How a check works
- Your claim becomes a search. An AI model turns "apple cider vinegar cures diabetes" into the terms researchers actually use.
- PubMed answers. That's the U.S. National Library of Medicine's index of medical research, 37 million papers. We take up to 8 of the best matches with their abstracts.
- The studies get weighed, not counted. A meta-analysis of many trials outranks one small pilot. A case report about one patient barely counts. The model reads the abstracts and says which studies it relied on.
- The verdict is stamped, with rules. "False" needs studies that contradict the claim. Nothing cited means "it's complicated," never a confident answer. Studies that only match the words don't count.
- Every verdict says what's still open. Research is rarely finished. The last line names the gap: who the studies left out, how short they ran, or what finding would change the answer.
Reading the evidence meter
- Strong: the verdict rests on a meta-analysis, systematic review, or randomized trial.
- Moderate: human studies or expert reviews, but no randomized trial behind it.
- Weak: only case reports, letters, or opinion pieces.
- No direct evidence: nothing found actually tests the claim.
Not medical advice. The model can misread an abstract, and PubMed doesn't index everything. Read the studies. Talk to a clinician about your own health.