How we measure what is viral

What counts as engagement

Each network exposes different numbers, so there is no single formula. These are the actual expressions that rank every leaderboard on the site:

Two of these deserve an explanation rather than a defence. Retweets count double because a repost puts the content in front of a new audience while a like does not. TikTok plays are divided by 100 because plays run roughly ten times likes, and without dividing them they would drown every other signal — a video with reach but no reaction would outrank one people actually responded to.

Where the numbers come from

Articles

We poll a curated list of RSS feeds every few minutes for new articles. For each article we then ask Facebook, Reddit and Pinterest how many times that exact URL was shared, and we keep asking on a decaying schedule while the article is young. These are real share counts reported by those platforms, not estimates, not a proxy from traffic.

Social posts

Posts are captured from the networks themselves, with the counters the platform displays at the moment of capture. We re-read them as posts age, so numbers move — a post that goes viral on day three shows the day-three figure, not the day-one one.

How current is a number

• New articles are discovered every 5 minutes.
• Share counts refresh hourly while an article is inside its tracking window; after that the last measured value is kept and no longer updated.
• Public ranking pages are cached for 6 hours and regenerated on a 4-hour schedule, so the page you land on is warm rather than rebuilt by your visit.

This means a public page can trail the live data by a few hours. Inside the product the same numbers are current.

Why a page might not exist

Not everything we track gets a page. A ranking built from too little data is worse than no ranking, so each surface has a floor, and below it the URL simply does not exist — it returns 404 rather than a thin page. These thresholds are read from the running code, so this table cannot drift from what the site enforces:

What we do not do

• We do not re-host anyone's work. Public pages carry a headline, a link to the original, and how many times it was shared. The traffic goes to the creator or publisher.
• We do not invent dates. When a platform obscures a post's publication time and we cannot resolve it, the page says the date is unknown. It never falls back to "now" — that would make an old post look like breaking news and float it to the top of every recent view.
• We do not fill gaps with estimates. A metric a platform did not give us is shown as absent, not modelled. On an article that spread on Reddit and nowhere else, the Facebook row is simply not there.
• We do not write the rankings. Nobody chooses who appears. The order is the formula above, applied to measured numbers.

What we cannot tell you

The honest limits of the data, because a ranking read without them is misread:

• Our corpus is what we track, not the internet. "Most shared publishers" means most shared among the feeds we follow. A great outlet we do not track cannot rank.
• Article counts measure our tracking window, not a publisher's output. An outlet we added last month will show fewer articles than one we have followed for a year, regardless of how much either publishes. That is precisely why publishers are ranked by how many articles travelled, never by how many exist.
• Share counts are what the platform reports. When a platform changes how it counts or rate-limits us, the number moves for reasons that have nothing to do with the content.
• Absence is not evidence. A creator, sound or publisher missing from a ranking usually means we have not tracked enough of them yet — not that they are not performing.