How we evaluate the evidence

We separate two things that get mixed together: what an expert claimed, and what the published research can actually support. Here’s exactly how that works.

The approach

We extract specific supplement claims from public source material — podcasts, videos, and transcripts — and attribute each to the expert who made it, with a link back to where they said it. Each claim is then compared against research indexed in PubMed, preferring human trials, systematic reviews, and meta-analyses over animal or mechanistic studies.

Claims are extracted with AI (Claude) and reviewed. An extraction-confidence rating (high / medium / low) reflects how cleanly each claim was captured from its source.

The evidence quality score (0–100)

Every supplement gets a 0–100 score and a letter grade, built from four equally weighted components worth 25 points each. It measures the quality and consistency of the evidence base — not whether a supplement is right for you.

Evidence depth

0–25

Scales with the number of PubMed studies referenced for the supplement. 20 or more studies earns full marks.

Study quality

0–25

Rewards the gold standard: up to 12.5 points for the share of randomized controlled trials, and 12.5 for the share of systematic reviews and meta-analyses.

Expert consensus

0–25

Up to 15 points for how often tracked experts mention it (50+ mentions = full) and 10 points for how many distinct experts do (5 experts = full).

Claim support

0–25

The share of assessed expert claims that published research supports or partially supports.

A80–100Excellent

B65–79Good

C50–64Fair

D35–49Limited

F0–34Insufficient

Evidence levels

Alongside the numeric score, each supplement carries a plain-language evidence label. Labels are a shorthand for research depth and consistency — not a recommendation.

🟢

Strong — Multiple RCTs or systematic reviews in humans with consistent findings

🟡

Moderate — Some RCT evidence or consistent observational data

🟠

Weak — Mostly animal / mechanistic or small pilot studies

🔵

Mixed — Conflicting results across multiple studies

⚫

Insufficient — Very limited or no human research

How we label each claim

Individual expert claims get their own verdict based on how the research compares to what was said:

Supported by research — Published studies back the specific claim.

Partially supported — Some evidence supports it, with caveats or limited scope.

Not supported by research — The available research does not back the claim.

Contradicted by research — Studies point the other way.

Insufficient evidence to assess — Too little human research to judge.

Sources

Research

PubMed / NCBI E-utilities. Every study is linked so you can read the source. Systematic reviews, meta-analyses, and human RCTs are ranked highest; animal and mechanistic work is included but weighted lower.

Expert claims

Extracted from publicly available podcasts, videos, and transcripts, attributed to the source and episode. We track Andrew Huberman, Rhonda Patrick, Peter Attia, Tracey Marks, and Gary Brecka.

Limitations & standards

· A strong label doesn’t mean a supplement is right for you; an insufficient label doesn’t mean a claim is false — only that the evidence is limited for the claim being evaluated.
· We avoid certainty theater. When research is mixed, low-quality, animal-only, or missing, the page says so.
· Supplements with very thin or experimental evidence are kept out of the main public index until coverage improves.
· Scores and labels are recomputed as new research and claims are added, so a supplement’s grade can change over time.

This is not medical advice. Abstraction Health is an evidence navigator — not a clinical tool. We don’t diagnose, prescribe, or recommend doses. Talk to a qualified healthcare provider before starting, stopping, or changing any supplement or medication.