YouTube Recommendations and the Watch-Next Chain
You chose the first video. The algorithm chose every one after that.
You searched for one video. A tutorial, a review, a clip someone sent you. You watched it. It was exactly what you wanted. Then the next video started automatically, chosen by the algorithm, and it was interesting enough to keep watching. Then the next one. And the next. An hour later, you are watching something you would never have searched for, and the original video is a distant memory.
The first video was yours. Everything after it was the algorithm’s.
How the recommendation system works
YouTube’s recommendation system uses deep neural networks to select videos based on your watch history, the behavior of similar users, video metadata and engagement signals [2]. The system is optimized for a specific outcome: watch time. It selects videos most likely to keep you on the platform.
This optimization is effective. Research on viewing patterns shows that recommendations account for a substantial share of total viewing time, and that the algorithm shapes viewing trajectories, guiding users from their initial choice into chains of related and sometimes progressively more niche content [5].
The agency problem
A qualitative study of YouTube users found that autoplay and algorithmic recommendations were the design features most strongly associated with reduced sense of agency [1]. Users described watching videos they had not chosen, continuing past the point of interest, and feeling that the platform was directing their viewing rather than facilitating it.
The experience is distinct from choosing to watch television. When you select a program, you are making a decision about what to watch. When the algorithm selects the next video and autoplay starts it, the decision is made for you. The feeling of continuous choice, “I could stop at any time,” is undermined by the reality that stopping requires action while continuing requires nothing.
A fictional example
The following is an illustrative example, not a real person’s story.
Marcus searches for a guitar tutorial. He watches a ten-minute lesson, which is what he needed. The next video begins automatically: a related tutorial by a different instructor. It is decent, so he watches. The third video is a performance by a professional guitarist. The fourth is a documentary about the history of the electric guitar. The fifth is a talk show clip featuring a musician. Marcus closes the app ninety minutes after opening it, having learned the chord he needed in the first ten minutes.
The algorithm served Marcus content he found interesting. That is exactly what it is designed to do. The question is whether ninety minutes of interesting content, which he did not choose and would not have sought, was a good use of his evening.
Watching intentionally
- Turn off autoplay. In YouTube’s settings, disable autoplay. This means each video ends with a still screen rather than a countdown, which restores the decision point.
- Use search, not the home page. The home page is a recommendation engine. The search bar is a tool. If you have something to watch, search for it directly and avoid the home page.
- Close when you are done. When the video you searched for is finished, close the app or tab. The recommended videos in the sidebar are designed to catch your eye. Closing the window removes them.
- Use a playlist. If you want to watch multiple videos, create a playlist in advance. Watch the playlist and stop when it ends. This gives you a curated session with a defined end.
- Notice the “one more” feeling. When a recommended video starts and you feel the pull to watch, recognize the moment: “The algorithm chose this, not me. Do I actually want it?” Sometimes the answer is yes, and that is fine. The point is that you answered rather than drifted.
If the watch-next pattern is driven by autoplay specifically, autoplay and the next-episode trap examines the design feature in more detail across streaming platforms.
When to talk to someone. If you regularly lose hours to recommended videos, if the pattern is affecting your sleep, work or relationships, or if you feel unable to stop despite wanting to, talk to a doctor or therapist. Extended, compulsive viewing can be a habit issue on its own or a symptom of something else that benefits from professional support.
Frequently asked questions
How does YouTube's recommendation algorithm work?
YouTube recommends videos based on your watch history, what similar users have watched, video engagement metrics and session context. The algorithm optimizes for watch time, meaning it selects videos most likely to keep you on the platform. This does not mean the videos are bad; it means they are selected for engagement, not for your stated goals.
Does YouTube's algorithm create filter bubbles?
Research is mixed. The algorithm tends to recommend content similar to what you have watched, which can narrow the range of content you encounter. Whether this constitutes a 'filter bubble' in a meaningful sense depends on how varied your initial viewing is and whether you actively seek diverse content.
How do I stop watching recommended videos?
Turn off autoplay. Use search to find specific videos rather than browsing the home page. Close the tab or app when the video you chose is finished. These steps return the viewing decision to you instead of the algorithm.
Is YouTube designed to keep you watching?
YouTube's business model depends on watch time, and the recommendation system is optimized to maximize it. This is not a conspiracy; it is the incentive structure of an advertising-supported platform. The design works as intended. Your job is to decide whether its intentions align with yours.
Sources and further reading
- Lukoff et al. — How the design of YouTube influences user sense of agency
Proceedings of CHI 2021. Autoplay and recommendations reduced users' sense of deliberate choice. DOI: 10.1145/3411764.3445467.
- Covington, Adams and Sargin — Deep neural networks for YouTube recommendations
Proceedings of RecSys 2016. Technical overview of YouTube's recommendation system architecture. DOI: 10.1145/2959100.2959190.
- Montag et al. — Addictive features of social media/messenger platforms and freemium games against the background of psychological and economic theories
International Journal of Environmental Research and Public Health, 2019;16(14):2612. Algorithmic recommendations identified as engagement-maximizing design feature. DOI: 10.3390/ijerph16142612.
- Lyngs et al. — Self-control in cyberspace: applying dual systems theory to a review of digital self-control tools
Proceedings of CHI 2019. Tools that modify the feed or remove recommendations were among strategies reviewed. DOI: 10.1145/3290605.3300361.
- Hosseinmardi et al. — Examining the consumption of radical content on YouTube
Proceedings of the National Academy of Sciences, 2021;118(32):e2101967118. Study of viewing patterns; recommendations shaped viewing trajectories. DOI: 10.1073/pnas.2101967118.
Sources checked October 16, 2026. This is an editorial summary, not a systematic review or individual advice.