YouTube Algorithm

• Comment: Whole bunch of WP:OR. Also lines like "If you need a whole guide to grow your YouTube channel on YouTube, click here" show this is likely a not so incognito advertisement. Sulfurboy (talk) 10:15, 22 August 2026 (UTC)

• Comment: In accordance with Wikipedia's Conflict of interest guideline, I disclose that I have a conflict of interest regarding the subject of this article. ActStuffOGWiki (talk) 09:59, 22 August 2026 (UTC)

---

The general recommendation system used by YouTube.

Overview

The YouTube algorithm (also called YouTube recommendations, YouTube recommendation model, YouTube feed, YouTube recommender system and YouTube video suggestion engine) is the general term for the computer systems that YouTube uses to decide which videos to show people. It helps select videos for the Home page, the Up Next section, Shorts, and search results. These systems are designed to show each viewer videos that they may want to watch and enjoy.

The YouTube algorithm is a group of systems that work together. Different parts of YouTube may use different signals and methods. For example, the Home page usually depends on a viewer’s wider interests, while Up Next is more closely connected to the video currently being watched.

In reality, YouTube does not publish the complete formula behind its recommendations. The company changes its systems over time, and the exact factors can vary according to the viewer, the type of video, and the part of YouTube being used. So, you kind of have to learn the new formula behind its recommendations.

Purpose

YouTube’s recommendation system tries to connect viewers with videos that are relevant and enjoyable to them. Its actual goal is, in fact, simply to produce as many clicks as possible. YouTube also says that it aims to improve long-term viewer satisfaction, meaning that it wants people to find videos they consider useful, interesting or entertaining.

This is actually important because a click does not always mean that a viewer liked a video. A person may click a video because its title or thumbnail looks interesting, but they might leave after a few seconds because the video does not meet their expectations. For this reason, YouTube considers what happens after the click as well as whether the click happens. The most important things according to YouTube are viewer satisfaction, Click-Through-Rate (CTR) and average view duration.

The system also helps viewers discover videos from channels they do not already know. Without recommendations, viewers would have to find most videos through direct searches, subscriptions, or links shared by other people. Since YouTube contains a very large amount of videos, recommendations help viewers navigate this large collection.

Main Parts

The YouTube algorithm actually isn't a singular system but a collection of recommendation engines that decide which videos appear on different parts of the platform such as the homepage, Shorts feed, search results, and suggested videos. It works by first gathering a large set of possible videos and then ranking them for each viewer based on the CTR, retention rate and viewer satisfaction, and then it sees how well the video performs when shown. YouTube uses around 80 billion signals for your recommendations.

Home Page

The Home Page (also called the Browse feed) is the main, personalized landing screen that appears when a person opens YouTube. It is designed to help viewers discover new content they are likely to enjoy, based mostly on their past behavior. The Browse Feed builds the feed using 80 billion+ signals such as the viewer’s watch history, videos they have clicked on before, channels they are subscribed to, and how recently those channels have uploaded a video. It also considers how often a viewer returns to certain channels and how satisfied they have been with similar videos in the past. A very important thing the 2026 system does better than before, is the fact that it doesn't reward clickbait or keyword stuffing as much, which is great. The system favors videos that have good performance from non-subscribers, high click-through rates on small thumbnails, and that lead viewers to watch more content on YouTube afterward. To get to the Home Page, you need to post a video to the right people, at the right moment, at the right time.

Up Next

The Up Next section (also called Suggested Videos) appears alongside or below the video a viewer is currently watching and recommends what to play next. Unlike the Home Page, this system is optimized for 'session continuation' rather than discovery, meaning it tries to keep the viewer watching by suggesting the most logical next video. The strongest signal is what other viewers have watched after the same video, known as the co-view signal. The algorithm also prioritizes other videos from the same channel, videos in the same series or playlist, and content that is topically similar to the current video. Personal watch history has less influence here than in the Home Page, because the goal is to answer; “what is good to watch after this?” rather than “what does this user usually like?” The system scores candidate videos using content similarity and probability that someone who watched the current video will continue their session time by watching the suggested video on the Up Next section.

