--- title: "How YouTube Recommendations Work (and How to Steer Them)" description: "Why does YouTube show you what it shows you? A plain-language look at what YouTube itself says about its recommendation system, what's still unknown, and how to use that to steer your feed." url: https://boslaai.co/blog/how-youtube-recommendations-work language: en --- # How YouTube Recommendations Work (and How to Steer Them) > Why does YouTube show you what it shows you? A plain-language look at what YouTube itself says about its recommendation system, what's still unknown, and how to use that to steer your feed. _The BOSLA AI team ยท 2026-10-16_ You watch one cooking video, and suddenly your Home page is full of recipes. You search for a phone review once, and comparisons follow you for a week. It can feel like the algorithm knows you โ€” or like it has completely misunderstood you. Neither is quite true. YouTube's recommendations follow patterns in what you and people like you do. Once you understand the basic inputs, the feed stops feeling random, and you get a much better sense of how to steer it. This article explains what YouTube itself has said about its recommendation system, what remains unknown, and what that means for you in practice. ## What "the algorithm" actually is There isn't one single algorithm. YouTube describes a recommendation system that works across different parts of the app, mainly: - **Home:** a mix of personalized recommendations, your subscriptions, and news when you're signed in. - **Up next:** suggestions while you watch, based on the current video and what YouTube thinks you might like. Search works differently. When you search, you've already told YouTube what you want. On Home, YouTube has to guess. ## What YouTube says it looks at YouTube has published explanations of its system, most notably a 2021 post by its VP of Engineering, Cristos Goodrow, and its "How YouTube Works" pages. The main inputs it describes are: ### What you watch Your watch history is one of the clearest signals. Watching a video, especially watching most of it, tells YouTube something about your interests. This is also why one curious click can reshape your feed for a while. ### Clicks and watch time โ€” but not on their own According to YouTube, early recommendations leaned heavily on clicks. That rewarded videos people clicked but didn't actually enjoy, so in 2012 YouTube started weighting **watch time** as well. The company says this change caused an immediate 20% drop in views, which it accepted because it believed the videos being recommended were more valuable to viewers. ### Satisfaction surveys Watch time alone isn't perfect either: you can watch something and still regret it. So YouTube asks some viewers to rate videos they watched. It describes counting only highly rated watch time as **"valued watch time,"** and using models to estimate how satisfied people would be, including people who never fill out a survey. ### Likes, dislikes, shares and subscriptions These also feed in, but YouTube says their weight depends on the person. For example, if someone shares nearly everything, each share tells YouTube less about that person's real preferences. ### Your direct feedback Choosing **Not interested** or **Don't recommend channel** sends a direct signal about what you don't want. YouTube says this feedback may be used to tune your recommendations. ### Quality and responsibility signals YouTube says it also considers channel and video quality, using human evaluators and automated classifiers. For news and similar topics, it says it promotes authoritative sources and reduces recommendations of "borderline" content โ€” videos that come close to breaking its rules without clearly doing so. ## What we don't know YouTube's public explanations describe the main ingredients, not the full recipe. The exact weight of each signal, how it changes over time, and how it differs between Home, Up next and Shorts aren't public. Many third-party guides make confident claims about "the algorithm" that go beyond what YouTube has actually said. It's also worth noting that independent research has questioned how much control users have. A [2022 Mozilla study](https://www.mozillafoundation.org/en/research/library/user-controls/) found that YouTube's controls do influence recommendations, but less than many people expect. So the honest summary is: we know the main inputs, and we know the system is trying to predict what you'll watch and be satisfied with. Beyond that, treat any precise claim with caution. ## What this means for you: 5 practical takeaways Understanding the inputs gives you a clearer picture of what you can change. 1. **Your watching is the strongest vote.** What you watch, and how long you watch it, likely matters more than any single button. Watching what you actually want is the most reliable way to steer your feed. 2. **One-off curiosity has a cost.** If you don't want a topic in your feed, remove it from your watch history afterward, or pause history before watching. 3. **Use direct feedback consistently.** "Not interested" and "Don't recommend channel" work best as a habit, not a one-time cleanup. 4. **Answer the surveys honestly.** If YouTube asks how satisfied you were with a video, your answer is one of the few ways to tell it the difference between "I watched this" and "I'm glad I watched this." 5. **Search when you know what you want.** Starting from search instead of Home means you're telling YouTube your intent instead of letting it guess. For step-by-step instructions on each control, see our guide to [customizing your YouTube feed](https://boslaai.co/blog/how-to-customize-your-youtube-feed). If you use YouTube for learning, our guide to [studying on YouTube without distractions](https://boslaai.co/blog/how-to-use-youtube-for-studying) puts these ideas into a simple workflow. ## FAQ **Does YouTube recommend videos just to keep me watching longer?** YouTube says it moved beyond raw watch time toward measures of satisfaction, including surveys. How these are balanced isn't public, so it's fair to say the system aims to predict what you'll watch and value, while recognizing that engagement is still a major input. **Why do I see things I've never shown interest in?** Recommendations also draw on what people with similar viewing patterns watch, on trends, and on your recent activity. A topic you've never searched can still appear because it's popular with viewers like you. **Does disliking a video help?** YouTube says likes and dislikes are among the signals it uses, but their weight varies. If you don't want more of something, "Not interested" or "Don't recommend channel" are more direct. ## Steering, not surrendering The recommendation system responds to what you do. That doesn't mean you have full control, but it does mean your choices count. Knowing which signals matter helps you spend your attention on purpose. We're building BOSLA AI around the same idea: you should be able to say what you want more of and less of, clearly and directly. [Read the principles behind BOSLA AI](https://boslaai.co/blog/your-choices-not-an-algorithm), or [create a free account](https://boslaai.co/sign-up) and try it, free in beta.