YouTube Will Let Users Build Their Own Algorithm With AI

YouTube is rolling out custom feeds, a feature that lets users describe in their own words what they want to watch instead of accepting the default recommendation stream.

Under the hood, YouTube uses Gemini to turn that description into a personalized feed built around the request. Part of the ranking logic therefore moves from a platform-controlled black box to something a viewer can state and refine in plain language.

The framing in TechCrunch's headline, building your own algorithm, captures that shift. Where viewers previously influenced recommendations mainly through behavior signals such as watching, liking, and subscribing, custom feeds add an explicit, language-based input.

Putting Gemini directly into feed creation is also a notable placement for Google's model inside a core consumer product. Rather than adding another chat box to the interface, the company is letting the model shape what content people actually discover and watch.

For creators, describable recommendations add a dimension that does not depend entirely on inferred watch-time and engagement metrics: if a viewer can ask for a particular kind of programming, matching content becomes findable on the viewer's own terms.

Several details remain unclear from the published information. Which regions and how many users get the feature first, what the rollout timeline looks like, and whether viewers can see or adjust Gemini's ranking logic are all still open questions.

What to watch next: whether custom feeds run alongside YouTube's existing recommender as a separate consumption entry point, whether Gemini-generated rankings are explained or adjustable, and whether Google extends the same pattern to other content products.