Recommended videos

Overview

Recommendations help you discover more of the videos you love, whether it’s a great new recipe to try or your next favorite song. The success of YouTube’s recommendations depends on accurately predicting the videos you want to watch. When our recommendations are at their best, they connect billions of people around the world to content that uniquely inspires, informs, and entertains.

You can find recommendations at work in two main places: your homepage and the “Up Next” panel.

• Homepage

Your homepage is what you see when you first open YouTube—it displays a mixture of personalized recommendations, subscriptions, and the latest news and information.

• Up Next

The Up Next panel appears when you’re watching a video. It suggests additional content based on whatever you’re currently watching, alongside other videos that we think you may be interested in.

How does YouTube’s recommendation system work?

We start with the knowledge that everyone has unique viewing habits. Our system then compares your viewing habits with those that are similar to you and uses that information to suggest other content you may want to watch.

Our recommendation system is constantly evolving, learning every day from over 80 billion pieces of information we call signals: for example, your watch and search history (if enabled), channel subscriptions, clicks, watchtime, survey responses, and sharing, likes, and dislikes.

• Clicks: Clicking on a video provides a strong indication that you will also find it satisfying. However, we learned in 2011 that clicking on a video doesn’t mean you actually watched it—which is why we also added watchtime in 2012.

• Watchtime: Your watchtime—which videos you watched and for how long—provides personalized signals to our system about what you most likely want to watch. Longer watchtime suggests that you will have found a video more valuable.

• Survey responses: To make sure you are satisfied with the content you’re watching, we measure what we call “valued watchtime”—the time spent watching a video that you consider valuable. We measure valued watchtime through user surveys that ask you to rate the video you watched from one to five stars, giving us a metric to determine how satisfying you found the content. Only videos that you rate highly with four or five stars are counted as valued watchtime. Based on the responses we get, we’ve trained a machine learning model to predict potential survey responses for everyone.

• Sharing, likes, and dislikes: On average, people are more likely to be satisfied by videos that they share or like. Our system uses this information to try to predict the likelihood that you will share or like further videos. If you dislike a video, that’s a signal that it probably wasn’t something you enjoyed watching.

• Personal activity signals: Personal activity signals like your channel subscriptions and watch and search histories make it easy to find videos you already like, improving your video recommendations. You can adjust what information is shared with recommendations in your privacy controls.

• Context signals: Context signals are determined by a user’s country and time of day. For example, this helps us show you locally relevant news.

The importance of each signal depends on you. If you’re the kind of person to share any video that you watch, including the ones that you rate one or two stars, our system will know not to heavily factor in your shares when recommending content. All of this is why our system doesn't follow a set formula, but develops dynamically as your viewing habits change.

How do YouTube’s recommendations help maintain a responsible platform?

Recommendations play an important role in how we maintain a responsible platform. They connect you to high-quality information and reduce the chances that you’ll see problematic content, and at the same time complement the work done by our Community Guidelines, which define what is and isn’t allowed on YouTube.

Limiting low quality content

Recommendations can be meaningful in helping stop the spread of harmful misinformation.

We’ve used recommendations to limit low-quality content from being viewed since 2011, and with a rise in the spread of misinformation in recent years, we’ve expanded the ways in which we use our recommendation system to curb the spread of harmful misinformation and “borderline” content - content that comes close to, but doesn’t quite violate our Community Guidelines.

To determine borderline content, evaluators around the world—trained using publicly available guidelines—look at whether the content is inaccurate, misleading or deceptive; insensitive or intolerant; and harmful or with the potential to cause harm. Any video classified borderline is not proactively recommended on YouTube unless you have subscribed to the channel.

We then use the results of these human evaluations to train our system to model their decisions, which means we can scale their assessments to all videos across YouTube. We are constantly working to improve our systems to reduce recommendations on borderline content and videos that could misinform users in a harmful way.

Raising quality content

In addition to reducing the spread of harmful misinformation and borderline content, we take the additional step of recommending authoritative videos to viewers on topics such as news, politics, medical, and scientific information.

We rely on human evaluators, trained using publicly available guidelines, who assess the quality of information in each channel and video. To decide if a video is authoritative, evaluators look at factors like the expertise and reputation of the speaker or channel, the main topic of the video, and whether the content delivers on its promise or achieves its goal. The more authoritative a video, the more it’s promoted in recommendations.

Controls to make recommendations more relevant to you

Our system sorts through billions of videos to recommend content tailored to your specific interests.

But of course, we also know not everyone wants to always share this information with us. So we’ve built controls that help you decide how much data you want to provide. You can pause, edit, or delete your YouTube search and watch history whenever you want.

We also provide ways for you to tell us when we’re recommending something you aren’t interested in. For example, buttons on the homepage and in the “Up next” section allow you to filter and choose recommendations by specific topics. You can also click on “not interested” to tell us that a video or channel is not what you wanted to see at that time.