YouTube is taking another step toward making recommendations feel less like a one-size-fits-all algorithm and more like a conversation. The platform is rolling out an AI feature that lets you generate a custom video feed by describing what you want to watch—then pinning that feed to the top of your YouTube homepage so it’s always within reach. It’s a small interface change on the surface, but it points to a bigger shift in how video discovery may work going forward: instead of only reacting to what you’ve already watched, YouTube is beginning to let you actively “author” your viewing experience.
At the center of the update is a simple idea. Rather than relying solely on your watch history, likes, subscriptions, and prior searches, YouTube now offers a prompt-based way to request a feed. You go to the “Your custom feed” tab at the top of the YouTube homepage, enter a description into an AI text box, and the system builds a feed around that request. You can also choose from suggested options, which helps if you’re not sure how to phrase what you’re looking for. The result is a personalized stream of videos tailored to your interests, your mood, or a specific topic you want to explore—something you can then keep pinned for quick access.
This matters because “recommendations” have always been a bit of a black box. Even when they’re good, they’re still constrained by what the system can infer from your behavior. If you suddenly want something different—say, you’re in the mood for calm productivity content instead of tech news, or you want a weekend deep dive into a niche subject—you often end up searching manually or scrolling through irrelevant suggestions until the algorithm catches up. A prompt-driven feed is designed to reduce that friction. It gives you a way to tell YouTube what you want right now, not just what you wanted last week.
The feature is currently rolling out with English language support to YouTube users in the US who are signed in, whether they’re using the mobile app or desktop. That rollout detail is important: it suggests YouTube is testing the experience carefully, likely because prompt-based personalization introduces new variables. When you ask for a feed in natural language, the system has to interpret intent, translate it into a set of content signals, and then assemble results that feel coherent rather than random. Doing that reliably across different user bases, devices, and content categories is non-trivial. So the limited initial availability reads as a controlled launch rather than a global flip of a switch.
What makes this feature particularly interesting is the way it reframes the homepage. Traditionally, the YouTube homepage is a curated mix of recommended videos, channels you follow, and content the system thinks you’ll click. With custom feeds pinned to the top, the homepage becomes more like a dashboard you can configure. Instead of only passively consuming what YouTube surfaces, you can create a “view” that matches your current goal. That’s a subtle but meaningful shift in power: you’re no longer just the input; you’re also the editor.
Think about how people actually use YouTube. Many viewers don’t come with a single question like “what should I watch?” They come with a context. They might want background audio while working, something to watch with friends, a short burst of entertainment after dinner, or a structured learning path for a topic they’re trying to understand. Those contexts change throughout the day. A prompt-based feed aligns with that reality because it’s easy to update. You can imagine creating multiple pinned feeds—one for “learning mode,” one for “relax mode,” one for “fitness and mobility,” one for “new indie game trailers,” and so on. Even if YouTube doesn’t explicitly market it as multi-feed management, the pinning capability implies that the feature is meant to be used repeatedly, not just once.
There’s also a psychological angle. When recommendations are generated from your own prompt, the feed feels more intentional. That can make the experience more satisfying even when the underlying mechanics are similar. In other words, the difference isn’t only the quality of the videos—it’s the sense of control. People tend to trust systems more when they can see the logic behind the output, and prompting provides that transparency. You can look at the feed and think, “This is what I asked for,” rather than “This is what the algorithm decided.”
Of course, the promise of prompt-based personalization comes with challenges. Natural language is flexible, but it’s also ambiguous. If you type something broad like “motivational videos,” what does that mean to you? Motivational could mean motivational speeches, fitness motivation, career advice, or even cinematic edits. The AI has to interpret your intent and then map it to content that matches your likely preferences. If it gets it wrong, you’ll notice quickly—because the feed is supposed to reflect your request, not just your history.
That’s where the suggested options become more than a convenience. They can act as guardrails, nudging users toward prompts that are easier for the system to interpret. Suggested options also help reduce the “blank page” problem. Not everyone knows how to describe what they want in a way that translates well into recommendation logic. By offering examples, YouTube can improve both the success rate and the perceived usefulness of the feature.