Shorts Feed

The Shorts feed is an entirely separate recommendation engine which is dedicated to short-form videos. The shorts-form algorithm operates independently from the long-form algorithm. Its more dominant recommender system in 2026 is swipe-through rate (or equivalently, swipe-away rate), which measures how many viewers watch a Short to the end instead of swiping away. A swipe-through rate above roughly 50% is considered strong, while below about 30% usually stops a Short from being shown to larger audiences. The system also looks at completion rate, rewatches, and average view duration, with high retention (often around 65–70% or more) helping a Short escalate to wider distribution. Distribution happens in stages: a Short is first shown to a small seed audience, and if it clears swipe and retention thresholds, it is released to progressively larger groups. Top-performing Shorts may also be promoted beyond the Shorts feed, appearing on the Home Page or in the Up Next section next to related long-form videos.

Search Results

The Search results page categorizes videos based on how well they match the user’s query and how satisfying they are for that search intent. The algorithm considers factors such as the video’s title, description, and tags, as well as how often viewers click on it for a given term and how long they watch it. Videos that keep viewers engaged and lead them to watch more related content tend to rank higher over time. Unlike the Home Page and Up Next sections, Search Results relies more heavily on the relevance of the content to the query itself, though personalization still plays a role based on the viewer’s history and preferences. This page is especially more important for long or short form content that answers specific questions or teaches specific skills.

Ranking System

The YouTube algorithm ranks videos by predicting how likely each viewer is to click on a video and how satisfied they will be if they watch it. It does this using a two-stage process: first selecting a large set of candidate videos, then ordering them with machine learning models that score each video for that specific viewer. The ranking is not the same for everyone; it changes based on the viewer’s history, the main parts (Home, Up Next, Search, or Shorts), and how the video performs when shown.

Viewer Satisfaction

Viewer satisfaction is the most important ranking signal in 2026 and has replaced raw watch time as the primary goal of the system. YouTube measures satisfaction through a combination of post-watch surveys, likes, shares, repeat views of the same channel, and whether viewers return to the channel within a week. The platform feeds millions of survey responses directly into its recommendation models to predict how satisfied each viewer will be with a video. A short video that is highly satisfying and shared widely can now outrank a longer video with more total watch time but lower satisfaction.

Watch Time and Retention Rate

Watch time remains a core signal but now functions as a supporting factor rather than the dominant one. The algorithm considers total minutes watched, average view duration (the percentage of the video watched), and how much of the video is viewed in the first 30 seconds. High retention, especially early in the video, signifies a strong hook and helps the video earn more impressions. Videos that keep viewers on YouTube longer by leading them to watch more content (session time) are also rewarded.

Click-Through Rate

Main article: Click-through rate.

Click-through rate (CTR) measures how often viewers click on a video after seeing its thumbnail and title. A higher CTR generally leads to more impressions and better rankings, but absolute CTR is not actually comparable across YouTube channels because it depends on context and audience. The algorithm tests every new video with a small audience and uses CTR along with early retention to decide whether to push it wider. Clickbait can generate more clicks, but poor retention ultimately doesn't do good to a video’s ranking because the system learns that viewers are not satisfied.

Personalization

Personalization plays a major role in how videos are ranked for each viewer. The recommendation model uses a viewer’s watch and search history, subscriptions, likes, and the not interested feedback to adjust recommendations. It also considers how often a viewer returns to certain channels and what topics they have engaged with in the past. On parts like the Home Page, personalization has more weight, while on Search, relevance to queries is more important. In 2025, YouTube started driving more towards AI personalization than before.

Shorts-Specific Signals

The Shorts feed uses a separate ranking system with its own dominant signals. Swipe-through rate (the percentage of viewers who watch a Short to the end instead of swiping away) is the most important factor. Completion rate, rewatches, and average view duration also influence how widely a Short is distributed. Shorts that clear retention thresholds are quietly showed to progressively larger audiences and may also appear on the Home Page or in Up Next section.

History

The YouTube algorithm has evolved from a simple chronological recommendation system in 2005 to a more complex, slightly AI-powered system that recently prioritized viewer satisfaction over raw watch time. This history can be divided into several major eras, each defined by a shift in how YouTube measured as well as how YouTube optimized for user engagement.

