Methodology

How mensresearch.ai indexes, ranks, and synthesizes peer-reviewed research on men's health.

What we index

mensresearch.ai aggregates peer-reviewed research from PubMed (36 million+ biomedical citations) and Semantic Scholar (220 million+ scholarly papers across disciplines), filtered to men's health relevance via MeSH term matching and embedding similarity to a curated seed corpus. As of publication, 1.3M papers are indexed.

Preprints and non-peer-reviewed sources are explicitly excluded.

How we rank

For each search query, we compute a blended relevance score:

Score = 0.7 × semantic relevance + 0.2 × log-normalized citations + 0.1 × recency decay

Papers are then quality-filtered: abstracts must exceed 100 characters, be published ≥ 2015 or carry citations, and be ≥ 80% ASCII (filters OCR noise and non-English junk).

How we synthesize

The top 10 quality-filtered sources for a query are sent to Claude Sonnet 4.6 (Anthropic) which returns a structured JSON response with four components:

Verdict — one of: supported / mixed / not supported / insufficient evidence
TL;DR — two-sentence summary
Findings — key data points, each cited to a source paper
Caveats + methodology notes — study-design limitations, population specifics

If fewer than 3 quality sources are available, the system returns "insufficient evidence" rather than synthesizing on weak ground.

What we don't do

Clinical diagnosis, treatment, or prescription advice
Personalized medical recommendations
Replace a clinician relationship

See our Medical Disclaimer for details.

Limitations

Corpus gaps: non-indexed preprints, non-English papers, and paywalled full-text content are under-represented.
Recency lag: PubMed sync runs daily; Semantic Scholar sync runs weekly. New papers can be indexed within 24-72 hours.
Synthesis error risk: AI-generated summaries can misrepresent sources. We mitigate via source grounding (every claim links to a PMID) and quality filtering, but we log corrections at /errata.

Versioning

Algorithm v1.0 went live 2026-04-22. Material changes — weighting tweaks, prompt updates, filter additions — are logged at /methodology/history (coming soon).

Contact

Editorial questions or corrections: [email protected].

AI-generated content disclosure: Research summaries on this site are produced by AI (Claude Sonnet 4.6 by Anthropic) from peer-reviewed sources. Summaries are grounded in the cited papers but are not reviewed by a clinician before publication. Treat them as research education, not medical advice.