YouTube Adds Custom Feeds That Build Themselves From a Written Request

YouTube is introducing custom feeds that let viewers type out what they want to watch and have the platform assemble a feed around that request, with Google's Gemini models handling the interpretation. The feature, reported by TechCrunch, marks a shift for a recommendation system that has spent years refining passive signals — watch time, clicks, session history — into a product users can now steer with an explicit instruction.

The mechanics are straightforward on the surface: a user describes the videos they want, and Gemini parses that description into a feed rather than relying solely on behavioral data. That matters because YouTube's main feed has long been criticized for being difficult to audit or adjust. Until now, viewers could nudge it with likes, dislikes, and 'not interested' flags, but they could not simply state an intent and get a result built around it.

For YouTube, the move also puts Gemini in front of an enormous consumer audience. The company has been threading its AI models through products including search, ad tools, and creator features, and recommendation is the highest-stakes surface it has touched so far. Any change to how the home feed is generated ripples across viewing hours, ad inventory, and the livelihoods of creators who depend on algorithmic distribution.

The announcement lands amid broader scrutiny of recommendation algorithms and their effects on what billions of people see. Regulators in the European Union have pressed platforms to offer users more control over feeds under the Digital Services Act, and researchers have argued that opaque ranking systems make it hard to study harms. A feature that lets users spell out their preferences in words is a partial answer to that pressure, though it remains to be seen how transparent YouTube will be about how Gemini translates a request into a ranking.

The obvious open questions are accuracy and abuse. A written prompt can be vague, contradictory, or gamed, and Gemini's job is to turn something like 'calm cooking videos without talking' into an actual slate of content that holds up over multiple sessions. YouTube has not said how it will handle prompts designed to surface borderline material, or whether custom feeds will be available to all users or tested with a subset first.

What to watch next is whether custom feeds become a genuine alternative to the default algorithm or a side panel most people ignore. If users adopt them, the feature could change how creators think about reaching audiences — optimizing for describable niches rather than broad engagement bait. If they do not, it becomes another control that exists on paper more than in practice.