Another challenge is balancing novelty with relevance. A custom feed built from a prompt could easily become either too narrow (only showing familiar content) or too broad (introducing unrelated videos that technically match keywords). The best version of this feature would thread the needle: deliver videos that fit the prompt while still feeling fresh enough to be worth watching. YouTube’s existing recommendation infrastructure already handles a lot of this balancing act, but prompting adds a new layer. The system must decide how much weight to give the prompt versus the user’s established viewing patterns. If it leans too heavily on the prompt, it might ignore long-term preferences. If it leans too heavily on history, it might fail to deliver the “right now” experience that prompting is meant to provide.
YouTube’s decision to let users pin these feeds suggests the company expects them to be stable enough to return to. Pinning implies continuity: you’re not just generating a temporary list; you’re creating a reusable entry point. That raises another question: will custom feeds evolve over time as you watch more videos? If they do, the feed could become even more accurate, but it might also drift away from the original intent. If they don’t, the feed might become stale. The feature’s design likely needs to address this tension, even if the user-facing details aren’t fully visible yet.
There’s also the broader ecosystem implication. YouTube is competing in a world where discovery is increasingly mediated by AI. Search engines are changing, chat-based assistants are becoming common, and social platforms are experimenting with AI-driven feeds. A prompt-based custom feed is a way for YouTube to stay relevant in that shift without forcing users to leave the platform. Instead of asking you to search for “productivity videos for night work” and then refine results, YouTube can generate a feed directly. That keeps users inside the YouTube experience and reduces the number of steps between intent and playback.
From a creator perspective, this could also change how content gets discovered. Traditional recommendations often rely on engagement signals and similarity to past viewing. Prompt-based feeds introduce a new discovery pathway: content might be surfaced because it matches a prompt’s intent, even if it’s not the closest match to a viewer’s history. That could benefit creators whose content aligns strongly with certain themes or moods, especially if their videos are well categorized and clearly described. At the same time, it could increase competition for attention within those prompt-defined spaces. If many users prompt for the same general topics, the system will need to decide which creators rise to the top—and those decisions will shape what audiences consider “the best” within a given mood or interest.
The feature also hints at how YouTube may handle personalization in the future. Prompting is a form of user intent capture. Over time, YouTube could potentially combine prompt intent with other signals—like time of day, device context, or even what you’ve been watching recently—to refine feeds further. For example, a prompt like “calm cooking videos” might behave differently in the evening than in the morning. Or “beginner guitar lessons” might prioritize structured tutorials when you’re in learning mode. While the current rollout focuses on the prompt-to-feed workflow, the underlying direction is clear: YouTube wants to treat intent as a first-class input.
It’s worth noting that the feature is currently limited to English and to signed-in users in the US. That limitation doesn’t necessarily mean the feature is less capable elsewhere; it likely reflects the complexity of rolling out AI features responsibly. Language nuance matters for prompts, and content availability varies by region. YouTube also has to ensure that the feed generation behaves consistently with its policies and community standards. A prompt-based system can produce unexpected interpretations, so careful rollout helps prevent edge cases from causing harm or confusion.
Even with those caveats, the core value proposition is compelling: fewer searches, less scrolling, and a more direct path from “what I want” to “what I’m watching.” For heavy YouTube users, that can be a quality-of-life improvement. For casual users, it could make the platform feel less overwhelming. YouTube’s catalog is enormous, and the hardest part for many viewers is not finding content—it’s finding the right content at the right moment. A custom feed feature is essentially a shortcut to relevance.
There’s also a subtle cultural shift implied by this update. For years, recommendation systems have trained users to accept that discovery is something done to them. YouTube suggests; you react. Prompt-based feeds suggest a different relationship: you instruct; the system responds. That’s closer to how people use search, but with a twist. Search returns results; a feed returns a flow. A feed is meant to be consumed continuously, which changes the experience from “find one thing” to “