YouTube is handing users a prompt box and letting them build their own recommendation algorithm. The company announced a feature called custom feeds on Wednesday at its Made On YouTube event, letting viewers type a natural-language description of what they want to watch and get a dedicated feed pinned to the top of their home page. The feature is powered by Google's Gemini model and rolls out on web and mobile starting next month.
The pitch is straightforward. Instead of accepting whatever the master recommendation system serves, a viewer can ask YouTube for a feed of video podcasts for a 30-minute train commute, or request "relaxing commentary videos to unwind with," and Gemini will assemble it. Users can specify what to prioritize, what to exclude, and how the feed should be shaped, with prompts as detailed as they want.
The custom feeds sit alongside the standard recommendation feed rather than replacing it. When a user opens the YouTube app, the main home feed remains, and each custom feed appears in its own tab at the top, one tap away. YouTube is positioning this as a discovery aid rather than a full algorithm override, letting viewers switch contexts without abandoning the default experience.
Key facts
- 01YouTube announced custom feeds on September 23, 2026, letting users describe recommendation feeds in natural language.
- 02The feature is powered by Google's Gemini model and pins each AI-built feed to the top of the YouTube home page in its own tab.
- 03YouTube's corpus now contains over 20 billion videos, per VP of Viewer AI Emily Moxley.
- 04Support for multiple custom feeds rolls out on web and mobile starting next month.
- 05The move follows similar prompt-based feed tools from Bluesky's Attie, Meta's Threads, Instagram, X, and Spotify.
The scale argument is central to why YouTube thinks this matters now.
That scale is exactly the problem custom feeds are meant to address. A 20-billion-video library is functionally infinite for any single user, and the default recommendation system optimizes for aggregate engagement rather than the specific mood or use case a viewer has in mind on a given evening. A prompt-based feed shifts that curation from an implicit signal war between watch history and clickbait to an explicit user instruction Gemini can act on.
YouTube is not the first platform to try this. Bluesky popularized user-built algorithms years ago as a differentiator against X, and rolled out an AI tool called Attie earlier this year to make feed construction easier. Meta's Threads and Instagram, X, and Spotify have all shipped their own feed-building features in recent months, most of them AI-powered. Spotify's Taste Profile launched only weeks ago and works on a similar principle for music and podcasts.
What YouTube brings to the category is the corpus. Threads and Bluesky are curating short-form text posts; Spotify is curating a music catalog measured in the tens of millions. Applying a natural-language filter across 20 billion videos, spanning long-form, Shorts, live streams, and podcasts, is a materially different retrieval problem, and it is one that plays to Gemini's multimodal strengths in a way pure text platforms cannot match.
The open question is how sharply the feeds actually reflect user intent versus reverting to safe, high-engagement recommendations dressed in a custom-feed wrapper. Prompt-based curation only works if the model honors constraints like "exclude reaction videos" or "prioritize channels under 100,000 subscribers," and platforms have historically resisted giving users tools that shrink watch time in the short run. YouTube has not disclosed how tightly the Gemini layer overrides the ranking system underneath it.
For YouTube, custom feeds are also a defensive move against the drift of habitual video watching toward TikTok on one end and podcast apps on the other. Giving users an explicit "commute feed" or "unwind feed" is a way to reclaim specific moments in the day that competitors have carved out, and it does so by leaning on Google's model advantage rather than by tweaking the recommendation black box further. If the feature works as advertised, it is the most consequential thing Gemini has shipped inside a consumer Google product this year, and it puts pressure on every other video platform to answer with something similar.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
