How community questions become an evidence-review queue

We rank what people repeatedly ask, including mistaken or low-quality claims. Ranking decides what deserves an answer; it does not decide whether a claim is true.

Method version 1.0.0 · Updated August 5, 2026

Popularity formula

Five signals, one bounded score

Frequency

35%

How often a distinct, de-duplicated question appears in the observation window.

Search demand

20%

First-party search patterns and approved query sources, normalized before comparison.

Cross-platform

20%

Whether the same question appears across independent source types rather than one community alone.

Engagement

15%

Available replies, views, shares, or likes, bounded so one viral item cannot control the list.

Recent growth

10%

Change during the current window compared with an equivalent earlier period.

Each component is normalized from 0 to 100. Ties use a stable code-point order so the same inputs always produce the same output.

Source scope

Approved observation channels

Manually reviewed public questions
Official YouTube API comment data
Publisher-approved RSS feeds
Google Search Console queries
Consented first-party questions
Manual Reddit review under the approved policy

Platform boundaries

What the sample cannot represent

No platform is treated as the public. Access rules, recommendation systems, moderation, language, and audience composition shape what is visible. Publication requires at least three source types, no single source may exceed 40% of a molecule's sample, and no popularity score becomes a prevalence estimate.

Privacy

Questions are separated from people

Names, birth dates, first-person medical histories, emails, phone numbers, handles, addresses, coordinates, private identifiers, and links are removed before human labeling. Advertising, obvious automated messages, and records with residual private data are excluded. Public pages contain only normalized questions, bounded counts, review links, and coverage labels.

Publication gate

A list can remain empty

The observation target applies to the complete de-duplicated molecule sample, not to each question. A molecule also needs at least three source types, no source above 40%, ten human-reviewed question-to-answer bindings, valid evidence pages, and editorial approval. Semaglutide and tirzepatide require 500 observations; other targets range from 150 to 250. Until every gate passes, the product page shows no ranking.

Corrections

Challenge the grouping, source, or interpretation

Send the page URL, the exact statement, and a primary source when available. We review substantive corrections and update the evidence-review date when the answer changes.

Submit a correction