Skip to content

How we work

The State of Supplement Evidence, November 2026: Most Re-checked Findings Rest on Limited Evidence

As of 17 November 2026, most supplement findings we have re-checked rest on limited evidence. Of 2,703 graded findings in our re-checked subset (rows checked against their studies 8 to 17 November 2026, not the whole site), 71.4% are grade C or D and 5.4% are grade A. Here is how we counted, and what a grade does and does not mean.

As of 17 November 2026. This is a dated methodology report. Its numbers describe only findings we have re-checked against their studies, are computed from our data as of that day, and will change as the checking goes on.

What we found

  • Most re-checked findings rest on limited evidence. Of the 2,703 graded findings in our re-checked subset (rows checked against their studies 8 to 17 November 2026, not the whole site), 1,930 (71.4%) are grade C or D. 147 (5.4%) are grade A, and 773 (28.6%) are grade A or B.
  • That does not mean most supplements fail. Grade C or D means the research is thin, so we can't be sure either way.
  • Many rest on one study. 36.1% of the same re-checked subset of 2,703 findings rest on a single linked study; among its grade C and D findings it is 47.9%.
  • Many found no effect. 43.1% of the same re-checked subset of 2,703 findings record no clear effect. Even among its grade A and B findings, 25.5% do: a strong grade can mean strong evidence that nothing changed.
  • Our guides look stronger, likely because they lead with the best-studied options. Counting the checked, graded claims in our published guides too, once each, gives a combined 3,499 re-checked claims, of which 31.9% are grade A or B, a little more than in the database subset alone.

How we counted

What a finding and a grade are. A finding pairs one supplement with one outcome, such as a supplement and blood pressure. Its grade, A to D, rates the published human research behind it, as how we grade defines it. A, strong evidence: several well-run human trials, or meta-analyses of them, point the same way. B, moderate evidence: good human studies support the finding, but there are fewer of them or they agree less. C, limited evidence: a small number of studies, or studies with clear weaknesses, so the finding could change. D, very limited evidence: early research only, not enough to draw a conclusion. A grade does not rate a product, a brand or an amount, and it does not say a supplement will work for you.

Only re-checked findings, and only a subset. We counted a database finding only if its grade, its verdict and its list of studies were checked against the cited abstracts on a recorded day by 17 November 2026, and it is not on hold for a correction. By then 10,650 rows had been checked; 2,703 of them carry a grade, and the other 7,947 were checked and left without one. A checked finding carries no grade when there is not enough human evidence to give one: only studies in people count, and studies in animals or cells can stay listed but don't earn a grade. Those 2,703 graded rows, checked 8 to 17 November 2026, are the subset this report describes. They are a subset, not the whole site: findings outside it are not counted here.

Guide claims, as a second set. In our 96 published guides, a graded claim counts when every study it cites is in that guide's record of checked citations. That gives 1,023 claims (46.5% grade A or B); 0 were left out because a cited study was missing from the record. When a guide claim has the same grade as a database finding and cites one of its studies, it is counted once (227 claims), which gives the combined 3,499, of which 6.4% are grade A.

What we left out. Findings not yet re-checked are not counted anywhere in this report, and neither are our rewritten data pages, whose evidence carries no letter grade our script can read. Tables by condition are left out of this post.

The snapshot. The figures on this page are computed from our evidence database each time the site is built, counting rows checked on or before 17 November 2026, so a later correction to one of those rows shows up here. They follow the same rules a separate script uses to check them. Shares are rounded to one decimal place, so they may not add up exactly. We count findings and claims, not people or products.

How strong is the evidence behind re-checked findings?

Counted over the 2,703 graded findings in our re-checked subset (rows checked against their studies 8 to 17 November 2026, not the whole site).

GradeRe-checked subsetShare
A (strong)1475.4%
B (moderate)62623.2%
C (limited)1,14342.3%
D (very limited)78729.1%
All graded2,703

Do the most-searched supplement pages have strong evidence?

We took the 20 supplement pages shown most often in Google search results from 19 October to 15 November 2026 and looked up each one's best grade among its re-checked findings only. 7 of them have a re-checked finding graded A or B, 11 reach only C or D, and 2 have no re-checked grade yet. A best grade can belong to any outcome, including a finding of no effect.

What safety information do pages carry?

Each reviewed supplement page lists its safety points, such as side effects, drug interactions and who should take care. Guides, comparison pages and tools that recommend or offer a supplement carry its shared, audited warning word for word.

What do we leave out for research integrity?

Some papers never count toward a grade, a direction or a summary. Our integrity list holds 2,967 PubMed records: papers by authors on our research-integrity list (matched by author name, so the list errs toward leaving papers out), including 160 retracted papers listed one by one. We don't name the authors here. Since the check began on 7 November 2026, 1,091 study links were taken off findings where the reviewer's reason cites a retraction or an integrity concern, changing 313 findings. That count leaves out any link that was later added back. How we handle retractions is in a separate post.

  • How we work

    How we handle retracted research

    A study can be retracted years after other papers have built on it. This post explains what a retraction and an integrity concern are, which papers we never let support a grade, why a meta-analysis that leans on a retracted trial can't carry a claim, and how you can check our fixes in the corrections log.

    Published

  • How we work

    Why some protocols have no core supplement

    Some of our protocol pages now open by saying no supplement has been shown to help, and their stack has no core. This post explains what a core supplement is, why a page can have none, why those pages carry no buy links, and how a page can gain a core later.

    Published

  • How we work

    Why some supplements show 'Studies not yet linked'

    Some findings on our supplement and condition pages now say 'Studies not yet linked' where they used to show 'Improves' and an arrow. This post explains what the label means, why we stopped showing a benefit with nothing behind it, how a study gets checked and linked to a finding, and what the label does not mean.

    Published

All posts