Direct answer
Go straight to PubMed (over 36 million publications, NIH-funded, free). Filter for meta-analysis or RCT, sample size >100, published in the last 5 years. For fresher preprints, Europe PMC. For citation impact, iCite (RCR score). Influencer sources like Bulletproof or Goop aren't research — they're marketing.
Deep dive
The three sources you actually need
PubMed — the standard. Over 36 million biomedical publications. Curated by the US National Library of Medicine, indexed per article with MeSH (Medical Subject Headings) for precise search. Upside: every study has a PMID — a unique numeric ID you can verify in any evidence-led forum or article. A PMID always references the same study, worldwide.
Europe PMC — the supplement. Mirrors PubMed plus preprints from bioRxiv/medRxiv plus open-access full text. Important when you want fresh research (preprints appear months before peer review). Caveat: preprints aren't peer-reviewed yet — effect sizes can shift after review, or the study may be retracted.
iCite — the impact metric. NIH tool that computes the Relative Citation Ratio (RCR). A study with RCR=2.0 is cited twice as often as the median in its field — that's a more robust impact indicator than the Journal Impact Factor (which measures the whole journal, not the individual study). Landmark studies typically have RCR >5.
The PubMed search that actually works
Beginners type "creatine memory" and scroll 200 hits. That's not research, that's a lottery. Do it this way:
creatine[MeSH] AND memory[MeSH] AND (randomized controlled trial[pt] OR meta-analysis[pt])
Then filter:
- Publication Date: Last 5 Years
- Species: Humans
- Article Type: Meta-Analysis, RCT, Systematic Review
What remains is often 5–20 high-quality studies instead of 200. Avgerinos et al. 2018 (PMID 29704637) is, for instance, the standard review on creatine and cognition.
Where the hype lies — or the data are thin
Three common traps:
- Selling animal studies as human evidence. "Resveratrol extends lifespan by 30%" — in mice, at a pharmacological dose. For a human you'd have to eat 2 kg of grapes daily. The Species:Humans filter prevents that.
- Industry-sponsored studies without disclosure check. A whey-protein study funded by a whey manufacturer has roughly a 2× higher effect bias than independent research (Lesser et al. 2007, PMID 17214504). On the PubMed detail page, scan for "Conflict of Interest Statement".
- Treating outdated data as consensus. A 2014 meta-analysis is not today's state of the art — especially for supplements (ashwagandha, NMN, lion's mane), 2020+ is mandatory.
Methodology — how we judge this
We rate studies by evidence hierarchy:
| Tier | Study type | Reliability |
|---|---|---|
| 1 | Meta-analysis (of multiple RCTs) | very high |
| 2 | Single large RCT (n>500) | high |
| 3 | Smaller RCT (n=20-100) | medium — effect sizes often overstated |
| 4 | Observational study (cohort) | correlational, not proof |
| 5 | Animal study | mechanistically interesting, not transferable |
| 6 | Anecdote / influencer | hypothesis, not proof |
On biohacking-ai.com/studien-karte you'll find our interactive 3D mapping of ~300 000 studies from PubMed + Europe PMC + OpenAlex — enriched with iCite RCR scores and topic clustering. The map is intentionally noindex (crawl budget), but we explain the research methodology in the methodology answer.
Sources
- Avgerinos et al. 2018 — Effects of creatine supplementation on cognitive function PMID 29704637 — example of a cleanly structured meta-analysis, good search-result template
- Lesser et al. 2007 — Relationship between funding source and conclusion among nutrition-related scientific articles PMID 17214504 — documents sponsor bias in nutrition studies
- Hutchins et al. 2016 — Relative Citation Ratio (RCR) PMID 27599104 — methodology paper for iCite's RCR score
Related answers
See below — auto-generated via relatedAnswers.