2005-2011

When YouTube launched in 2005, the homepage displayed videos in chronological order, with the most recently uploaded content appearing first. As the platform grew, this system became impractical, and YouTube began experimenting with ranking videos based on view count. During this period, the most-viewed videos were promoted a lot, which led to a focus on clickbait titles and thumbnails designed to maximize clicks rather than viewer enjoyment. This era was characterized by a simple metric; the more views a video had, the more likely it was to be recommended.

2012-2015

In 2012, YouTube announced a major shift in its algorithm, moving from view count to watch time as the primary ranking signal. This change was made to encourage creators to produce content that kept viewers watching for longer sessions, rather than just attracting clicks. The platform began to favor videos and channels that generated more total minutes of watch time, which led to the rise of long-form content and series. During this period, YouTube also introduced features like subscriptions and notifications, which helped personalize the user experience.

2016-2019

Around 2016, YouTube began replacing its older recommendation systems with deep neural networks built on Google’s machine-learning infrastructure. In September 2016, engineers from Google Brain published a paper describing a two-stage system: a candidate network that narrowed YouTube’s catalog down to a few hundred videos for each viewer, followed by a ranking network that ordered those videos based on hundreds of features. This era saw the introduction of a more sophisticated personalization; with the algorithm taking a viewer's watch history, search history, and other behavioral signals into consideration. The 'Adpocalypse' in 2017 also led to tighter monetization rules and changes in what content was promoted. Many YouTubers said it ruined their revenue.

2020-2025

In the early 2020s, YouTube began to place more emphasis on viewer satisfaction, recognizing that watch time alone did not always correlate with a positive user experience. The platform started to incorporate signals such as likes, dislikes, shares, and survey responses into its recommendation models. This period also saw the introduction of Shorts in 2020, which required a separate recommendation engine optimized for short-form content. By 2025, YouTube had begun to test and implement systems that weighted satisfaction signals more than raw watch time.

Present Changes

The YouTube algorithm in 2026 is a collection of slightly AI-powered recommendation systems that operates across multiple surfaces, including the Home Page, Up Next section, Shorts feed, and Search results, each with its own ranking signals. YouTube has confirmed that satisfaction signals now outweigh traditional metrics like total watch time, marking a huge shift in how videos are recommended.

Primary Signal is User Satisfaction

In 2026, viewer satisfaction is the most important ranking factor for YouTube recommendations. This shift was publicly confirmed by YouTube in 2025 and fully implemented in 2026. Satisfaction is measured through four main signals: survey responses from viewers who rate videos on a five-star scale, repeat views where people come back to watch a video again, shares where someone sends a video to another person, and session continuation where the viewer keeps watching YouTube even after the video ends. The formula YouTube uses is; watch time + user satisfaction = session contribution, meaning that a video's ability to keep viewers on the platform is more important now than how long they watch a single video.

First 30 Seconds Rule

In one year, YouTube elevated the first 30 seconds of a video from a not-so-important metric to an important ranking input. This means that what happens in the opening half-minute of a video now plays a decisive role in whether the algorithm will recommend it to more viewers. Videos that fail to engage viewers in the first 30 seconds are less likely to be promoted, even if they have strong overall watch time. Creators are highly advised to focus on creating engaging and interesting intros and clear value propositions in the opening moments of their videos.

Shorts Feed is Separated from Long Form Feed

The Shorts algorithm in 2026 operates on a completely separate recommendation algorithm from long-form content. This 'decoupling' means that performance in Shorts does not directly affect how a channel's long-form videos are recommended, and vice versa, which is great. The Shorts algorithm prioritizes swipe-through rate (the percentage of viewers who watch a YouTube Short to the end instead of swiping away), completion rate, and rewatches. This separation allows creators to specialize in either format without worrying about cross-format performance penalties.

Artificial Intelligence Disclosure

Starting in May 2026, YouTube introduced new internal signals to help identify AI-generated content. If a creator does not specify whether they used AI, the platform may auto-label the content based on its own detection systems. For long-form videos, the AI disclosure label now appears directly below the video player, above the description. For Shorts, the label appears as an overlay on the video itself.

