TDG Programme Blog DH Concierge Book a Call
Health Literacy

How to Check a Health Claim in Five Questions

A reference list makes a product look serious. That is what it is for. But a citation only tells you a paper exists — not that it says what the label implies, not that it was done in humans, and not that it used anything resembling what you are being sold. Five questions separate the two, and they take about two minutes.

STEPHEN DUNCAN FDN-P BSC HONS MSC · DETECTIVE HEALTH · AUGUST 2026

Something has changed in the supplement market over the last few years, and on the face of it, changed for the better.

Products now come with references. Proper ones — journal names, DOIs, author lists. Some companies publish a page per ingredient. It looks like exactly the accountability people have been asking for.

The difficulty is that a reference list produces trust whether or not the papers support the claim. Almost nobody reads them, which the people compiling them know. And most of the time nothing dishonest has happened — the papers are real, the findings are real, and something has quietly gone wrong in the distance between the study and the sentence on the bottle.

Here is how to close that distance yourself. I use the same five questions on my own material, and they have cost me several claims I liked.

Question 1 — Is it in humans?

The single highest-yield question, and often answerable from the title alone.

Mouse and rat studies are genuinely valuable. They are how mechanisms get worked out, and they are frequently the only ethical way to start. But a mouse is not a small person, and the graveyard of interventions that worked beautifully in rodents and did nothing in humans is enormous.

Worked example

Cilantro for heavy metals. This is repeated constantly with real confidence. The supporting work is three rodent studies. There are no human trials at all.

And there is a detail that makes the point sharper: in every one of those studies the coriander was given during ongoing metal exposure. That tests prevention. The clinical claim is that it mobilises metal already stored in your bones and brain — a different question that nobody has asked in an animal, let alone a person.

What to look for: mice, rats, rodent, murine, in vitro, cell line, or a named cell type like MCF-7 or Caco-2. Any of those means the study was not in people.

Question 2 — Is it the same form?

This one catches more claims than any other, because the substance in the study is very often not the substance in the bottle.

Two worked examples

Astragalus and telomeres. The compound in the research is cycloastragenol — a purified, concentrated derivative of astragaloside IV. Astragalus root, tea or standard extract is not that, and there is no evidence any of them delivers a meaningful dose of it. A claim about the derivative is being attached to the herb.

KPV, and a citation that argues against itself. A product I looked at recently cites a paper on delivering KPV via hyaluronic acid-functionalised nanoparticles. That system exists precisely because plain oral KPV does not survive digestion. The product does not use nanoparticles. So the reference offered as support quietly documents the problem the product has.

What to look for: extract versus whole herb, one salt or ester versus another, methylcobalamin versus cyanocobalamin, a purified derivative versus its natural source. And read what the paper actually administered, not what it is about.

Question 3 — Is it the same dose?

Mechanism is dose-independent. Effect is not. A compound doing something interesting at a concentration you cannot reach is not doing it to you.

Worked example — and this is the commonest error in the whole field

Turmeric. The evidence for knee osteoarthritis is genuinely good: sixteen randomised trials, 1,810 adults, effects comparable to NSAIDs with fewer adverse events. I use it and I cite it.

Those trials used standardised extracts at roughly 500–1,500 mg of curcuminoids daily, usually with something added to aid absorption. Culinary turmeric is about 3% curcuminoids by weight and poorly absorbed. Cooking with turmeric is a fine thing to do. It is not the intervention that was tested, and the trial evidence does not transfer to it.

What to look for: the dose in the methods section, then the dose on the label. If the label does not state the standardised active content, that is itself informative.

Question 4 — Is it for the same purpose, and the same outcome?

Two failures hide here, and the second is subtle enough that serious people get caught by it.

The obvious one is purpose drift. KPV reduces experimental colitis in mice. That is a disease model. The product is sold for "comfort and tissue homeostasis as part of a daily wellness routine" — a healthy person, taking it indefinitely. Those are different propositions, and evidence for the first is not evidence for the second.

The subtle one is surrogate endpoints, and it is worth understanding properly because it is the trap that catches the well-informed.

Surrogate versus outcome — the one I got wrong

I used to argue that methylfolate should replace folic acid for preventing neural tube defects. The reasoning felt airtight: 5-MTHF raises red blood cell folate at least as well as folic acid, often better, and avoids unmetabolised folic acid entirely.

