YouTube is starting to turn one of the internet’s most powerful recommendation systems into something viewers can actively steer with a prompt.
At Made on YouTube 2026, the platform introduced Custom Feeds, a feature that lets users describe the kind of videos they want to see and save the result as a dedicated feed on the YouTube homepage. Instead of relying only on behavioral signals such as watch history, likes and searches, users can explicitly tell the system what they want next.
Custom Feeds turn recommendation into a conversation
In YouTube’s official September 23 viewer announcement, the company said users will be able to create dedicated homepage tabs around specific moods, routines or interests. A viewer might ask for witty podcasts for a 30-minute commute, long documentaries for an evening wind-down or new DIY creators to discover.
The important change is not that YouTube suddenly has personalization. It has been one of the defining characteristics of the platform for years. The change is that the user can now provide a direct natural-language instruction instead of waiting for the recommendation system to infer intent from past behavior.
The feed does not replace YouTube’s main algorithm
TechCrunch reported that Custom Feeds are powered by Google’s Gemini AI model and will appear as saved tabs on the homepage. Users can create multiple feeds and describe them in detail, including what content to prioritize or exclude.
The standard YouTube Home feed remains in place. Custom Feeds sit alongside it, which means YouTube is adding an explicit layer of user control without abandoning the behavioral recommendation system that already drives discovery across the service.
This is part of a larger shift from inferred taste to declared taste
For years, recommendation systems largely watched what users did and tried to predict what they would do next. The new generation of AI interfaces gives users another option: simply state the desired outcome.
That distinction sounds small, but it changes the relationship between a user and an algorithm. Behavioral recommendation can trap people inside patterns created by old interests, accidental clicks or one-off viewing sessions. Prompted feeds can give the user a way to declare a new direction immediately.
YouTube has more than 20 billion videos to organize
YouTube says its corpus contains more than 20 billion videos and that more than 20 million new videos are uploaded each day. At that scale, search alone cannot organize the experience. Recommendation becomes an interface layer between viewers and an effectively unmanageable media library.
Custom Feeds are therefore less about creating one perfect algorithm and more about allowing one person to maintain several algorithms for different contexts: work, entertainment, learning, commuting, hobbies or whatever temporary obsession happens to matter that week.
Creators now have to think about prompts as a discovery surface
The feature also creates a new question for creators. If viewers increasingly describe the type of content they want instead of navigating through broad categories, titles, thumbnails, descriptions and topic signals may be interpreted by AI systems trying to match videos to a much more specific request.
BitcoinVersus.Tech recently covered YouTube adding AI draft feedback and video A/B testing for creators, AI voice translation moving into YouTube live streams, and LISA using YouTube Premium as part of an IMAX-to-streaming release strategy. Together, those changes show YouTube pushing AI into both sides of the platform: how creators make media and how audiences discover it.
The algorithm is becoming editable
The most interesting cultural change is that recommendation systems are becoming visible as products in their own right. Users are no longer only asking, “Why did the algorithm show me this?” Platforms are beginning to offer a second question: “What do you want the algorithm to show you instead?”
Custom Feeds do not eliminate the black box behind recommendations, and YouTube still controls the models, ranking systems and available inventory. But giving users an explicit prompt layer is a meaningful shift away from a media environment where the only way to train the recommendation engine was to keep watching until it figured you out.
BitcoinVersus.Tech
Advertisement
Editor’s Note
We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support to help further secure the integrity of our research initiatives, please donate here: 3C9o19EH5HSiwEPyCTmEKzxhNCbo2X6TTb
BitcoinVersus.tech is not a financial advisor. This media platform reports on financial subjects purely for informational purposes.
Leave a comment