Research Guide · Evidence & Claims

How to Read a Health Claim

A supplement label can say almost anything about how a product "supports" your body without ever proving it works. This guide answers the most common questions about evaluating a health claim, synthesized from the reporting in What the Label Doesn't Tell You.

Why do so many people assume supplements are already tested?

Survey after survey finds the same pattern: most Americans believe, reasonably, that a product sitting on a shelf next to regulated medicine has cleared some kind of government safety and effectiveness check before it got there. Most of the other products on that shelf did. The supplement sitting between them, in most cases, did not — a gap that has nothing to do with what's inside the bottle and everything to do with a legal classification decision made in 1994.

What does "clinically studied" actually mean?

Nothing specific, and that's the honest answer. The phrase carries no legal definition and no minimum evidentiary bar — it can describe a small, unpublished, industry-funded pilot study on a dozen people, or a large, independent, peer-reviewed randomized trial. The label won't tell you which.

What's the difference between an anecdote and evidence?

A personal story — "it worked for me" — can be entirely genuine and still not establish that a supplement caused the improvement. Four ordinary explanations can produce the same story: confirmation bias, the placebo effect, regression to the mean, and the natural fluctuation of symptoms and disease. Anecdotes are useful for generating a hypothesis worth testing. They cannot establish that a product actually works.

"Anecdotes can generate hypotheses. They cannot establish effectiveness."What the Label Doesn't Tell You

What does a scientifically rigorous claim actually look like?

Evidence sits on a hierarchy, not a single flat category. Laboratory studies and animal studies sit at the base — useful for a possible mechanism, many steps removed from a human body. Observational studies find real associations without ruling out other explanations. Randomized controlled trials, which assign people by chance to a product or a placebo, can rule out most of those other explanations. Systematic reviews, combining many trials, sit at the top.

Evidence typeWhat it can tell you
Testimonial / anecdoteA hypothesis worth testing — nothing more
Laboratory / animal studyA possible mechanism, untested in humans
Observational studyA real association, with confounders unresolved
Randomized controlled trialWhether the product itself likely caused the effect
Systematic review / meta-analysisWhat the full body of trials shows, not just one

Can a plausible biological mechanism be wrong?

Yes — and one of the clearest documented cases shows exactly how. In the early 1990s, researchers had strong reason to believe beta-carotene, an antioxidant, would protect smokers from lung cancer: the biochemistry was sound, and population data was encouraging. Two major randomized trials tested the hypothesis directly on tens of thousands of smokers.

The result ran in the opposite direction. One trial found an 18 percent increase in lung cancer diagnoses and an 8 percent increase in deaths among participants given beta-carotene — a pattern serious enough that both major trials stopped their beta-carotene arms early. A later review found an overall 20 percent increased risk, concentrated most heavily among smokers.

"The mechanism didn't predict the outcome. The trial produced a surprising outcome, and the mechanism had to be revised to explain it."What the Label Doesn't Tell You

Why should a reversed medical recommendation increase trust, not decrease it?

The beta-carotene story is, in the end, a story about science doing exactly what it's supposed to do. Nobody involved was dishonest — the hypothesis was reasonable, the trial was run at real cost specifically to test it, and the recommendation changed the moment the evidence demanded it. A system incapable of that kind of correction would be far more dangerous than one willing to admit it was wrong.

Every figure on this page is drawn from the sourced reporting in What the Label Doesn't Tell You. See the book for full citations.