty schedule and Notice of Penalty Offenses, FDA warning-letter database, Amazon seller policy, public litigation record.
Guidance
This is general information, not legal advice. The data does suggest a clear order of operations.
Fix the paperwork first.
The most common failure in our scan data is paperwork, not language: a missing FDA disclaimer, which is required alongside structure/function claims, costs nothing, and is absent at 455 brands. Third on the list is paperwork too: a Non-GMO claim with no document behind it, at 398.
Hunt named conditions.
Two in five brands carry a claim that mentions or implies a disease. These are the findings that can end a listing or start a lawsuit. Search your own copy for condition names, and for their synonyms and abbreviations, everywhere: product page, packaging renders, review responses, backend keywords.
Audit your certifications.
“GMP certified” appears on the listings of 284 brands. When the certification cannot be verified, the badge is itself a claim at risk. If you display it, be able to document it.
Treat default marketing language as suspect.
“Boosts immunity”, “clinically proven” and “balances hormones” read as normal because everyone uses them. Everyone using a phrase is not evidence that it is compliant. It is evidence that the fail rate is 76%.
Reread the detail page against the marketplace policy, not just the label.
Amazon prohibits disease and treatment claims on supplement detail pages under its Restricted Products policy, and that covers backend keywords as well as the copy a shopper reads. Most listings were written against the label alone.
Pharmacist-in-the-loop
Is your copy in this data?
The audit answers in five business days: every claim, every channel, pharmacist signed.
Cite this report
Citing this report
Every figure on this page is free to cite with attribution to “Claims Verified, The State of Supplement Claims, Q3 2026” and a link to claimsverified.org/research/state-of-claims/. Figures reflect the September 2026 database snapshot and update as scanning continues. For category-level breakdowns, methodology questions or interviews, email us.
audit@claimsverified.org →Media and methodology questions: audit@claimsverified.org · © 2026 Claims Verified · Data from the claims database, snapshot 27 Sep 2026 · Reproduction with attribution permitted · Cite as “Claims Verified, State of Supplement Claims, Q3 2026” · Not legal advice
See where your own listing sits. One Amazon listing or your website, checked free. No call, no account.
Check my listingSay the most the rules allow. Then keep selling it, everywhere, at any spend.
Clean copy can run on every channel at once with nothing for a machine to switch off, so you can push budget knowing what a reviewer can find. And it is the brand a careful shopper can believe.