For most of YouTube's life, you've managed "the algorithm" indirectly. You watch a few videos, click a few thumbnails,...
.webp "YouTube Custom Feeds: How AI Lets You Build Your Own Algorithm")
Now YouTube is trying something different. You can just tell it what you want.
YouTube is rolling out Custom Feeds, a feature where you describe the videos you want in your own words. You type a prompt, Google's Gemini builds a feed from it, and that feed shows up as a pinned tab on your home page. Your regular home feed doesn't go anywhere. This is an extra lane you can switch into when you feel like it.
That sounds like a small quality-of-life update, and in one sense it is. But it also changes how you deal with a recommendation system. It shows where product design is going in 2026: away from personalization the system guesses at, and toward personalization the user asks for.
What YouTube Is Shipping
Put simply, YouTube stops guessing from your watch history. You describe the feed you want, and it builds that feed.
YouTube's own examples are a feed of video podcasts for a 30-minute commute and a feed of relaxing commentary videos for winding down. You can get specific, too: what to prioritize, what to skip, what mood you're after, and what you never want to see.
Gemini turns that description into a feed, and the feed gets its own tab pinned at the top of the home page. YouTube says support for multiple custom feeds will roll out on web and mobile starting next month.
Why This Matters More Than the Usual AI Feature
Most AI features are bolted on. They're nice, but the product underneath works the same way it always did.
This one is different because it changes how you talk to the algorithm in the first place.
Recommendation systems have always been good at maximizing engagement and bad at understanding context. In a single week, you might open YouTube to learn a specific tool, to put something on in the background while you decompress, and to go deep on a niche interest through long interviews. The main feed mashes all of that into one stream. That's usually when people start feeling like the platform is pushing stuff at them.
With Custom Feeds, YouTube is conceding that one feed can't represent a whole person. You end up using the app in modes, like a feed for your commute and a separate one for learning Figma. For a product team, that's a real improvement, because users finally get something to steer with.
"Build Your Own Algorithm" Raises the Bar on Trust
Once a platform tells you that you can build your own algorithm, you expect it to follow your instructions.
If you set up a feed for "AI product updates with no hype," and it still shows outrage thumbnails, you're not going to shrug and blame the algorithm. You're going to decide the feature is broken, because you asked for something specific and didn't get it.
The feature succeeds or fails on two questions. Can users quickly tighten, loosen, or fix a feed when it drifts? And does the feed stay on track over weeks and months, or does it slowly slide back toward whatever keeps people watching longest?
Prompts can be long and detailed, covering what to include, what to exclude, and what to prioritize. That's a lot of flexibility. It also means YouTube has made a promise it now has to keep, every day.
Filters Are Turning Into Sentences
If you build products, this part applies to you even if you never open YouTube.
Custom Feeds is one example of a broader shift. With an LLM in the loop, you can swap a pile of settings for a single text box.
Most SaaS and consumer apps give users control through filters, toggles, sort options, category pickers, and saved views. Those work, but they're rigid, and they get unwieldy once your catalog or list of options grows large. The alternative is letting the user describe what they want in plain language, and the product translates it into a view they can save and come back to.
That's basically what YouTube built for discovery: saved views of the recommendation engine that act like separate feeds.
What Happens Next
Once people get used to making a feed for a specific purpose, they'll want the same thing in other apps. For example:
- Shopping: "show me durable, mid-range options and hide sponsored junk"
- News: "give me updates, not opinion pieces"
- Learning: "show me beginner-friendly content with practical examples"
- Hiring: "only roles with salary ranges, remote-first, and no take-home tests"
Oddly enough, this might be the most realistic fix for people's complaints about algorithms. Nobody has to build one perfect feed. Users can make a few feeds that match whatever they're doing at the moment.
YouTube is keeping its algorithm. It's handing users some of the controls, and that's a bigger deal than it looks.
If you're building a product where users are fighting filters and toggles instead of getting what they actually want, ATX Soft can help you design a natural-language personalization layer that holds up over time, not just in the demo.
Frequently Asked Questions
What are YouTube Custom Feeds?
You describe the videos you want in plain language. YouTube uses Gemini to build a feed from that description and pins it as its own tab on your home page.
Does this replace the main YouTube algorithm?
No. The standard home feed stays. Custom Feeds are extra feeds you can switch to for specific interests or moods.
How do you create a Custom Feed on YouTube?
Type what you want the feed to cover into the prompt box, and YouTube uses Gemini to generate it.
Can you create multiple Custom Feeds?
Yes. YouTube says support for multiple custom feeds will roll out on web and mobile starting next month.
Why is YouTube doing this now?
YouTube's reason is scale. It says it has more than 20 billion videos, and describing what you want is easier than searching when you don't know exactly what to type.
.webp)