All of that is true. It is also beside the point, because red blood cell folate is a surrogate. Preventing a neural tube defect is the outcome. Every trial that established that prevention — the MRC Vitamin Study, Czeizel — used folic acid. And the precise mechanism by which folic acid prevents these defects is still not understood.

When you do not know the mechanism, you cannot assume a different molecule that moves the marker better will move the outcome the same way. If you are pregnant or planning to be, take folic acid at the recommended dose. I was wrong, I have corrected it, and it is the mistake I am most glad to have caught.

What to look for: did the study measure a marker, or something that actually happened to someone? Cholesterol is a marker. A heart attack is an outcome. Both matter; they are not interchangeable.

Question 5 — Who paid, and does the number even exist?

Two quick checks that take seconds.

Funding. Not disqualifying — someone has to pay for research, and often that is the manufacturer. But it belongs in the picture.

Worked examples

Rose hip for joint pain. A meta-analysis pooled three trials, 287 patients, and found a small but real benefit. All three were supported by the manufacturer, and the authors themselves called for independent replication. Eighteen years later it has not arrived. That does not make the finding wrong. It makes it unconfirmed.

And check the number is in the paper. A widely quoted study on descending stairs reports a 13% fall in LDL cholesterol. Chasing it, I found the figure is not in the published abstract at all — it circulates through health media. It turned out to be genuine, restated by the same research group in a later open-access paper. But I could not have known that without looking, and the version circulating was several steps removed from anyone who had read the study.

What to look for: the funding and competing interests statement, usually at the foot of the paper. And if a specific number is being quoted, find it in the source rather than in the article quoting the article quoting the source.

The Pattern Underneath All Five

Once you have run these a few times you start seeing the same shape, and it is almost never fraud.

A mechanism found in cell culture becomes a property of the compound. Then a property of the plant. Then a clinical indication. And by the time it reaches a label, nobody has recorded which step it stopped being true at.

I call it escalation, and every person in the chain may be acting in good faith. The researcher reports what they found. The reviewer summarises. The formulator reads the summary. The copywriter reads the formulator. Nobody lies. The claim simply grows by one degree at each handover, and the citation at the bottom stays technically accurate the entire way.

Which is why the reference list is not the safeguard people take it for. The citation is real. The distance is the problem.

The Two-Minute Version

Ask of any claim
  • Was it done in humans?
  • Was it the same form of the substance?
  • Was it a dose I could realistically take?
  • Was it for my situation, and did it measure an outcome or a marker?
  • Who paid — and is the quoted number actually in the paper?
Phrases that should slow you down
  • “Studies show” with no study named
  • “Clinically proven” with no trial cited
  • “Supports healthy…” anything
  • A mechanism described at length with no outcome given
  • A reference list where every paper is preclinical
  • Confident dosing advice with no dose in the cited study

None of this requires a science degree. It requires reading the methods section, which is usually four paragraphs, and asking whether it matches the sentence that sent you there.

Or Ask, And Check My Working

Plenty of people would rather not do this, which is entirely reasonable — nobody should need a research method to buy a supplement.

AIdan, the clinical concierge on this site, applies exactly these five questions. Ask it about a compound and it will tell you whether the evidence is human or preclinical, whether the studied form matches what is sold, and where a popular claim does not hold up. It carries a corrections block for precisely that purpose.

But I would rather you did not take that on trust either, so here is what sits behind it. Every claim on this site is checked against primary literature before publication, and when something fails, the correction is published rather than quietly removed. That list currently includes the astragalus telomerase claim, ALP as a zinc marker, treating an in-range TSH as a diagnosis, movement screening as injury prevention, and the folate claim above — all things I taught or wrote before checking properly.

That log is the only reason to trust any of it. A practitioner who has never published a correction has either never been wrong, or has never looked.

And if you are weighing up a peptide specifically — the category where all five of these questions fail most often — the Peptide Readiness Assessment runs them for you, compound by compound.

Educational content, not medical advice. Nothing here is a reason to stop a prescribed treatment; if a claim about your medication does not survive these questions, that is a conversation with your prescriber rather than a decision to make alone.

Related

How to Check a Health Claim in Five Questions Peptide Readiness Assessment Food as Medicine — Plants That Work Across Seven Systems More from the blog →

Test, don’t guess

Applies to the claims as much as to the client. If you want a second opinion before spending, a discovery call is free and twenty minutes.

Ask AIdan → Book a discovery call