The most-quoted number in this field was made up

What the testing found

1. The most-cited statistic about supplement quality was retracted for data fabrication

   Who was studied

   A 2013 DNA barcoding study in BMC Medicine, the source of the widely repeated claim that 59% of herbal products contain plant species not on the label and that only 2 of 12 companies sold clean products

   The paper was retracted in July 2024. The notice states that an investigation by the University of Guelph found evidence of data fabrication and that the editor no longer has confidence in the presented data. Four of the five authors disagree with the retraction and the fifth has not stated a position.

   What this does not show

   That herbal products are fine, or that species substitution does not happen. Substitution is independently documented by other groups using other methods, and the ashwagandha result below is one of them. What it means is narrower and sharper: the 59% figure should not be cited, and any article or brand still citing it has not checked its references since mid-2024. A great deal of supplement journalism, and a great deal of supplement marketing that positions a brand against the industry, rests on that number.

   Review of studiesRetraction Note: DNA barcoding detects contamination and substitution in North American herbal products (2024)

2. Melatonin gummies: 22 of 25 were inaccurately labelled and one contained none at all

   Who was studied

   25 melatonin gummy brands sold in the United States, analysed by liquid chromatography at the University of Mississippi

   88% were inaccurately labelled, with only 3 within 10% of the declared quantity. Actual melatonin ran from 74% to 347% of label, and measured content per serving from 1.3 mg to 13.1 mg. One product had no detectable melatonin but did contain 31.3 mg of cannabidiol. A Canadian analysis of 31 products six years earlier, by a different group with a different method, found content from 83% below to 478% above label and lot-to-lot variation within a single product reaching 465%.

   What this does not show

   A market-wide failure rate, and not a clinical harm. Both samples were small and purposively chosen, and the US study tested gummies, whose mixing problems may not transfer to compressed tablets. One internal detail argues against blaming the format: cannabidiol content in those same gummies was accurate, at 104 to 118% of label. The problem looks specific to melatonin. It matters because a product labelled 3 mg that delivers 13.1 mg is delivering roughly four times a common dose, in a category frequently given to children.

   Study in peopleQuantity of melatonin and CBD in melatonin gummies sold in the US (2023)

3. Two NMN products sold as NMN contained no measurable NMN

   Who was studied

   18 nicotinamide mononucleotide supplements and 5 urolithin A supplements, blinded and randomly coded, measured by triple-quadrupole mass spectrometry with five aliquots per sample

   Deviation from label ran from plus 28.6% to minus 100%. Two NMN capsule products contained no quantifiable NMN, and one urolithin A capsule contained no quantifiable urolithin A. Liposomal NMN formulations did worst, at minus 86.3% and minus 33.4%. Four powder NMN products were accurate within about 7%. Price predicted nothing: NMN ranged from 1.1 to 17.5 euro per gram and urolithin A from 5.8 to 65.4.

   What this does not show

   That NMN does not work. It shows that some products sold as NMN do not contain NMN, which is a different claim and in some ways a worse one, because it means a person can run a careful self-experiment on an empty capsule. The sample was 18 and 5 products bought in one market at two time points, and the study cannot separate manufacturing failure from degradation, which matters because NMN is chemically unstable. A larger industry report reaching a similar conclusion was funded by a company selling the competing molecule, nicotinamide riboside, which does not make it wrong but does need saying.

   Study in peopleTesting the amount of nicotinamide mononucleotide and urolithin A as compared to the label claim (2024)

4. Only 2 of 25 ashwagandha products were confirmed to be made from root, which is what the trials used

   Who was studied

   25 commercial ashwagandha supplements plus raw plant samples, tested for three steroidal lactones against both British and United States pharmacopoeial standards

   Over 44% failed British Pharmacopoeia standards and 60% failed the US standard, with only 10 of 25 meeting both. Only two products were confirmed root-derived. The rest contained varying proportions of leaf and stem, disclosed or not.

   What this does not show

   That 60% of ashwagandha is bad, and the authors say so themselves. A large part of the measured failure is an analytical artefact: the pharmacopoeial method mis-measures one of the markers when leaf material is present, so those failure rates are partly a statement about the test. The finding that survives is the plant part. Ashwagandha trials are run almost entirely on standardised root extract, so if a product is substantially leaf its chemistry differs and the trial evidence does not transfer to it, even though the label says ashwagandha truthfully. This is the cleanest example anywhere of the gap between a compound having evidence and a product having evidence.

   Review of studiesA comparison of British and US pharmacopoeia standards for quality of ashwagandha dietary supplements in commerce (2026)

5. Vitamin D pills from sealed bottles held 9 to 140 percent of their stated dose

   Who was studied

   Individual cholecalciferol pills, five from each of 15 sealed over-the-counter bottles at 1,000, 5,000 and 10,000 IU, plus bottles from the same and different lots, plus compounded tablets

   Over-the-counter pills contained 52 to 135% of the expected dose, and across different lots the range widened to 9 to 140%. Compounded 1,000 IU tablets ran 23 to 146%. Just over half of the over-the-counter pills and about a third of the compounded ones met USP standards. Pill-to-pill variation inside a single bottle reached a coefficient of variation of 28%, although two-thirds of bottles stayed under 10%.

   What this does not show

   That your vitamin D is wrong, or that certification fixes this. The one USP-verified manufacturer in the sample averaged 101.7% with all five pills within 10%, which is encouraging and is also five pills from one company, so it cannot support a general claim about certified products. This is a research letter with a small non-random sample and no clinical outcomes. For vitamin D at ordinary doses a moderate potency deviation probably does not matter much, which is precisely why it is a good illustration: the variation is real and its importance depends entirely on the compound.

   Study in peopleOver-the-counter and compounded vitamin D: is potency what we expect? (2013)

What each seal actually checks

A seal is worth something, but only something specific. Two of the best-known programs test every single batch for banned substances. They make no claim about whether the label is right. Two others check the label, but only once a year. Read the row, not the logo. Each entry below comes from the program’s own published rules.

Two limits apply to all of them. Seals are optional, and the maker pays for them. So a product without a seal is not a bad one. It is an untested one. And a lab report from a seller (a certificate of analysis) describes the batch it came from, not always the batch in your hand.