When YouTube first showed Custom Feeds at Made on YouTube in September 2026, it was easy to file as an experiment: a text box on the home screen where a viewer could describe the videos they wanted and have Gemini assemble a feed to match. A quarter later, the shape of the thing changed. In an update rolling out through late October and into November 2026, custom feeds stopped being a tucked-away novelty and became part of the furniture — they moved into the top navigation bar where the old topic chips lived, viewers can now save and name several of them, and the whole feature got the rules of a real product: a cap on how many you can keep, an expiry on the ones you ignore, and sync across your devices. That is the difference between a feature and a surface. A viewer no longer tries custom feeds once; they live in a small set of standing, self-authored lineups they return to. This guide is about what that promotion means for the people making the videos. It is deliberately not a rerun of the discovery mechanics — the legibility, single-intent packaging, and machine-readable video that decide whether you get matched into a described feed are worked through in the companion piece on viewer-built algorithms. This page is about the layer above that: once custom feeds are a finite, permanent set of slots on a viewer's personal navigation, the creator's job shifts from gaming one shared feed to programming for a lineup — occupying a slot in the feeds your audience builds around your topic, their mood, and their routine. It covers exactly what the expansion changed and what is still US-only and early, why the feed limits are the part that actually matters strategically, the new creator-side product-recommendation shelf that shipped in the same wave, and the honest limits of planning around a behavior nobody has measured yet.
YouTube announced Custom Feeds at its Made on YouTube event on September 23, 2026: a feature that lets a viewer describe, in their own words, the videos they want to see and has Gemini build a recommendation feed to match. At launch it read like an experiment — a text box, a few example prompts, a feed that appeared alongside the normal recommendations. In an update rolling out through late October and into November 2026, the feature grew up. Custom feeds moved into the top navigation bar, taking the place where the older non-custom topic chips used to sit; a viewer creates one by tapping a "+" at the top of the home feed and describing a topic, a set of channels, or a vibe; and the saved feeds now sync across devices. In the same redesign, YouTube pulled the subscriptions feed up into that top bar too. The change is small to describe and large in consequence: custom feeds stopped being something a viewer tries once and became something that lives in the primary navigation, as a set of standing lineups they return to.
That promotion is what this guide is about. It is written for creators, and it deliberately does not re-cover the discovery mechanics — how a video gets read and matched into a described feed, why single-intent packaging and real transcripts decide whether you surface. Those are the subject of the companion guide on YouTube algorithm customization with AI, and they still apply. This page is about the layer on top: once custom feeds are a finite, permanent, saved surface rather than a novelty, the creator's job changes shape. It stops being "get promoted into the one feed everyone shares" and becomes "be the obvious occupant of a slot in the small set of feeds your audience actually keeps." That is a programming problem, and the expansion is what makes it one.
Precision matters on a rollout, so here is what moved and what did not. The headline is placement: custom feeds now sit in the top navigation bar and replace the non-custom topic options that used to live there, which means they are no longer a thing a viewer opts into from a menu — they are part of the first row of controls on the home screen. Creating one is a "+" tap followed by a prompt, and an existing feed can be refined by editing the prompt box at the top of it. The feeds sync across a viewer's devices, including desktop. And the broader home redesign moved the subscriptions feed up into the same top bar, so the whole top-of-home surface is now a strip of competing feeds — subscriptions, the default recommendations, and however many custom feeds the viewer has built. What did not change: custom feeds are additive. The ordinary recommendation feed you have always optimized for is still there, and watching inside a custom feed still feeds your regular Home recommendations. The two surfaces are linked, not walled off.
The strategically important part of the expansion is the one that sounds like fine print. Custom feeds now have a cap and an expiry. A free account can keep up to three active custom feeds; a YouTube Premium subscriber can save up to eight. Any feed a viewer stops using expires automatically after roughly 30 days of inactivity, and the feeds are private — a viewer cannot share one with a friend. Read those three rules together and they describe a scarce, self-pruning surface. A viewer is not building an endless shelf of feeds; they are curating a small, standing lineup — three slots for most people, eight for the committed — and the expiry means the slots that survive are the ones a viewer actually comes back to. That scarcity is the whole game for a creator. If discovery for your audience increasingly happens inside a set of three-to-eight feeds a person deliberately keeps, then being in that set is worth far more than a transient appearance in an infinite scroll, and the competition is not every video on the platform — it is whatever else is clean enough to fill the same slot.
