Nobody outside Google knows YouTube’s ranking function. What YouTube does publish, in its Help Center and through the Creator Insider channel, is a consistent description of the shape: there is no single algorithm, there are separate systems for Home, Suggested and Search, and all of them are trying to predict viewer satisfaction rather than reward creators.
The most useful line YouTube keeps repeating is that the system follows the audience, not the other way around.
What YouTube says it optimises for
YouTube’s “Search and discovery on YouTube” Help page states that Home is based on video performance, described as “how well your video has interested and satisfied similar viewers, among other factors,” plus the viewer’s own watch and search history.
Suggested videos are “ranked to offer your audience videos they’re most likely to watch next,” combining relevance to the current video with personalisation.
Search ranks on “how well the title, description, and video content match the viewer’s search” and “what videos drive the most engagement for a search.” Note the second half: YouTube search is not purely a text-matching problem.
Across all of these, YouTube says the recommendation system considers hundreds of signals, and names watch behaviour, likes and dislikes, “not interested” feedback, and search queries.
Note: Figures here were verified against YouTube Help “Search and discovery on YouTube” and YouTube’s creator communications as of September 2026. Platforms change these without notice.
What signals does YouTube rank on?
Ordered by surface, as YouTube presents them.
| Surface | Signals YouTube names | What it is predicting |
|---|---|---|
| Home | Video performance with similar viewers; your watch history; your search history; how often a video has already been shown to you | Will this viewer start and stay |
| Suggested | Relevance to the video being watched; the viewer’s history; performance | What you watch next |
| Search | Title, description and video content match to the query; engagement driven by videos for that query | Query satisfaction |
| All | Watch behaviour, likes and dislikes, “not interested” and “don’t recommend channel” feedback | Satisfaction, not just clicks |
The “similar viewers” phrasing is the part most guides skip. Performance is not evaluated in the abstract. It is evaluated against how comparable audiences responded.
What changed most recently
Stated by YouTube: the surface-by-surface description above is the current Help Center position, and the explicit inclusion of negative feedback (“not interested”, “don’t recommend channel”) as a ranking input.
Widely reported, not confirmed with numbers: that Shorts and long-form are ranked by separate systems with separate performance expectations. YouTube’s creator communications describe them as distinct experiences with different viewing behaviour, which is consistent, but no published document gives you a weighting.
Practitioner heuristic: click-through rate and average view duration as the two dials to watch in Studio. Both are real metrics YouTube gives you. Neither is confirmed as a direct ranking input in the form creators use them. They are proxies for what YouTube says it measures, which is interest and satisfaction.
Myths this kills
Shadowbans. YouTube does not confirm a mechanism by that name. It does have documented ineligibility: content that is not advertiser-friendly, not suitable for all audiences, or in violation of guidelines gets restricted, and Studio tells you. A video that simply is not being recommended is usually a video that did not hold the viewers it was shown to. See what a shadowban actually is.
Engagement pods. Unfounded and unusually risky on YouTube, because artificial views are explicitly against the Terms of Service and get removed. Worse: if YouTube shows your video to an audience that leaves immediately, that is a signal against the video.
“The algorithm hates links.” No published support. Descriptions with links are normal. What YouTube demotes is spam and misleading metadata, which is a different thing.
Posting time is decisive. Least true on YouTube of any platform on this list, because a video accrues views over months and the Home feed is not chronological. Timing affects the first-day subscriber surge and little beyond it. Our best time to post on YouTube guide says as much, and the YouTube posting time tool converts the windows. Cadence is covered in how often to post on YouTube.
What actually moves views, in priority order
- Whether the video satisfies the people it was shown to. YouTube’s stated target. Everything else is downstream.
- Title and thumbnail as a matched pair. They decide whether the video gets sampled at all, on every surface.
- The first 30 seconds. Where most of the loss happens.
- Topic consistency across the channel. “Similar viewers” is a comparison the system can only make if it knows who yours are.
- Packaging for search intent, when the video is a search video. Titles and descriptions genuinely do match text on this platform.
- Publishing regularly. Not a ranking signal. It builds the audience history the system uses.
What to do
Decide before you record whether the video is a browse video or a search video, and package it accordingly. Make the first 30 seconds deliver on the thumbnail. Then keep the channel narrow enough that YouTube can identify who a “similar viewer” is.
Scheduling YouTube alongside everything else
Long-form is not really a scheduling problem, but Shorts are: the same vertical cut usually belongs on TikTok and Reels too. BulkPublish publishes to YouTube alongside 14 other platforms from one calendar, so a Short goes out three places from one upload. There is a guide on scheduling YouTube Shorts.
The parts you control
Format, hook, consistency. On YouTube the hook is literally two objects: the title and the thumbnail. The format decision is browse versus search. And consistency is what lets a system built on “similar viewers” work out who yours are.