wyper.io logowyper.io Blog Fact-Check Add to Chrome
Home › How to fact-check a health or pandemic claim

How to fact-check a health or pandemic claim

Health is the field where being wrong costs the most, and where the most confident posts have the thinnest sources.

Fact-Check · last reviewed 2026-08-24

The short answer

To fact-check a health claim, get to the primary study rather than the article about it, then check four things: whether it is peer reviewed, how many people it covers, whether the risk is stated in absolute or relative terms, and whether the conclusion matches what the abstract actually says. Most misleading health posts cite real research and describe it wrongly.

The failure is usually the retelling, not the research

The typical misleading health post is not invented. It points at a genuine paper and then misstates it: a relative risk becomes an absolute one, a mouse study becomes a human finding, a preprint becomes established science, a correlation becomes a cause. Every step feels small and the sum is a false claim with a real citation attached.

This is why retrieval matters more than rhetoric. If the search that supports a check only reaches the reporting, the check inherits the retelling. We measured ours against 24 claims with the deciding document named in advance: 63 percent of everything returned was a primary document, including journals, meta-analyses and datasets, and the deciding root was reached in 23 of 24 cases.

Four questions that settle most of them

  1. Peer reviewed, or a preprintBoth are legitimate, but they carry different weight. A preprint is a manuscript that has not yet been checked by other researchers, and headlines rarely say so.
  2. How many people, and whoTwelve volunteers is a signal, not a finding. Note whether the population resembles the one the claim is being applied to, and whether it was humans at all.
  3. Absolute or relative riskA doubling from one in a million to two in a million is a 100 percent increase and almost nothing. If a post only gives you the percentage, it has chosen the more dramatic of two true numbers.
  4. Does the abstract support the headlineOpen the paper and read the conclusion. This one step disposes of a large share of health claims, because the authors are usually more careful than the people quoting them.

What a truth score can and cannot tell you here

It can tell you whether a specific assertion is supported by the evidence found, with each statement either backed by a sentence located verbatim in the cited source or marked as unsourced. It cannot tell you whether a treatment is right for you, and it does not rule on intent.

It also cannot fix a genuinely open question. Where the evidence is contested, the honest output is the disagreement, not a number that hides it. Full Spectrum is built for exactly that case: counter-positions, the spread of sources, and the questions still open, with a Triple-AI reading over one shared evidence set.

This is verification of claims, not medical advice. For a decision about your own health, the source is a clinician who knows your history.

Why source diversity decides outcomes in this field

Health claims travel across languages faster than almost anything else, and a check that only reads English is blind by design. Our measurement included non-English roots on purpose and reached them. Sources are never ranked by origin: no bonus for legacy media, no preference for official bodies, and no shortcut through other fact-checkers, who accounted for 1 percent of everything returned.

That rule cuts both ways deliberately. A ministry statement carries no automatic advantage over a peer-reviewed paper that contradicts it, and a widely shared blog carries no automatic disadvantage if it cites the primary document correctly. What counts is what the source is, not who published it.

Questions

What is the difference between a preprint and a peer-reviewed study?
A preprint is a manuscript the authors have published before other researchers checked it. It can be entirely sound or badly flawed, and nothing about the format tells you which. Peer review is a filter, not a guarantee, but its absence is worth knowing when a headline treats a preprint as settled.
Why does absolute versus relative risk matter so much?
Because both are true and they feel completely different. A rise from one in a million to two in a million is a 100 percent relative increase and a one in a million absolute increase. Reporting only the relative figure is the most common way to make a small effect sound alarming.
Can wyper check a scientific paper?
It can check a claim against sources including journals and datasets, and it verifies whether a quoted sentence really appears in the document credited with it. It does not replace reading the paper for a decision that matters, and it does not give medical advice.
How do I check a health claim in another language?
Search for the primary document in its original language rather than for coverage of it in yours. Our own retrieval measurement deliberately included non-English roots, and the verdict comes back in your language across twelve of them.
What if the evidence is genuinely unclear?
Then the correct answer is that it is unclear, and a tool that hides that behind a confident number is doing you harm. Full Spectrum reports the counter-positions and the open questions instead of resolving them artificially.

Check a claim and look at the evidence yourself

Free: 5 checks a day, no account needed. Every statement tells you whether a quote backs it.

Open the web app Add to Chrome, free

Read on

Ask any fact-checker for their error rateRetrieval, quote verification and model agreement, measured with raw data. How to verify war footageWhere automation helps in a crisis and where it has to stop.