Keep the scope exact, because it bounds what you can act on. The feature is rolling out in English, to users 13 and older, in the United States, and a viewer needs both watch history and search history enabled for it to work. Tablets are not supported yet, and YouTube has not said when that changes. Reporting on the exact geography has been inconsistent — some write-ups described a wider rollout — but the US-English-only read is the better-supported one, and YouTube's own framing of the underlying feature has been experimental. So this is an announced-and-rolling-out surface in one market, not a global, settled distribution channel. Do not rebuild a strategy around it. Treat it as a strong directional signal — YouTube is investing real navigation real estate in viewer-stated intent — and make the moves that are cheap and correct regardless of how fast custom feeds spread.
Here is the structural shift the expansion forces. The classic YouTube growth model treated the home feed as one enormous shared surface: in principle everyone in your potential audience was reachable through the same recommendation system, so the job was to make something with broad enough pull to be promoted into it. A finite set of saved, viewer-authored feeds fractures that. Your audience's discovery is increasingly spread across a long tail of narrow feeds — "beginner sourdough," "calm woodworking, no talking," "pre-2000 F1 race breakdowns" — and because each viewer only keeps a handful, the question is no longer "can I out-perform the corpus for a slot in the big feed" but "am I the obvious thing to put in the slot this viewer built." That is channel programming, not virality. A programmer does not try to win every household; they own a reliable slot in the lineup of the households they serve. The equivalent move here is to be legibly, specifically, consistently the channel a viewer's described feed is asking for — which, for a focused creator, is a more winnable competition than out-shouting everything on the platform, and a lesson the platform has been signaling for a while in social media discoverability beyond followers.
The scarcity also reframes what a "loss" looks like. In the shared-feed era, a video that underperformed simply did not get promoted this time; there was always the next upload. In a saved-feed era, the cost of being unprogrammable is structural: a channel that is vague, scattered, or hard to read as being about one thing matches no viewer's specific request cleanly, so it never enters the candidate pool for the feeds its potential audience keeps — not because a single video lost, but because the channel is illegible to the slot. Broad, do-everything content was always a weaker growth play; a finite set of audience-authored feeds raises the price of it, because there is no wide shared feed left to carry it on general appeal.
The example prompts YouTube keeps using are the clearest brief a creator has for this surface, and they reveal an axis most channels never package for. "Witty podcasts for my daily 30-minute commute." "Evening wind down with long documentaries." "Relaxing commentary to unwind with." Those are not topics — they are contexts. A viewer is describing a slot in their day, a duration, and a feeling, not just a subject. Which means there is a real, underused programming opportunity in signaling the context a video fits: its length relative to a commute or a wind-down, its tone, whether it is background-friendly or demands attention, whether it is a quick hit or a long sit. Most creators describe what a video is about and stop there; far fewer declare when and how it is meant to be watched. A finite, routine-driven set of feeds is the first discovery surface that rewards saying so, because the feeds themselves are increasingly organized around moments in a day, not just niches. Programming a channel for this means thinking in terms a television programmer would recognize — a morning-commute block of short, punchy pieces; a long-sit evening block — and packaging each piece so its context is as legible as its topic.
The custom-feeds expansion arrived in a wave that also carried creator-facing changes worth knowing, because they point the same direction — more of the creator's operation moving inside YouTube's own surfaces. YouTube merged Ask Studio, its AI channel-guidance chat, with its general Help so creators get advice and support in one place, and it began testing a personalized product-recommendation shelf inside the YouTube Studio shopping hub. Rene Ritchie, YouTube's Head of Editorial, described it plainly: "YouTube Studio is rolling out a brand new personalized product recommendation shelf right inside your shopping hub." For eligible affiliate creators, the shelf suggests brand partnerships they may be well-suited to based on their past content, audience engagement, and niche trends, and shows an estimated commission rate for each suggestion. It is a test, with limited availability, not a fully live program — treat the specifics as provisional. But the throughline with custom feeds is clear: YouTube is building AI layers that match supply to intent on both sides of the screen — viewers to the feeds they describe, creators to the partnerships their catalog fits. The common requirement, for the creator, is a body of work legible enough for an AI to read what it is about and place it correctly. That is as true of the shelf matching you to a brand as it is of the feed matching you to a viewer.
