YouTube is done making you train its algorithm by watching things. Starting next month, users will be able to type a plain description of what they want to see, such as “video podcasts for a 30 minute commute” or “relaxing commentary videos to unwind with,” and get a standing feed built around that request, pinned to its own tab on the home page.
The company calls the feature custom feeds, and it announced the tool at its Made On YouTube event on Wednesday. Emily Moxley, YouTube’s VP of product management for Viewer AI, told reporters in a pre-brief that the pitch is scale: “There’s over 20 billion videos in the YouTube corpus, so you’re always one search away from a treasure trove of topics.” That number, YouTube’s own, is the justification for a search box that behaves less like search and more like a standing subscription.
Under the hood, custom feeds runs on Google’s Gemini model to interpret the request and assemble a matching list of videos. Users can write long, specific instructions, covering what to include, what to exclude, and what to prioritize, and the system is meant to hold onto that context rather than treat each session as a fresh query. The feed sits alongside YouTube’s main recommendation feed rather than replacing it, so the personalized tab is additive: users still get the algorithmic homepage YouTube has always shown them, plus however many custom ones they build.
That structure matters more than the interface. A prompt box that builds a durable feed is not a new recommendation engine, it is a new instruction layer sitting on top of the one YouTube already runs, and the two will compete for the same attention. Whether YouTube’s default feed keeps its position when a semantic alternative sits one tab away is the open question the company has not addressed.
YouTube is late to this specific move. Bluesky built its reputation partly on user-created custom feeds years before this, and added an AI assistant called Attie this year to make building them easier without writing feed logic by hand. Threads added public custom feeds in 2025. Instagram, X, and, this same week, Spotify have each shipped their own feed-building tools, several powered by AI in the same style YouTube is now using. Prompt-built feeds are no longer a differentiator; they are close to table stakes for any platform that wants to claim it gives users control over discovery.
The pattern across all of these products is that a feed described in language is easier to build than a feed tuned by hand, but harder to audit. None of the companies shipping this feature, YouTube included, has published data on how often a described feed actually matches what the user meant, or how it handles a request that conflicts with what performs well for engagement. YouTube’s announcement, like the others, describes the feature in terms of what a user can ask for, not what the system does when a request and the platform’s own incentives point different directions.
For creators, a feed layer that responds to specific written requests changes what gets discovered outside YouTube’s existing recommendation logic, particularly for niche or long-form content that struggles to surface in a general-purpose feed. Any channel with a narrow, well-defined audience should treat the next few months as a test of whether custom feeds route around the algorithm’s existing biases or just replicate them with extra steps.
Sarah Perez reported this for TechCrunch on September 23, 2026.
