There is no single YouTube algorithm. YouTube runs recommendation systems for each place a viewer can come across a video: the homepage, the Up Next panel, the Shorts feed, search results, the Subscriptions feed, and channel and destination pages. Its help page on how YouTube recommendations work says those systems compare each viewer's habits with people who watch in similar ways, using signals it lists by name: watch history, search history, subscriptions, likes, dislikes, “Not interested” and “Don't recommend channel” feedback, and satisfaction surveys. A companion page on content performance sorts what the system learns about a video into three buckets: appeal, engagement and satisfaction.
One word, several systems
In the Creator Insider episode Behind the Algorithms, a YouTube team member explains that separate systems serve Home, Suggested and other pages, and that a creator cannot optimize for a traffic source, only for the people using it. A later episode on what creators need to know about the algorithm suggests swapping the word “algorithm” for “audience” whenever you ask what the system wants.
The recommendations help page spells out how inputs shift by surface. Up Next uses the video someone is watching as its main signal. The homepage relies primarily on watch history, so a viewer with watch history off and little past history sees a homepage with the search bar and side menu instead of recommendations. The Shorts Feed is personalized to what YouTube expects the viewer wants next, and some shelves on channel pages and on destination pages such as Music are personalized too.
Search follows its own rules. YouTube's page on how YouTube search works names three elements whose importance varies by type of search: relevance, from how well the title, tags, description and video content match the query; engagement, such as the watch time a video earns for that query; and quality, meaning channels that show expertise, authoritativeness and trustworthiness on the topic. It adds that YouTube does not accept payment for better organic placement.
YouTube's recommendation notes for creators say each surface is tuned to a different viewer objective: the Shorts feed may lean on recency, and the Subscriptions tab lists the newest videos first. The system also learns from context such as time of day and device type, so one person may see different videos on a phone in the morning than on a TV at night.
The viewer signals YouTube names
The recommendations help page says the system learns every day from over 80 billion pieces of information that YouTube calls signals, and it lists the primary ones:
- Watch history: the videos someone has watched, which also lets YouTube remember where they left off.
- Search history: what someone has searched for on YouTube.
- Channel subscriptions: the channels a viewer has chosen to follow.
- Likes and dislikes: likes help predict interest in similar videos, while dislikes tell the system what to avoid.
- “Not interested” and “Don't recommend channel”: explicit feedback that steers recommendations away from a video or a whole channel.
- Satisfaction surveys: ratings viewers give after watching, which help the system judge satisfaction rather than watch time alone.
A post on the official YouTube blog, On YouTube's recommendation system, explains how the signals stack. Watch time was added to clicks because a click does not prove anyone watched. Surveys measure what YouTube calls valued watch time, with a model predicting answers for viewers who never take one. Shares, likes and dislikes are weighted per person, so a viewer who shares nearly everything has shares counted less.
The How YouTube Works page on recommendations adds a channel layer: YouTube weighs a channel's reputation and quality when deciding how, when and to whom its content is surfaced, uses external evaluators to judge how an average viewer might perceive a video, and gives extra prominence to high-quality sources for news, personal finance, and medical and scientific topics. Apart from the metadata that search reads, these inputs come from viewers or from quality assessments, not from settings you can switch on in YouTube Studio.
Appeal, engagement and satisfaction
The content performance page reduces a video's performance to three questions. Appeal: did people choose to watch it, or ignore it or click “Not interested”? Engagement: once they started, did they stick around? Satisfaction: did they enjoy it? The exact signals differ by format, but YouTube says the goal stays the same, which is to predict whether a viewer would be interested in a video and feel satisfied after watching.
YouTube's Content tab analytics tips map a similar funnel onto Studio metrics: impressions click-through rate for appeal, views for whether people watched, and average view duration for satisfaction. Treat them as proxies, because survey answers are not shown in a creator's Analytics.
The impressions and click-through rate FAQ shows the buckets working together. YouTube says it recommends a video to viewers when it is relevant to them and its average view duration indicates they find it interesting, and it describes the clickbait pattern: high click-through rate, low average view duration and fewer impressions than expected. A Creator Insider episode on how videos get discovered adds that those two metrics are among dozens of signals used for search and discovery.
Signal-by-surface map to YouTube Analytics
The traffic source report described on YouTube's video reach page names each surface, so you can see which one sends your views and what YouTube says it responds to:
| Surface | What YouTube says shapes it | Where to look in YouTube Analytics | Source |
|---|---|---|---|
| Home (Browse features) | A personalized mix of recommendations, subscriptions and news, relying primarily on the viewer's watch history | Reach tab, “How viewers find this video”: the Browse features source, then impressions, click-through rate and average view duration | How YouTube recommendations work |
| Up Next (Suggested videos) | The video being watched is the main signal; Creator Insider describes videos often watched together and topically related videos | Reach tab: the “Content suggesting this video” card and the Suggested videos source | Creator Insider: Behind the Algorithms |
| Shorts feed | Personalized to what YouTube thinks the viewer wants next, and may favor recent content | The Shorts traffic source and the “Stayed to watch” metric for each Short | Recommendation notes for creators |
| YouTube search | Relevance, engagement and quality, weighted by type of search, plus the viewer's own history when it is turned on | Reach tab: the “YouTube search terms” report | How YouTube search works |
| Subscriptions feed | Videos from channels the viewer subscribes to, most recent first | The Browse features source, which YouTube says includes subscriptions | Video reach page; Recommendation notes for creators |
| Notifications | Alerts to subscribers, including those who chose all notifications and enabled them on a device | The “Notifications” and “Notifications to belled subscribers” sources, and the “Bell notifications sent” card | Video reach page |
| Channel pages | Some shelves are personalized to what YouTube thinks will interest the viewer most | The “Channel pages” traffic source | How YouTube recommendations work |
What YouTube says does not decide your reach
Several popular theories have direct answers in YouTube's recommendation notes for creators and on Creator Insider:
- Monetization status. YouTube says recommendations do not prioritize monetized videos, and that switching off monetization on a video shows no change in search or recommendation traffic. Revenue questions belong in our map of every YouTube revenue stream.
