How Our Data Is Made
The pipeline behind every rating — and its limits.
We separate what an expert claimed from what the research can support, and we try to be transparent about exactly how that happens — including where the process relies on AI.
1. Expert claims
We extract specific supplement claims from publicly available podcasts, videos, and transcripts using AI (Anthropic’s Claude). Every claim is attributed to the expert who made it and linked back to its source, and carries an extraction-confidence rating (high / medium / low) so you can see how cleanly it was captured.
2. Research retrieval
For each supplement we pull studies from PubMed / NCBI, prioritizing human randomized controlled trials, systematic reviews, and meta-analyses over animal or mechanistic work. Every study links out to its source, with sample size, population, funding, and disclosed conflicts of interest shown where available.
3. Evidence comparison & scoring
Each claim is compared against the research and labeled (supported, partially supported, insufficient, and so on). Supplements get an overall evidence rating and a 0–100 quality score from our published methodology.
Limits you should know
- · AI extraction can misattribute or misphrase a claim. That’s why every claim shows an extraction-confidence rating and links to its original source — so you can verify it yourself.
- · Evidence ratings reflect the depth and consistency of research, not whether a supplement is right for you.
- · We are actively expanding credentialed human review of the AI-extracted claims. Where a page has been reviewed by a qualified professional, that is shown on the page; where it hasn’t, we don’t claim it has.
- · Found something wrong? Report a correction — we fix confirmed errors.
See also our scoring methodology and editorial independence statement. This is educational information, not medical advice.