News at a Glance
YouTube’s new custom feeds let users describe the videos they want to see in their own words, then use Gemini to build a personalized feed around the request.
What Happened
On September 23, 2026, YouTube officially launched "Create Your Own Feed," an AI-powered personalized feed generation tool built on Google's Gemini multimodal model. Users can input natural language commands (e.g., "I want in-depth videos about quantum computing") to build customized feeds, with the system parsing intent and dynamically filtering and ranking content. The feature supports real-time adjustments, allowing users to modify preference descriptions to update recommendations. YouTube stated the tool is available to select US users in beta, with a full rollout planned by end of 2026. The feature runs parallel to YouTube's existing recommendation algorithm, and use
Background
YouTube's recommendation algorithm has long been dominated by machine learning models based on viewing history and interaction signals. However, user demand for transparency and control over recommendations has grown, with surveys in 2025 showing over 60% of YouTube users wanting more direct influence over recommended content. Meanwhile, generative AI has accelerated in content understanding and personalization, with Google's Gemini model achieving breakthroughs in video semantic understanding, providing the technical foundation for natural language-driv
Deep Dive
Liu Gong believes the core innovation lies in shifting recommendation systems from "passive prediction" to "active construction." Traditional algorithms rely on implicit modeling of user behavior, while this tool converts explicit natural language preferences into structured recommendation signals via Gemini's semantic understanding. This demands new heights in video semantic comprehension, intent parsing, and dynamic ranking. Liu Gong notes that Gemini's multimodal video understanding is key, as it parses visual, audio, and textual information to match abstract user descriptions. However, Liu Gong predicts initial quality fluctuations, as ambiguous natural language may lead to results deviating from user expectations, requiring continuous optimization through feedback loops.Liu Gong argues the commercial value lies in boosting user engagement and ad precision. User-constructed feeds may yield higher watch time and interaction rates, increasing ad exposure opportunities. The tool could also become a differentiator for YouTube Premium, driving subscription growth. Liu Gong predicts that if adopted widely, it will alter creator strategies—creators must optimize titles and descriptions for natural language queries rather than keywords alone, potentially spawning new SEO tool markets.Liu Gong believes this will intensify AI recommendation competition among video platforms. TikTok and Netflix may be forced to follow suit or risk user churn. Liu Gong points out that YouTube's move may redefine "personalization" standards—from algorithm-led to user-led—with profound implications for the entire content distribution industry.
Data Points
- Comparison with existing algorithm: The current algorithm recommends based on implicit behavioral data (watch time, likes), while the new tool uses explicit natural language descriptions, shifting logic from "predicting behavior" to "executing commands."
- Competitor comparison: Netflix's 2025 "AI Recommendation Assistant" only supports preset options, whereas YouTube's tool allows fully free natural language input, significantly higher interaction freedom.
- Technical specs: Built on Gemini 2.5 Pro, video semantic understanding accuracy improves ~3
Impact & Outlook
For the industry, in the short term (3 months), YouTube will collect user feedback to refine the algorithm, while other platforms may assess feasibility of similar features; in the medium term (1 year), if successful, it could push the entire video recommendation sector toward user-led models, making AI recommendation tools standard. For ordinary users, in the short term (3 months), they gain more transparent recommendation control but may face quality fluctuations; in the medium term (1 year), they will experience more precise content discovery while ad
Angles to Explore
Points to consider
- Will natural language recommendation weaken algorithmic tracking of user behavior, thereby affecting ad precision?
- Is this feature an opportunity or threat for small and medium-sized content creators?
- How will YouTube balance user control with content ecosystem diversity?
Source & Original
This item comes from TechCrunch AI (published 2026-09-23 14:30:00). Wenfeng Radar provides a Chinese/English summary and commentary on overseas AI developments; the original content is owned by its author.
Daily aggregation of overseas AI developments with in-depth analysis. Bookmark us to never miss an important signal; back to homepage.