The through-line of this guide is that a finite set of audience-authored feeds turns discovery into channel programming: the win is occupying a durable slot in the small lineup your audience keeps — for their topic, their mood, and their routine — not winning one shared feed. Programming a lineup is a supply problem before it is a strategy problem, and it is the part Kompozy addresses. Kompozy is an AI content generation and multi-platform publishing engine; it does not build viewers' feeds, place you in anyone's custom tab, or control YouTube's navigation — those are YouTube's surfaces, and no tool programs them for you. What it changes is the cost of producing a consistent, context-aware lineup worth programming into them.
The programming angle is where an engine earns its place. A viewer's feeds are narrow and routine-driven, so the move is not one broad channel but several reliably-distinct streams, each unambiguously about one topic, mood, or moment — the morning-commute short, the long-sit explainer, the niche-specific review — produced at a cadence that keeps each slot fresh. Kompozy's AI Influencer persona pool lets one workspace run several consistent branded identities rather than one, so you can program a distinct, recognizable stream for each slice of your audience instead of blurring them into a single channel the feeds can't cleanly place. Persona Shorts and Persona Frames generate the short-form and brand-templated video that fill the quick-hit and signature blocks, each with the auto-captions and transcript that make its topic and context machine-readable — which is how a feed knows the piece belongs in the slot a viewer described. Because a single Persona Brief governs voice, topic, and the fact and banned-word rules for every output, each stream stays reliably about its one thing across a long run instead of drifting into the unprogrammable sprawl a finite feed can't hold.
Sustaining a programmed lineup, rather than shipping one good video, is the part that does not scale by hand for a small team. Autopilot keeps each stream flowing on a schedule so the slots stay active — which matters directly when an unused feed expires after 30 days and the slots that survive are the ones kept warm by a steady supply. Every output clears quality gates that hold the Persona Brief in context and reject invented statistics and off-brand language before anything publishes, so the topical and tonal consistency that makes you a clean match stays intact across volume. And because Kompozy generates across output buckets and fans to the social platforms plus blog and email, the topic you program on YouTube is declared, in readable form, on the other surfaces a viewer and an answer engine also read — so the same catalog that fills a custom-feed slot is also the body of work the Studio product shelf, and an AI search answer, can read and place. Kompozy does not decide what your channel should be known for or make a weak video worth finishing; it lowers the cost of producing the consistent, context-legible, multi-stream lineup that a surface built on feeds viewers write themselves actually rewards.
Custom feeds are personalized YouTube recommendation feeds a viewer builds by describing, in plain language, the kind of videos they want — "witty podcasts for my 30-minute commute," "new DIY creators" — and having Gemini assemble a feed to match. First shown at Made on YouTube in September 2026, they appear as saved, named feeds the viewer returns to, sitting alongside the standard Home recommendations rather than replacing them.
The late-2026 update moved custom feeds into the top navigation bar, where they replace the older non-custom topic options, and a "+" at the top of the home feed creates one. Viewers can now save several feeds (not just try one), those feeds sync across devices, and the subscriptions feed moved up into the same top bar. The practical effect is that custom feeds went from a one-off novelty to a persistent place a viewer lives in.
Free accounts can keep up to three active custom feeds; YouTube Premium subscribers can save up to eight. Feeds are private and can't be shared, and any feed you stop using expires automatically after about 30 days of inactivity. Watching inside a custom feed still feeds your regular Home recommendations, so the two surfaces are linked, not walled off.
It turns discovery into channel programming. A viewer only holds a handful of standing feeds, so the game stops being "win a slot in one giant shared feed" and becomes "be the obvious occupant of a slot on your audience's personal lineup" — the feed they built around your niche, their mood, or their routine. That rewards being specifically and consistently about one thing, and it rewards packaging for context — length, tone, when a video is meant to be watched — not just topic.
In the same late-2026 wave, YouTube began testing a personalized product-recommendation shelf inside the YouTube Studio shopping hub for eligible affiliate creators. It suggests brand partnerships a creator may be well-suited to based on their past content, audience engagement, and niche trends, and shows an estimated commission rate for each. It is a test with limited availability, not a fully live program.
YouTube's custom feeds — AI-built recommendation feeds a viewer describes in plain language — graduated from a September 2026 experiment to a permanent surface in late 2026: they moved into the top navigation bar, and viewers can now save up to three (free) or eight (Premium) feeds that expire after about 30 days of inactivity. For creators this turns discovery into channel programming: occupying a finite slot on the audience-authored lineup for your niche, context, and routine, rather than gaming one shared feed.
Get started → · ← All guides · Compare Kompozy vs other tools