YouTube Lets You Prompt AI to Build a Custom Video Feed
YouTube is rolling out an AI feature that generates a personalized video feed from user-entered text prompts or YouTube-suggested options. The feature moves discovery from passive algorithm to active intent.
Original sourceYouTube is introducing an AI-driven feed customization tool that lets users type in a prompt — or select from YouTube-suggested options — to generate a tailored video feed on the fly. Rather than waiting for the recommendation algorithm to infer preferences over weeks of watch history, users can now express what they want to see in natural language and get a curated feed in return.
The feature represents a meaningful shift in how YouTube handles discovery. The existing homepage algorithm operates as a black box, surfacing content based on implicit signals like watch time, likes, and search history. This new tool introduces an explicit layer on top of that — users can request things like 'relaxing cooking videos' or 'deep dives into mechanical keyboards' and receive a feed built around that intent.
YouTube has not fully disclosed the underlying model or how prompts are resolved against its video index, but the feature appears to blend semantic understanding with its existing recommendation infrastructure. Suggested prompts from YouTube itself are also offered, which hints at a guided discovery mode for users who don't know exactly what they want but want more control than the default feed gives them.
The rollout follows a broader industry push toward conversational and intent-driven interfaces in content platforms. For YouTube, where the algorithm has long been both a superpower and a source of user frustration, giving users a direct input mechanism is a notable product acknowledgment that passive recommendation alone isn't always enough.
Panel Takes
The Skeptic
Reality Check
“This is a thin wrapper over YouTube's existing recommendation engine with a text input box on top — the prompt probably routes to the same signals the algorithm already uses, just with an extra NLP parsing step. The scenario where this breaks is obvious: a user types 'no drama, just calm tech reviews' and gets Logan Paul's latest upload because the model is optimizing engagement, not intent. YouTube's incentives and the user's intent are structurally misaligned, and a prompt box doesn't fix that. What kills this in 12 months: Google ships a genuinely better version inside Search or Gemini that actually understands nuance, and this feature gets quietly deprecated as a beta experiment.”
The PM
Product Strategy
“The job-to-be-done here is clear and real: 'I know roughly what I want to watch but the algorithm keeps showing me the wrong things' — that's a problem millions of users have every day. The interesting product tension is whether YouTube lets the prompt fully override engagement-optimized ranking or just nudges it, because if it's the latter, users will feel tricked after the first few sessions and stop using it. The 'suggested prompts from YouTube' feature is a smart onboarding move — it lowers the blank-canvas anxiety and teaches users the prompt vocabulary without requiring them to figure it out themselves.”
The Futurist
Big Picture
“The falsifiable bet here is: users will shift from passive consumption to active curation as AI lowers the cost of expressing intent, and platforms that expose that interface first will capture the habit. The second-order effect nobody is talking about is what this does to creator strategy — if viewers can prompt away from 'viral' content toward niche intent, the optimal video title stops being SEO-bait and starts being a precise semantic signal about what the video actually delivers. YouTube is riding the trend of conversational interfaces replacing browse UX, and they're on-time to it, not early — but being on-time when you own the distribution is often enough.”
The Creator
Content & Design
“As a viewer, this is genuinely useful — being able to say 'show me 20-minute documentary-style food videos, not vlogs' and get a feed that respects that is a real quality-of-life improvement over the current algorithm roulette. The taste layer here is delegated almost entirely to the user, which is the right call for a discovery tool, but the suggested prompts YouTube provides are where the platform's own editorial sensibility will show — if those suggestions are generic ('comedy,' 'music,' 'gaming'), this feature has no taste at all. The editing surface is the concern: if you get a bad feed from a prompt, is there a way to refine it iteratively, or do you just re-prompt from scratch?”