How the YouTube Algorithm Works, According to YouTube
Published 2026-09-25 · 10 min read · YTDetective
TL;DR
- YouTube says its algorithm has two goals: help each viewer find videos they want to watch, and maximize long-term viewer satisfaction. It learns from what each viewer does and from how a video performs when it is offered.
- YouTube says publish time is not known to impact long-term performance, that taking a break is not penalized, and that one video that does not land will not necessarily hurt a channel.
- A 2021 post on YouTube's official blog says recommendations drive more viewing than channel subscriptions or search.
- No page we found gives the weight of each signal. Claims such as a 7-second rule or a 'seed audience' test do not appear on the YouTube pages we read.
Most advice about 'the algorithm' is one creator's guess repeated by the next. YouTube has published more than people assume: a help page on the recommendation system, an FAQ on recommendations, a Shorts discovery page, a click-through-rate FAQ and two official blog posts. This guide puts what those pages say in one place and keeps YouTube's claims apart from our own reading of them.
The two goals and the two kinds of signal
YouTube Help says the system aims to identify the most relevant content for each user at any given moment, and that it has two goals: help each viewer find videos they want to watch, and maximize long-term viewer satisfaction. It says the algorithm follows the audience, and that creators should ask 'Does my audience like this?' instead of 'Does the algorithm like my content?'
The signals fall into two groups.
- Viewer personalization. Watch history covers the videos a viewer chooses, ignores or dismisses with 'not interested', how long and how much of a video they watch, likes, shares, comments, survey responses, subscriptions and the language of what they watch. Interest affinity covers favored themes, topics and formats, and what similar viewers also enjoyed.
- Content performance. How a video does when it is offered to viewers: whether they choose to click, watch or engage positively.
YouTube Help adds that the system analyzes each viewer's profile in real time, considering device, time of day and past habits. A second help page says the same video can rank differently on a phone in the morning and on a TV at night.
What the 2021 blog post adds
In 2021 a YouTube VP of Engineering wrote a long post on the official blog. Three claims from it are useful with numbers attached.
| Signal | What the post says about it |
|---|---|
| Clicks | A strong indication a viewer will find a video satisfying, but YouTube learned in 2011 that a click does not mean the video was watched |
| Watch time | Added in 2012 so a click on a misleading thumbnail does not count as success |
| Survey responses | Viewers rate a video from one to five stars, and only four and five stars count as 'valued watchtime'. A model predicts responses for viewers who do not answer |
| Shares, likes, dislikes | On average people are more likely to be satisfied by videos they share or like, and a dislike suggests they were not |
Signals named in YouTube's 2021 post on its recommendation system
It says the system learns from over 80 billion pieces of information it calls signals, that it does not run from a 'recipe book', and that 'the importance of each signal depends on you'. That last point is why nobody outside YouTube can publish a formula. It also says recommendations appear mainly on the homepage and in the Up Next panel, and that they drive a significant amount of viewership, even more than channel subscriptions or search.
What YouTube says it does not do
YouTube's own pages answer several of the most repeated worries.
| Belief | What YouTube says |
|---|---|
| Post at the right hour | Publish time is not known to impact a video's long-term performance. For a Premiere or a live stream, consider when your audience is active |
| A break hurts the channel | The algorithm does not penalize creators for taking a break |
| One flop damages the channel | The algorithm uses fresh performance data for each video, so a single experiment that does not land will not necessarily hinder the channel. Frequently sharing content that does not resonate can affect overall performance over time |
| Mixing Shorts, VODs and lives confuses it | Experimenting with formats will not inherently confuse the algorithm or hurt a channel. A viewer who likes your long videos might see your Shorts, though YouTube notes this is not common behavior |
| Monetized videos are favored | The recommendation algorithm does not prioritize videos based on whether they are monetized |
| Changing a thumbnail resets the video | The systems respond to how viewers react to the new title or thumbnail, not to the act of changing it |
| Uploading as unlisted first hurts | Uploading as unlisted before making a video public should not significantly impact its overall performance |
Common beliefs and what YouTube's pages say
Two nuances matter. YouTube also says long-term channel performance can be affected if a viewer consistently stops watching a channel's videos when they are recommended, or if viewers increasingly engage with other channels. And its advice on frequency is to prioritize consistent quality content over a high frequency of uploads.
Reading your own numbers
YouTube's click-through-rate FAQ gives the only benchmark it publishes: half of all channels and videos have an impressions click-through rate between 2% and 10%. New videos, and videos with under 100 views, can fall outside that range, and a video that gets many impressions, such as on the Home page, naturally has a lower rate. YouTube says to avoid checking the rate right after uploading and to avoid reacting to small changes.
