How does the YouTube algorithm work? What YouTube says
Short answer
YouTube describes one recommendation system with two goals: help each viewer find videos they want to watch, and maximize long-term viewer satisfaction. It learns from viewers’ signals, such as watch history, search history, subscriptions, likes and surveys, and from how a video performs when it is shown: whether people click, how much they watch and whether they enjoyed it. Its pages say the system favors no format and has no ideal length, and that publish time is not known to affect long-term performance.
What the system is for
Read on 11 October 2026, on four YouTube Help pages and the How YouTube Works site. None shows a date.
YouTube’s recommendation system page says the system aims to identify the most relevant content for each user at any given moment, and that its algorithm has two goals: help each viewer find videos they want to watch, and maximize long-term viewer satisfaction. It analyzes each viewer’s profile in real time, considering device, time of day and past habits.
The same page puts the creator’s side plainly: creators often focus on “the algorithm”, but what matters is what viewers enjoy, and they should ask whether the audience likes it. It sorts the signals into two kinds. Viewer personalization is what tells the system a person’s preferences. Content performance is how well a video does when it is offered, such as whether viewers choose to click, watch or engage positively.
How YouTube recommendations work names where recommendations appear: the homepage, Up Next, the Shorts player, destination pages such as Music, and channel pages. It adds that features lean on different signals: Up Next uses the video you are watching as its main signal, and the homepage primarily relies on your watch history.
The signals YouTube names
| Signal | What YouTube says |
|---|---|
| Watch history | What viewers choose to watch, ignore or dismiss, how long and how much they watch, and the languages they watch in. |
| Search history | What you search for influences future recommendations. |
| Subscriptions | The channels a viewer subscribes to help recommend other videos. |
| Likes and dislikes | Likes help predict interest in similar videos; dislikes inform what to avoid recommending. |
| “Not interested” and “Don’t recommend channel” | Feedback the system uses to avoid similar videos or the channel. |
| Satisfaction surveys | Ratings help the system understand satisfaction, “not just watch time”. |
| Interest affinity | Themes, topics and formats a viewer favors, and what similar viewers also enjoyed, as in “viewers of video A also watched video B”. |
| Device and time of day | The same video may rank higher on the device where it historically performs better for that viewer. |
YouTube says the system learns every day from over 80 billion pieces of information, which it calls signals.
How YouTube says it reads a video’s performance
The Search and discovery tips page says the systems look at how a video does when it is recommended: whether viewers choose to watch it, ignore it or click “not interested”, and whether they stick around. It names average view duration and average percentage viewed as signals that inform ranking, then likes and post-watch survey results. The Shorts edition lists the percentage of viewers who chose to view, as well.
It also names three outside factors that explain where an audience comes from:
- Topic interest. How many people worldwide are interested in and watching a topic, which changes over time.
- Competition. YouTube’s systems rank videos from all channels a viewer might watch, so a video with good metrics can still get fewer impressions if others do better.
- Seasonality. Traffic changes through the year and around holidays.
The How YouTube Works page adds that, as on Search, the system considers the reputation and quality of a channel to determine how, when and to whom content can be surfaced, and uses external evaluators to understand how an average viewer might see it.
Does it matter? What the pages say, item by item
| Question | What YouTube’s page says |
|---|---|
| Time of upload | Publishing when your audience is most active “might lead to more immediate views”, but YouTube has not observed evidence that it affects long-term viewership: “publish time is not known to impact a video’s long-term performance”. For a Premiere or a live stream, consider your audience’s active times. |
| Video length | There is “no universal ‘ideal’ length”. Use the audience retention curves in analytics. |
| Format | “Our system has no opinion about what type of video you make, and doesn’t favor any particular format.” |
| Trying new formats | Shorts, VODs or live streams “will not inherently confuse the algorithm”. A perceived effect usually comes from how viewers react to the new content. |
| Posting often | “Prioritize consistent quality content over a high frequency of uploads.” For Shorts: no minimum posting cadence is required. |
| Taking a break | Breaks are encouraged, and the algorithm doesn’t penalize creators for them. |
| One weak video | An individual video’s underperformance does not penalize a channel overall. |
| Many weak videos | Frequently sharing content that doesn’t resonate “can, over time, affect your channel’s overall performance”. A second page adds that long-term performance can be affected if a viewer consistently stops watching a channel’s recommended videos. |
| Changing a title or thumbnail | Performance can shift, because the systems respond to how viewers react, not to the act of changing. It is generally advised when click-through rate is low and impressions are below usual. |
| AI-made videos | They “can be recommended as broadly as videos made without AI”. |
| Monetization | The recommendation algorithm does not prioritize videos based on whether they are monetized. |
| Unlisted first | Uploading as unlisted before making a video public should not significantly affect its performance. |
| Shorts and long videos | Shorts performance doesn’t negatively affect long-form recommendations; the Shorts page says Shorts can help with discovery. |
| Subscriber count | It is the number of times people subscribed, not the size of the current active audience. The Unique Viewers metric in the Audience tab shows the active audience. |
What YouTube advises
The plainest recommendation, on the search and discovery page: “Focus on what your audience likes. If you do that and people watch, the recommendations will follow.” The recommendation system page and the good-to-know page add a list:
- Be sustainable. Regular delivery, and a consistent title and thumbnail style, make videos recognizable.
