YouTube is expanding its custom feeds experiment. Users type a short prompt to create a personalized video feed, pin it to the top of Home, and switch between feeds there. The first test launched in November, expanded in May, and now supports up to eight active feeds per user — with unused prompts expiring after a few weeks, and an editable text box atop each feed to adjust its focus. The updated experiment is rolling out in the US to signed-in viewers on mobile (Android and iOS) and desktop.
The most consequential detail isn't the feature itself — it's that custom feeds will replace the standard topic chips at the top of the Home feed in the mobile app. YouTube isn't adding an option; it's swapping part of its own algorithmic content display for direct user input. That's a striking move given that algorithmic matching has driven massive engagement gains across social apps. YouTube's bet is that combining explicit signals (prompts) with implicit ones (watch behavior) makes a more compelling feed — importing the conversational-prompt habit people learned from AI tools into feed design.
YouTube reach has flowed through search, Home recommendations, and subscriptions. If custom feeds stick, a fourth doorway opens: feeds users defined themselves. Getting picked up there requires your channel and videos to be legible in interest-language — titles, descriptions, chapters, and captions that align with the words users actually type into a prompt. "Cooking vlog for people living alone" beats a vague "vlog"; the topic definitions people would realistically build a feed around become the target.
Eight feeds are eight interest rooms. If users split their viewing into "home café," "investing basics," and "running," a sharply focused niche channel that fits one room precisely earns repeat exposure over a general channel that fits none. It's the same direction as YouTube's per-creator 24/7 channels experiment — see YouTube Expands Stations — Every Creator Gets a 24/7 Channel. YouTube is reorganizing viewing from an algorithm-fed infinite feed into rooms users enter with intent.
Platform algorithms have always been black boxes that marketers reverse-engineer from the outside — the exercise we documented in We Read X's Algorithm Source Code — 17 Filters Matter More Than the Weights. Custom feeds make part of that game transparent: what users will type is researchable, not guessable. Ask what prompts your customers would build feeds around — answer it with keyword research and customer interviews — and align your channel's topic definition and metadata to that language. It's a US-only experiment for now, but if it lands, one piece of YouTube SEO grammar changes.
Feeds users create by typing a short prompt, pinned to the top of Home and switchable there. Up to eight can be active per user, prompts are editable to refocus recommendations, and the experiment is rolling out to signed-in US viewers on mobile and desktop.
Custom feeds will replace the standard topic chips at the top of the mobile Home feed — YouTube swapping part of its algorithmic display for explicit user input, combined with implicit watch signals.
Make your channel legible in the interest-language users would type into a prompt. Align titles, descriptions, and chapters with the topic definitions customers actually use, and favor sharp niche positioning over general coverage — feeds reward channels that fit one interest room precisely.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
TikTok’s Road to Black Friday series shows how top TikTok Shop sellers prep for BFCM: Princess Polly’s 6,000 creator videos a month, Dossier’s sampling engine and SharkNinja’s LIVE strategy.
Meta shared how small businesses use its Muse AI agents: analyzing ad performance, building campaigns through conversation and automating workflows with Shopify, Canva and HighLevel. Prompts included.
Reddit outlined its U.S. midterm election plans, including the Up the Vote hub, AI detection that cut spam exposure 20% and removes nearly 2 million fake upvotes daily, and AMA rules for political ads.