A worked example, using invented numbers: a video is shown 10,000 times. At a 2% click-through rate it gets about 200 views, and at 10% about 1,000. The whole 2% to 10% range is normal for half of channels, so a single rate says little until you compare it with your own earlier videos that had similar traffic sources.
| If you see | YouTube's pages point to |
|---|---|
| High click-through rate, low average view duration, fewer impressions than expected | YouTube gives exactly this pattern as the sign of a clickbait thumbnail |
| Strong click-through rate and view duration, still few impressions | External factors: competition from other videos, topic interest and seasonality |
| Low click-through rate and fewer impressions than usual | A title or thumbnail change can help. If it does not, the topic may appeal to a smaller audience |
| Most views seem to come from a small core | YouTube says subscriber count reflects how many people subscribed, not the active audience, and that viewers often skip most videos in their subscription feed. Check Unique Viewers in the Audience tab |
Patterns and what YouTube's pages point to (our arrangement of YouTube's statements)
How Shorts differ
YouTube Help says the system has no opinion about what type of Short you make and does not favor a format. Shorts are ranked on performance and viewer personalization. It names three performance signals: the share of viewers who chose to view, average view duration and average percentage viewed, plus likes and post-watch surveys. The Shorts tab ranks by performance and relevance to each viewer, the Subscriptions tab shows the newest Shorts from subscribed channels, and creators who use sounds from the Audio Library can increase their chances of appearing on Sounds pages. Another help page says the Shorts feed may tune up on recency, which helps new content get discovered.
YouTube also says there is no minimum posting cadence for Shorts to perform, and that Shorts performance does not negatively impact long-form recommendations.
One change affects the numbers you see. YouTube says that from August 24, 2026, a view is counted from the first frame in every format, and the previous definition is now called an engaged view. It says recommendations and the metrics built on engaged views, such as click-through rate, average view duration and retention, are unchanged. Views before and after that date are not measured the same way, so compare across the date with care. Our guide to Shorts monetization covers qualified views.
- 01
Package for one viewer
Write the title and thumbnail for the person who would enjoy the video, and make sure the video delivers what they promise.
- 02
Judge a video by its retention curve
YouTube points to audience retention in Analytics to see where viewers leave, and says there is no ideal length.
- 03
Compare like with like
Look at click-through rate after enough impressions, and against videos with similar traffic sources.
- 04
Track your own patterns
The [Channel Audit](/tools/channel-audit) scores each recent upload against your own median so you can see what tends to work.
- 05
Treat a dip as a question first
Our [views drop guide](/blog/why-did-my-youtube-views-drop) gives a diagnosis order, and the [Shadowban Checker](/tools/shadowban-checker) is one input among several.
Confidence note
Everything above is YouTube's own account of its system, taken from its help pages and two blog posts, and we cannot verify it independently. The 2021 blog post is older than the help pages, so where they differ the help pages describe the current position. The table of patterns is our arrangement of YouTube's statements and not a diagnosis. Nothing public shows the signals for your channel or the weight of any signal.
FAQ
Does the time I post matter?
YouTube says publish time is not known to impact a video's long-term performance. It suggests considering your audience's active times for Premieres and live streams. See our best time to post guide for the details.
What is a good click-through rate?
YouTube says half of all channels and videos have an impressions click-through rate between 2% and 10%, and that the rate varies by content, audience and where the impression was shown. Compare your videos with each other over time.
Is there a 7-second rule or a seed audience test?
Guides often describe both, but they do not appear on the YouTube pages we read. YouTube names the signals it uses for Shorts, such as the share of viewers who chose to view and average view duration, and does not publish thresholds.
Do subscribers decide who sees my videos?
They are one signal among many. YouTube says recommendations drive more viewing than subscriptions or search, and that viewers often skip most videos in their subscription feed.
Will Shorts hurt my long-form videos?
YouTube says Shorts performance does not negatively impact long-form video recommendations, and that experimenting with formats will not inherently confuse the algorithm.
Does a break make the algorithm forget my channel?
YouTube says the algorithm does not penalize creators for taking a break, and suggests telling your audience so they can come back smoothly.
Related reading
Sources
- YouTube Help: YouTube's Recommendation System
- YouTube Help: Good to know about recommendations for YouTube's recommendation system
- YouTube Help: Search & discovery tips - Shorts
- YouTube Help: Impressions & click-through-rate FAQs
- YouTube Official Blog: On YouTube's recommendation system
- YouTube Official Blog: What are 'Engaged' Views on YouTube?