- Unlisted before public. Search and recommendations look at audience activity while a video is public, so an unlisted period first should not significantly change how it performs.
- One weak video sinking a channel. A single underperforming video does not penalize the channel, although long-term performance can slip if a viewer keeps skipping your videos when they are recommended.
- Trying a new format. Shorts, long-form videos or live streams will not confuse the system, because each piece is evaluated on its own and shown to viewers who prefer that format.
- A magic length. There is no universal ideal length, only the length that delivers the value without filler, checked against retention curves. Our guide to writing a YouTube script with a retention check turns that into a drafting routine.
- Tags. YouTube's performance page calls the title, thumbnail and description the important metadata and says tags mainly help with common misspellings. The full routine lives in our YouTube SEO checklist.
- The hour you publish. The Creator Insider algorithm episode argues that if the audience does not care what time a video goes up, the algorithm should not either. Testing timing on your own channel is covered in finding your best time to upload.
Editing a title or thumbnail falls in between. The notes say the system reacts to how viewers respond to the new packaging, not to the edit itself. YouTube advises leaving alone what works and trying a change when a video has a lower click-through rate and fewer impressions than usual; if nothing improves, the topic may simply appeal to a smaller audience.
Why a good video can still stall
Strong numbers do not entitle a video to more impressions. YouTube's notes name three outside factors: competition, since a video is ranked against everything else a viewer might watch; topic interest, since some subjects have far larger audiences; and seasonality. The Creator Insider discovery episode makes the same three points.
Click-through rate also falls as a video travels. YouTube's guide to decoding CTR and impressions works through an example in which impressions grow tenfold over a week while click-through rate drops from 9% to 3.5%, and calls that success because the video reached viewers beyond its core audience. Early click-through rate is often inflated by loyal fans, so read it next to impressions and traffic source.
Returning viewers count in their own way. On its page about new, casual and regular viewers, YouTube says people who come back regularly are more likely to be recommended more of your videos, though the segments themselves do not affect reach. Its channel health tips add that no universal benchmark exists for click-through rate, so compare against your own history.
A monthly review sheet for the signals you can see
You cannot see survey answers or anyone else's homepage, but you can see how each surface responds to your uploads. Once a month, compare the latest four weeks with the four before and keep notes in one running document.
| Check | Where in YouTube Studio | Question to answer | If the answer is yes |
|---|---|---|---|
| Traffic mix | Reach tab for your top videos, “How viewers find this video” | Did the share of views from Browse features, Suggested videos, Shorts or search shift? | Open the videos behind the shift and note what their topics and packaging share |
| Appeal on Home and Suggested | Reach tab, impressions and click-through rate by traffic source | Did click-through rate fall while impressions rose? | Read it as wider reach unless average view duration fell as well |
| Clickbait pattern | Reach and Engagement tabs | Does any video show high click-through rate, low average view duration and fewer impressions than usual? | Rework the title or thumbnail so it matches what the video delivers |
| Engagement | Engagement tab, audience retention and key moments | Do the biggest dips cluster at the same point across your top videos? | Move the strongest material earlier in the next script |
| Shorts | Each Short's Reach tab, “Stayed to watch” | Are more viewers swiping away in the opening seconds than last month? | Test a new opening on the next batch of Shorts |
| Search | Reach tab, “YouTube search terms” | Are viewers finding you with accurate terms your titles and descriptions never use? | Work those terms into future metadata where they describe the video truthfully |
| Returning viewers | Audience tab, new, casual and regular viewers | Did regular viewers shrink? | Check whether recent uploads drifted from your usual topics or format |
| Underperformers | Content tab sorted by impressions | Is a video getting a lower click-through rate and fewer impressions than your usual? | Try one packaging change, then recheck it next month before touching anything else |
After a few months the notes show whether a change in views came from one surface or from all of them, which tells you where to spend effort. When the review points to bigger changes in what you make, use our YouTube channel growth review system.
Limitations of this explanation
YouTube does not publish the weight of any signal, and its help pages change without notice, so check the date on any Creator Insider episode you rely on; several cited here are a few years old. Everything above comes from YouTube's public statements, not access to its systems, so it describes intent rather than a formula. Your Analytics show only part of the picture: survey responses and other viewers' histories stay hidden, small channels may see limited data, and one month from one channel is a small sample in which topic and season can outweigh anything you changed. Treat each pattern as a hypothesis to test on the next upload.