How grading works
We use a precision evaluation system which includes GRADE (Grading of Recommendations, Assessment, Development and Evaluation), the method used by the World Health Organization, Cochrane and hundreds of guideline groups. GRADE rates our certainty in an answer, not how exciting a study sounds. Then the best human experts in the field review the answers, weigh in with their judgment and sign off.
1. Frame the question
Every answer starts with a PICO: the Population, the Intervention or exposure, the Comparator and the Outcome. Certainty is rated for an outcome across all the relevant studies, not for one paper.
2. Pick a starting level
Randomized trials start at High. Observational studies start at Low. Animal, cell and mechanistic studies sit at Very Low for human outcomes; they can generate ideas but cannot prove an effect in people.
3. Rate down for problems
Five domains can lower certainty by one level (serious) or two (very serious): risk of bias, inconsistency between studies, indirectness (for example a lab marker instead of real events, or mice instead of people), imprecision (few events, wide confidence intervals) and publication bias.
4. Rate up when it is earned
Observational evidence can move up for a very large effect, a dose-response gradient, or when confounding would only shrink the effect.
5. The four levels
| High | We are very confident the true effect is close to the estimate. "We know." |
| Moderate | The true effect is probably close, but could differ substantially. "Probably." |
| Low | The true effect may be substantially different. "May." |
| Very Low | The true effect is likely substantially different. "We really don't know yet." |
6. AI first, experts next
Initially answers are drafted by an AI engine that reads live PubMed records and applies these rules. Every study link is pulled from PubMed, never invented. These drafts are labeled as not yet human reviewed. The most asked questions are reviewed and signed off by independent experts, and reviewed answers replace the AI drafts with a green ✓ Human expert reviewed mark, so you can always tell a human-reviewed answer from an AI draft.
Sources: GRADE Handbook (Schunemann et al., 2013); GRADE guidelines series, J Clin Epidemiol 2011 onward; Cochrane Handbook, chapter 14.