New AI Features (Made on YouTube 2025)

YouTube also introduced several AI-powered features in 2026 that affect how content is created and discovered. "Ask YouTube" is a new Gemini-powered conversational search feature that allows users to find videos using natural-language queries. AI video remixing with Gemini Omni enables background swaps, outfit changes, and in Shorts, there is a very similar blending. Veo 3 inside Shorts allows creators to animate photos, clips, and generate AI Shorts faster. YouTube also launched a "Likeness Detection" hub that helps creators manage their digital identity and control how their likeness is used in AI-generated content.

Controversies

Main article: Criticism of YouTube.

The YouTube algorithm has been the subject of numerous controversies since the platform’s launch, with critics arguing that its design prioritizes engagement and watch time over user well-being and content quality. These controversies span issues related to children’s safety, misinformation, political radicalization, and the impact on creators’ livelihoods. While YouTube has made changes in response to some of these criticisms, many concerns remain unresolved.

How the Algorithm Works

The YouTube algorithm is a two-stage recommendation system that selects and ranks videos for each viewer from billions of options in under 200 milliseconds. It operates by first narrowing down the entire catalog to a few hundred candidate videos, then scoring and ordering those candidates to display the most relevant 20 or so on the user’s screen.

Stage 1: Candidate Generation

The first stage of the YouTube algorithm is candidate generation, a.k.a retrieval. In this stage, the system starts with billions of videos in the database and uses a two-tower neural network to quickly filter them down to a few hundred reasonable candidates for each viewer. The goal here is surprisingly not to find the best videos, but to remove those that have no chance of being relevant. This stage uses collaborative filtering search to identify videos that match a viewer’s watch history, search history, subscriptions, and the behavior of similar viewers. The system also considers factors like video freshness, first 30 seconds rule and regional preferences to ensure a diverse set of candidates.

Stage 2: Ranking

The second stage is ranking, where the few hundred candidate videos are passed to a massive neural network for precise scoring. This stage uses a richer, more expensive model to evaluate each candidate based on hundreds of features, including click-through rate (CTR), average view duration, satisfaction signals, and session watch time. The ranking model predicts the expected valued watch time (EVWT) for each video, which is a measure of how much watch time the video is likely to generate while accounting for viewer satisfaction. Videos are then assigned a heavy score, and only the top 20 make it to the user’s screen. This stage is where individual video performance metrics matter most, with stronger watch time and retention signals leading to higher rankings.

Personalization

The algorithm is personalized. Meaning, it is giving recommendations to each viewer based on their unique behavior and preferences. It considers a viewer’s watch history, search history, demographic signals, and the patterns of similar viewers to generate a set of candidates. However, the system also balances personalization to avoid creating filter bubbles, where users are only shown content that aligns with their existing interests. This is achieved by including a mix of familiar and new content in the candidate set, ensuring that viewers are exposed to a variety of topics and creators. The algorithm also accounts for the cold start problem, where new videos or channels have limited historical data, by using collaborative filtering and other techniques to make initial recommendations.

See Also (Wikipedia)

• Recommender system
• YouTube Creator Awards
• Timeline of online video
• YouTube Rewind
• PewDiePie vs T-Series
• Click-through rate
• TikTokification
• YouTube
• Censorship of Google

See Also (External)

• 5 YouTube algorithm changes every creator should expect in 2026 – MilX
• YouTube wants to be TikTok — and it's bad for business
• YouTube Average View Duration: What Is a Good AVD?
• YouTube Revenue for Full-Year 2025 Topped $60 Billion, Making Video Platform Bigger Than Netflix
• YouTube will pay crores to creators if they don't work with Netflix, those who do may face consequences
• YouTube Algorithm 2026: 7 Ranking Signals Every Creator Must Know

This article "YouTube Algorithm" is from Wikipedia. The list of its authors can be seen in its historical and/or the page Edithistory:YouTube Algorithm. Articles copied from Draft Namespace on Wikipedia could be seen on the Draft Namespace of Wikipedia and not main one.

---