TECHNOLOGY · AI · CYBERSECURITY · THE SYSTEMS SHAPING BUSINESS
YouTube says it will let viewers create custom feeds by describing the kind of videos they want to see. The feature uses Google's Gemini models to interpret a natural-language request and assemble a dedicated recommendation feed, which can then be pinned as a tab on the YouTube home page. A user might create a feed for a specific hobby, a commute or a temporary interest without retraining the main recommendation system through weeks of viewing behavior.
The idea sounds modest, but it changes the relationship between the user and one of the internet's most powerful ranking systems. YouTube's default feed has historically inferred intent from behavior. Custom feeds allow the viewer to state intent directly and create multiple recommendation contexts on purpose.
YouTube Is Making the Recommendation Feed Promptable
The immediate event is only the visible edge of a broader shift, and the details become more important when placed in that larger context.
RelatedBehavioral Personalization Is Powerful but Sticky
Recommendation systems learn from signals such as clicks, watch time, likes and subscriptions. That makes them good at identifying stable preferences, but it can also make them slow to adapt when the user's goal changes. Watching a handful of videos for a research project can distort a feed; a short-lived obsession can become a long-lived recommendation category.
Prompted feeds offer a way to compartmentalize those interests. A person can ask for a temporary feed about marathon training or a narrow professional topic without necessarily changing the identity the main algorithm has inferred. That gives the user a degree of control that traditional tuning sliders and 'not interested' buttons never fully provided.
A Custom Algorithm Is Still a Platform Algorithm
The phrase 'build your own algorithm' should not be taken literally. Users are not choosing ranking weights or inspecting recommendation code. They are supplying a natural-language specification that YouTube interprets using its own models and policy systems. The platform still decides which candidate videos are eligible, how quality and safety are evaluated and how commercial content is surfaced.
That distinction matters for transparency. A prompt can make a feed feel user-controlled even when large parts of the ranking process remain opaque. YouTube will need to make it easy for people to understand why results appear and to refine a feed when the model interprets the request too broadly.
The Feature Could Reshape Discovery for Creators
Custom feeds may create new paths for niche content. A creator who does not fit a large category could still appear if a viewer describes an interest semantically. That could benefit specialized educational, technical or hobby channels whose audiences are clear in language but hard to identify through conventional demographic or engagement clusters.
It could also intensify competition around metadata and content understanding. If AI models become the layer translating prompts into recommendations, creators will care about how accurately the system understands their videos. Titles and thumbnails may remain important for clicks, but semantic representation becomes more important for being selected in the first place.
Custom feeds are a small but meaningful reversal: instead of the platform deciding what the user is, the user can describe what the feed should become.
User-Programmable Feeds May Become a Standard Product Pattern
Other platforms are likely to experiment with similar controls because natural language is an efficient way to express a complicated preference. A user can specify topic, tone, length, freshness and context in one sentence rather than manipulating a dozen settings. The model turns that request into a dynamic ranking policy.
There are still risks. A tightly specified feed can become an intentional filter bubble, and malicious prompts may be used to seek harmful or misleading material. Safety systems will therefore remain above the user's instructions. The most useful implementation will balance user agency with clear boundaries rather than pretending the user has absolute control.
YouTube's experiment is a sign that recommendation systems are becoming conversational products. The feed is no longer only something a platform computes in the background. It can become something the user actively describes, edits and maintains—an important shift in how people may interact with algorithmic media.
Custom Feeds Make Recommendation Algorithms More Explicit
YouTube's prompted feeds give users a rare ability to state the objective of a recommendation system instead of merely influencing it through watch history. That can be useful when a person wants a temporary mode—learning a language, following a sports tournament or avoiding a topic—without permanently reshaping the main home feed. In effect, the user creates a lightweight algorithm for a specific purpose.
The feature also exposes a design tension. Recommendation systems are effective partly because they infer preferences people do not articulate. A natural-language instruction is explicit but incomplete; the system still has to decide how broadly to interpret it, how much novelty to introduce and when to stop showing similar content. The AI layer therefore does not remove algorithmic judgment. It gives users a more understandable way to steer it.
If the model works well, other consumer platforms are likely to copy the pattern. Shopping, news, podcasts and social feeds all contain the same mismatch between rigid filters and complex human intent. The important product shift is from a single opaque feed toward multiple user-defined views that can be created and discarded conversationally.
What to Watch Next
What happens next with YouTube Will Let Viewers Shape Their Recommendation Algorithm With AI will depend on how the capability behaves outside carefully selected examples. Real users bring ambiguous instructions, incomplete data and edge cases that are difficult to represent in benchmark sets. The strongest products will make uncertainty visible, recover gracefully from failure and give people an easy way to inspect or reverse what the system did. Reliability is becoming a competitive feature in its own right.
The data boundary will be equally important. More capable assistants want more context, and more context can produce better results, but every new source creates questions about consent, retention and cross-use. Consumer products need controls people can understand without reading a policy manual. Enterprises need administrative rules that can be enforced across thousands of users. The companies that solve both layers will have an advantage as AI moves deeper into everyday work.
There is also a distribution question. A model may be technically impressive but economically weak if users must leave the tools where they already spend their time. Conversely, an embedded assistant can gain enormous reach even when a specialist competitor performs better on a narrow benchmark. YouTube Will Let Viewers Shape Their Recommendation Algorithm With AI should therefore be watched as part of a larger contest over who owns the interface through which people ask computers to do work.
Over time, the market will likely separate features that are merely generated from experiences that are genuinely agentic. The latter maintain state, use tools and complete multi-step objectives. That increases value and risk simultaneously. As the distinction becomes clearer, product leaders will need to decide which actions remain suggestions, which can be automated and which require explicit human authorization. Those governance choices will shape adoption as much as model intelligence.
