YouTube Custom Feeds Use Gemini AI to Personalize Recommendations
YouTube's Gemini-powered custom feeds let users prompt personalized video tabs on web and mobile, with support for multiple feeds rolling out next month.
Summary
At Made On YouTube on Wednesday, September 23, 2026, YouTube introduced Custom Feeds, a Gemini-powered feature that turns detailed natural language prompts into personalized recommendation feeds. Users can define subjects, purposes, priorities and exclusions, including video podcasts for a 30-minute train commute or relaxing commentary for unwinding. Each feed appears in a dedicated tab pinned atop the YouTube home page, while the existing main recommendation feed remains unchanged.
Multiple Custom Feeds will begin rolling out next month on web and mobile. Emily Moxley, YouTube's vice president of Product Management for Viewer AI, positioned the feature as a way to explore a catalog exceeding 20 billion videos. YouTube follows Bluesky, which pioneered customizable feeds and introduced its Attie AI tool in 2026, as well as Meta's Threads, Instagram, X and Spotify, which have launched similar feed-building controls, many powered by AI.
Positives
- Gemini accepts lengthy instructions covering desired subjects, priorities, exclusions and viewing purposes.
- Multiple Custom Feeds will be available across YouTube's web and mobile services starting next month.
- More than 20 billion YouTube videos can be explored through narrowly tailored recommendation feeds.
- Dedicated tabs pinned atop the home page provide direct access to feeds designed for specific interests, moods or routines.
Risks & concerns
- Custom Feeds do not replace YouTube's main recommendation feed, limiting users' control over the platform's core algorithm.
- Support is not immediately available, with the web and mobile rollout beginning next month.
- Bluesky, Threads, Instagram, X and Spotify already offer comparable feed-building controls, leaving YouTube behind several rivals.
- Gemini interprets users' prompts and assembles the recommendations, rather than providing direct manual control over every feed rule.