How we rate health claims
Every claim is rated with the same published method, so the same claim always gets the same verdict.
1. Evidence grade A–D
Each claim is graded by the highest-quality evidence that supports or refutes it, following E-GRADE, a GRADE-based framework developed by Korea's National Cancer Center to evaluate medical claims on YouTube.
| Grade | Certainty | Evidence accepted |
|---|---|---|
| A | High | Systematic reviews, meta-analyses, guidelines from recognized bodies |
| B | Moderate | Randomized trials and large, well-designed cohort studies |
| C | Low | Small studies, case reports, animal or cell studies, mechanisms |
| D | Very low | Anecdotes, testimonials, unsupported claims |
Evidence not from human studies is capped at grade C.
When we grade down (as in GRADE): even top-level evidence is lowered one grade if it (1) addresses the claim only indirectly, or (2) pools observational studies rather than randomized trials. Example: a meta-analysis of observational studies → B. We state the reason in each verdict.
2. Five verdicts
| Verdict | Meaning |
|---|---|
| ● True | High-quality (A) evidence supports the claim as stated. |
| ◐ Mostly true | Supported, but the effect or population is narrower than claimed. |
| ? Insufficient evidence | Not yet confirmed in people. This does not mean "false". |
| ✕ False | Reliable studies (A or B) show the opposite. |
| ! Harmful | Following it could cause harm, such as delaying treatment. |
We do not issue verdicts that contradict public health authorities (WHO, national agencies). When new research differs from guidance, we present both.
3. Process
- Collect questions people actually ask and claims that are spreading. We never single out a specific creator or person.
- Search guidelines, then systematic reviews and meta-analyses, then trials, then observational studies.
- Assign a grade and a verdict, and explain why in plain language. Steps 1–3 are drafted by AI (Claude).
- Publish only after the verdict passes automated evidence checks: every cited paper exists in PubMed with matching DOI, author and year; none is retracted; every number appears in the abstract; the grade follows section 1; and a second, independent AI re-checks the wording against the abstracts.
- High-risk verdicts ("Dangerous", anything about medicines, treatment or stopping treatment, or anything flagged by the checks) are published only after the operator (public health PhD, RN) approves them.
Each verdict shows how it was checked.
4. Journals we monitor
Each month we review MEDLINE-indexed journals that are listed in Web of Science (SCIE/SSCI) and rank near the top of their field, including NEJM, The Lancet, JAMA, BMJ, Annals of Internal Medicine, Nature Medicine, the Cochrane Database of Systematic Reviews, The Lancet Public Health, and the American Journal of Public Health.
We cannot review every journal or every study. Research published outside our list, or after our latest check, may not be reflected. If we missed something, tell us and we will review it and log the change.
5. Verdicts are maintained
- A monthly evidence update is published, including months with no changes.
- Every verdict shows its last review date. Changes are kept as history, never deleted.
- We separate corrections (we were wrong), verdict changes (new research changed the answer), and evidence updates (new studies, same answer).
- If a cited paper is retracted or official guidance changes, we re-review within 72 hours.
6. Independence
We do not accept sponsorship or advertising from products we might rate (supplements, devices, etc.), and we do not recommend specific clinics, pharmacies, or products.