YouTube launches AI‑powered custom feeds to let users design their own recommendation algorithm

YouTube unveiled a new “custom feeds” feature that lets users craft algorithmic recommendations by describing desired videos in plain language, leveraging Google’s Gemini AI. The feed appears as a pinned tab on the home page and will roll out to web and mobile users starting next month.

Key points

• Users type a natural‑language prompt to define the feed’s focus, such as “relaxing commentary videos to unwind with.”
• Gemini builds the feed, allowing detailed inclusion and exclusion criteria.
• The feature coexists with the main recommendation feed; it does not replace it.
• Rollout begins next month across YouTube’s web and mobile platforms.

What happened

In a pre‑brief ahead of the Made on YouTube event, Emily Moxley, VP of Product Management for Viewer AI, announced the custom feeds capability. She said, “One of the amazing things about being on YouTube is that the world is always at your fingertips,” emphasizing the platform’s massive video corpus of over 20 billion items.

The feature works like a prompt box where users enter a description of the videos they want to see. Gemini interprets the request, curates a personalized stream, and pins the resulting feed to the top of the home page in its own tab. Because the process is AI‑driven, users can provide lengthy, nuanced instructions about content to prioritize or exclude.

YouTube confirmed that the new feeds will not replace the existing home‑page recommendations. Instead, viewers can switch between the default feed and any number of custom feeds they create. Support for multiple feeds will be available on both web and mobile interfaces beginning next month.

The backstory

The move follows a broader industry trend toward user‑controlled recommendation engines. Decentralized network Bluesky first popularized prompt‑based feed customization, later adding its Attie AI tool. Meta’s Threads, Instagram, X, and Spotify have all rolled out similar AI‑powered feed‑building features in recent months. These efforts aim to give users more agency over the content they consume while opening new monetization pathways for platforms.

YouTube’s own recommendation system has long relied on opaque algorithms that balance watch time, engagement, and advertiser interests. By exposing a prompt interface, the company shifts part of the curation burden to the viewer, echoing the same shift seen across other social and streaming services.

Why YouTube is betting on AI‑driven curation

YouTube’s 20 billion‑video library presents a discovery challenge for both casual viewers and power users. Allowing users to articulate their preferences narrows the signal‑to‑noise ratio, potentially increasing session length and ad impressions. The Gemini model provides the scalability needed to process natural‑language prompts at YouTube’s scale, turning vague user intent into concrete recommendation lists.

The feature also serves as a defensive move against competitors that already offer granular feed control. By integrating the capability directly into the core UI, YouTube reduces friction for adoption and positions itself as the most flexible video platform for personalized discovery.

Potential impact on creators and advertisers

Creators may see traffic redirected to niche custom feeds that match their content’s specific themes, improving relevance but fragmenting audience metrics. Advertisers could target ads within custom feeds, gaining access to highly defined viewer intents, but they will need to adjust bidding strategies to account for the new segmentation.

The dual‑feed model preserves the existing recommendation engine, so creators with broad appeal will still benefit from the default algorithm. However, those whose content aligns with common custom‑feed prompts may experience a boost in discoverability, while others could face reduced visibility if users opt for narrowly defined feeds.

What builders and businesses should do

• Experiment with prompt language to see how Gemini surfaces your content in custom feeds.
• Align video metadata and tags with likely user prompts to improve placement.
• Monitor performance dashboards for traffic coming from custom feeds versus the default feed.
• Adjust ad targeting to include custom‑feed audiences where relevant.
• Prepare communication for stakeholders about how these feeds may affect viewership metrics.

The bottom line

YouTube’s custom feeds launch next month, and early adoption metrics will reveal whether AI‑driven personalization reshapes video discovery and advertising on the platform.