Editorial Policy

Last updated: 2026-09-21

PubMed Trivia exists to deliver primary science literature to general readers, accurately. This page discloses, transparently, our editorial process, quality controls, use of AI, and disclaimers.

1. Source material

  • Every article is based on peer-reviewed papers from the NIH database PubMed.
  • Priority: systematic reviews / meta-analyses > narrative reviews > RCTs. Case reports and opinion pieces are excluded.
  • Retracted papers are excluded once flagged in metadata.
  • Preference is given to papers published within the last ~30 years (classic-paper deep dives are labeled explicitly).

2. Production pipeline

  1. Selection: PubMed E-utilities are queried per category, sampled at random, and de-duplicated against already-published articles.
  2. Fetch: Abstract, journal, publication year via NIH E-utilities.
  3. Adaptation: Anthropic's Claude (Sonnet 4.5 / Haiku 4.5) adapts the abstract for general readers. English and Japanese are generated independently — no re-translation.
  4. Post-check: NG-word screening (misleading efficacy claims, unqualified recommendations, causal overreach). Failures trigger regeneration.
  5. Publication: Site, SNS, and XML sitemap.

3. AI use — full disclosure

Articles on this site are adapted with AI (Claude by Anthropic) as an assistant. AI replaces conventional translation/summarization labor so a larger volume of papers can be surfaced efficiently.

  • Prompts explicitly constrain the model to facts within the source abstract.
  • The rule "do not translate association as causation" is enforced.
  • Efficacy claims (e.g. "cures", "prevents", "guaranteed") are blocked by NG-word filters.
  • AI errors are possible in principle, so every article carries the source PMID and PubMed URL, letting readers verify the primary source directly.

4. Deep-dive evidence reviews

In addition to single-paper articles, we publish 1-2 evidence-review articles per month that synthesize ~10 systematic reviews / meta-analyses on one theme. These can be found via archive or search.

5. Medical disclaimer

Content here is general-education trivia, not medical advice, diagnosis, or treatment.

  • For health decisions, consult a qualified professional (physician, pharmacist).
  • Statements about specific diseases, drugs, or interventions are hedged; efficacy is not asserted.

6. Corrections

For factual errors, mistranslations, or bad citations, email sciencepubmedjp@gmail.com. We will annotate or remove content after verification. Substantive corrections are dated at the bottom of the article.

7. Ads and affiliate links

This site displays ads via Google AdSense, Amazon Associates, and Moshimo Affiliate. See our privacy policy for details. The presence or placement of affiliate links has no influence on the content or conclusions of an article.

8. Publisher information

7. Ongoing Editorial Quality Improvements (Update Log)

We continuously work to improve editorial quality after publication. Key workflow enhancements:

  • 2026-10-11 (Sat): Of 625 total articles, the editorial team used Claude Sonnet 4.5 to add four analytical sections (Background, Mechanism, Limitations and critical perspective, Implications for readers) to 297 articles whose body text was under 1,500 characters (early 1-paper summary articles). All articles now exceed 2,000 characters, with average length now over 4,000 characters. Prompts strictly constrain the model not to deviate from the original abstract's facts; a human sanity check follows each batch.
  • 2026-10-13 (Mon): Fixed a bug in the FAQ-section JSON-LD structured data output. Rich Result support for deep-dive articles with FAQ sections is now improved.
  • 2026-10-13 (Mon): Added BreadcrumbList / MedicalScholarlyArticle / WebSite + SearchAction / Person (author) JSON-LD schemas. Verified all items valid via Google's structured-data test.
  • 2026-10-08 (Wed): Appended `&s=review-rank` to all affiliate-link URLs across 625 files / 1,875 links so readers see highly-reviewed items first, reducing decision friction.

8. Commitment to Error Correction

We commit to responding to error reports within 48 hours (excluding weekends and holidays). Corrections are recorded in the article's frontmatter as `dateModified` and auto-pinged to Google Search Console.

Please report errors to: sciencepubmedjp@gmail.com (include the PMID in the subject line for faster handling)

9. Editorial Team

The PubMed Trivia editorial team is responsible for policy decisions, prompt design, NG-term list maintenance, article sampling review, and error-correction response. See the editor page for details.