How YTDiscovery Scores 100,000+ YouTube Videos
Every statistic we publish comes from our own discovery index. This page documents where the data comes from, how outlier scores are calculated, how often everything refreshes, and where the approach has limits. If you cite our numbers, cite this page with them.
What is in the index?
As of July 18, 2026, the index holds 106,871 long-form YouTube videos across hundreds of niches, and it grows daily. Videos enter the index through niche seed searches: a self-refreshing list of topics that expands as trending niches are detected, plus the channels behind videos that are already overperforming.
For every video we store the title, channel, publish date, view count, and subscriber count at crawl time, then keep the view counts fresh on a fixed refresh schedule. All data comes from the official YouTube Data API. We do not scrape private or non-public information.
How outlier scores are calculated
An outlier score answers one question: how far did this video beat its own channel’s normal performance? We calculate it by dividing the video’s views by the channel’s median recent views. The median (the middle value of the channel’s recent uploads) resists distortion from one earlier hit or flop, which an average would not.
Scoring against the channel’s own baseline is the core design choice. It removes subscriber count from the equation entirely, so a 20x breakout on a 500-subscriber channel ranks beside a 20x breakout on a 500,000-subscriber channel. On this basis, 2x is a real outlier, 5x is strong, and 10x or more is viral relative to the channel.
When a channel is too new or too sparse for a reliable median, we fall back to a proxy: views compared against expected reach for the channel’s subscriber count. Scores computed this way are marked “est” in the product, because the proxy is weaker: dead subscribers deflate it and browse-traffic channels inflate it. We never present a proxy score as a true channel-median score.
How often does the data refresh?
The crawler runs every 6 hours against the official YouTube Data API, pulling new videos from seed searches and from channels already represented in the index. View counts on indexed videos refresh twice daily, so outlier scores track a video’s climb rather than a stale snapshot. Trending niche summaries are re-synthesized once a day. Any statistic we publish in a report is re-verified against the live index on its publish date, and dated.
What an outlier score claims, and what it does not
A high outlier score claims exactly one thing: this video pulled dramatically more views than its channel normally gets. That is strong evidence the idea, title, or thumbnail resonated beyond the channel’s existing audience, which makes it a useful research signal.
It does not claim the video is high quality, accurate, or monetizable. It does not claim the format will work for you, on a different channel with a different audience. And it does not predict the future: a 45-day-old breakout may already be saturated by imitators. Outlier data tells you where proven demand is; judgment about what to do with it stays with the creator.
Known limitations
- Small-sample medians. A channel with only a handful of recent uploads has a noisy median, which can overstate or understate a score. Very new channels fall back to the subscriber proxy, marked “est”.
- Coverage is not exhaustive. The index grows from seed niches outward. A breakout in a niche we have not seeded yet will not appear until the crawler reaches it.
- Deleted and private videos are pruned. When a video disappears from YouTube it is removed from the index, so historical rankings can shift slightly after the fact.
- Shorts are excluded from the long-form index. Shorts distribution behaves differently enough that mixing the two would corrupt the medians. Long-form outlier scores only compare long-form videos.
- View counts lag by up to half a day. Refreshes run twice daily, so a fast-moving video’s live count can be ahead of ours between refreshes.
Citing this data
Journalists, researchers, and creators are welcome to cite statistics from the YTDiscovery index. Please attribute them to “the YTDiscovery index” with the date the number was published, and link this page as the methodology reference. Our published analyses live on the research blog, and the definitions behind the terms are in What Is a YouTube Outlier?
Last updated:
See the index in action
Search any niche and watch true channel-median outlier scores rank what's actually working. Free to start.