- Know your audience. The Audience tab shows preferred formats and the channels and content your audience also watches.
- Help viewers find more. Series, clear calls to action, playlists and end screens. “Positive viewer interactions are a strong signal for the recommendation system.”
- Experiment. The system uses fresh performance data for each video, so one that doesn’t land won’t necessarily hinder the channel.
- Build a library. For new channels, a substantial library lets a viewer who finds one video go deeper, which YouTube says signals value to the algorithm.
What the pages do not say
- A number of views, a click-through rate or a watch-time percentage that a video needs. They name the signals, not the thresholds.
- How the signals are weighed against each other, beyond which feature relies on which signal.
- How many impressions a new video gets at the start.
- A date: none of the pages shows when it was written, so we cannot say how current each statement is.
Mellow Hub and YouTube
Hub does not change how YouTube ranks a video. It publishes what you made. What Hub connects for YouTube: A Google account with a YouTube channel. Uploads count against the channel's daily API quota. Hub’s cabinet shows the views and likes YouTube reports for each published video. It does not show Studio’s retention, traffic sources or the Audience tab that YouTube’s advice points to; for those, see how to see YouTube analytics.
YouTube’s own pages tie impressions to how a video is packaged. Mellow Hub, whose answer this is, has a YouTube packaging guide, an AI YouTube thumbnail generator that spends credits, and a free YouTube thumbnail tester.
The first 3 posts are free, however many networks each one goes to, and need no card. After that Hub is paid: plans start at $19 a month (Solo), after a 7-day trial with a card on file that costs nothing if you cancel before it ends, and a monthly plan's first month is 20% off. Publishing is unlimited on every plan, under a fair-use ceiling that stops automated floods (900 publications a month on Solo), where one publication is one post on one account, so a post sent to five networks counts as five toward it. There are no automatic overage charges.
Questions people also ask
Is there one YouTube algorithm?
YouTube describes one recommendation system that is tuned for each surface. Search strives to surface results relevant to the query and looks at which videos have driven the most engagement for it, the Shorts feed may lean on the recency of content, and the Subscriptions tab lists videos from subscribed channels by most recent.
Does YouTube punish you for not uploading, or for a video that flops?
YouTube’s page says the algorithm doesn’t penalize creators for taking breaks, and that one video’s underperformance does not penalize the channel. It also says that frequently sharing content that doesn’t resonate can affect overall performance over time.
Do likes, comments and subscribers help a video get recommended?
Likes are one of the viewer signals YouTube names, and it says positive viewer interactions are a strong signal. It gives no weights. It also says the subscriber count is not the size of the active audience, and points to Unique Viewers in the Audience tab.
Does YouTube treat Shorts differently from long videos?
Its Shorts page says the system doesn’t favor any format, ranks Shorts on performance and viewer personalization, and that Shorts performance doesn’t negatively affect long-form recommendations. It names the percentage of viewers who chose to view, average view duration and average percentage viewed as Shorts signals.
Does YouTube recommend videos made with AI less?
YouTube’s page says videos made with AI can be recommended as broadly as videos made without AI. Its How YouTube Works page separately says creators must disclose realistic altered or synthetic content, and that a label then appears on the video or in its description.
Sources
- YouTube Help: YouTube’s recommendation system (undated)
- YouTube Help: How YouTube recommendations work (undated)
- YouTube Help: Search & discovery tips, video and Shorts editions (undated)
- YouTube Help: Good to know about recommendations for YouTube’s recommendation system (undated)
- How YouTube Works: Algorithm-based recommendations on YouTube (undated)
Other companies’ features change. Every statement about them rests on these pages, as they read on the date above.