Breakout YouTube channels by niche

How this board is ranked

The number on the right of each row is the breakout score. It replaced an earlier ranking by raw views per subscriber, which had no time term: measured on this dataset, that older metric put videos with a median age of 114 days at the top (oldest 212 days), so it rewarded “a small channel once had a hit” rather than “this is being pushed right now”.

score = views ÷ √(age in days × subscribers) × engagement credibility, then divided by the dataset median, so1.0× is the median signal.

• Time (÷ age): a video with 10M views over 300 days and one with 10M views over 10 days are not the same event. Age is floored at 3 days so brand-new uploads cannot divide their way to the top.
• Audience (÷ subscribers): kept from the old metric — the point is distribution reaching past an existing subscriber base, not raw size. Subscribers are floored at 1,000.
• Damping (square root): the two factors are combined with a square root rather than multiplicatively. Chosen by measurement: between exponents 0.3 and 0.5 the top 20 overlaps 80–100%, while 0.75 and above degenerates into “newest wins” (overlap falls to 50%, then 20%).
• Engagement credibility: likes per 100 views against the same niche's median, clamped to 0.4×–1.5×. It exists because views alone can be bought: 934 of 3536 signals here sit below 0.6× their niche's engagement. Niche-relative because ASMR, music and documentary legitimately have lower like rates than tutorials. The clamp means engagement can never move a video more than 2.5× — it is a guard, not the ranking.

What this score cannot do. Every metric on this site is a stored capture, not a live counter, and there is only one capture per video — so “views per day” is the average since publish and cannot tell climbing from already-peaked. The score is also relative to the current dataset, so it changes when the dataset is re-scraped. Engagement is a proxy for authenticity, not proof: it demotes videos whose views look detached from their likes, and nothing more. Median engagement across the dataset is 1.01 likes per 100 views.

Reach multiples describe distribution, not money. To put a multiple like these into a value under your own assumptions, the YouTube channel value calculator applies an RPM you choose to real view counts and lists high-scoring channels from this same dataset as comparables.

How this board is filtered

Nothing here is hand-picked, but it does arrive by two different routes — and both are held to the same metric floors, while only one of them takes a single video per channel.

• Keyword search: 1037 entries (29% of the dataset) came from a fixed keyword pool run against the YouTube Data API v3 search endpoint. This route keeps one video per channel per run.
• Channel pulls: 2499 entries (71%) came from following a channel already in the dataset and taking several of its recent uploads. This route takes more than one video per channel on purpose — the deepest single channel contributes 35 entries, which is why one channel can appear several times on a board.

Floors are applied to a video when it is first ingested, and this board accumulates across rounds instead of being rebuilt from scratch each time. The dataset can therefore hold entries that would not enter today: currently 1 of 3536 fall outside at least one floor.

• Window: a 180-day publish window is applied when a video is first ingested, not to the board as a whole — this dataset accumulates, so an entry admitted inside the window stays after it ages out.2944 of 3536 entries are still inside 180 days today, and the oldest here was published 398 days ago. What the stored data cannot do is verify the window that was in force at ingest: each row keeps one set of metrics, the newest capture, and no row records the age it had when it was accepted. So this page reports the current age distribution rather than asserting the window still holds.
• Capture dates: the pull date is written by the channel-pull route only, so 3215 of 3536 rows carry one (2026-09-06 to 2026-10-11). The other 321 were found by keyword search, which records no date at all — not a missing value that used to exist, a route that never had the field. Where a date exists it is the first capture and is preserved across re-pulls rather than refreshed. Ages shown anywhere on this site are recalculated from each video's publish date at build time (2026-10-11), not carried over from capture time.
• Minimum views: 20.0K — below that, a video is too small to prove demand.
• Maximum subscribers: 100.0K — larger channels break out because of their existing audience, not because of the format.
• Entry floor: 20.0K views, 3× views per subscriber and at least 1 like, checked at capture time. These floors decide what enters the dataset; the breakout score decides the order.
• Minimum likes: 1 — zero-like videos are usually a metrics artefact.
• Shorts excluded: uploads under 60 seconds and titles carrying Shorts/Reels/TikTok hashtags are filtered out.
• One video per channel per keyword run: applied to keyword-search candidates only — channel pulls take several uploads from the same channel (see above), so a channel with 35 tracked entries is a legitimate outcome of the second route rather than a broken rule.
• Label and auto-generated channels removed: VEVO and “- Topic” channels are excluded (120 excluded so far), because their reach comes from distribution deals, not from a copyable format.
• Promotional, brand and derivative uploads removed: 107 excluded so far. The floors above are the ones a corporate channel clears most easily — a company's new channel starts with few subscribers and can buy or earn enormous reach, so “low subscriber count” does not screen it out. The filter that does the work is therefore content form rather than scale: official trailers and teasers, product launches, brand and distributor channels, and re-creations of other people's films (fan-made trailers, concept trailers, remasters) are excluded — the last group for rights reasons as much as for relevance. What is left is what a faceless creator can actually copy. The excluded rows are not deleted: they are kept in archive/quarantine-editorial.json in the repository, and the build fails if one of them reappears in the dataset.

What this snapshot contains: 3536signals across 1334 channels, with a median reach of15.6× views per subscriber, 4.2K/day at the median and a median view count of 347.6K. By runtime, 18% of the tracked videos run 3 minutes or less and 30% run over 30 minutes — the rest sit in between. Two limits belong next to those numbers: the tracker discards anything at 60 seconds or less and any title tagged #shorts, so there are no YouTube Shorts in this dataset, and "3 minutes or less" is a length band rather than a claim that a video is a Short (the Data API does not return aspect ratio, so vertical format cannot be verified from here).