MARKETS
Blog

YouTube will let you build your own algorithm with AI

A
Auto Bot
Editorial Team
September 24, 2026 • 8 min read
YouTube will let you build your own algorithm with AI

YouTube’s AI-Powered Custom Feeds: The Algorithm Just Got a Prompt Box

YouTube just rolled out Gemini-powered custom feeds. Type a prompt—like a topic or interest—and Gemini spins up a dedicated tab pinned to your homepage. Next month, you’ll be able to create multiple of these. That’s not a tweak. That’s a bet that users will start treating their YouTube homepage like a dashboard of AI-curated playlists, each one a niche interest or mood.

The framing is straightforward: YouTube's vast user base is drowning in content, and the main algorithm, for all its sophistication, can't cater to every whim. Custom feeds are YouTube’s answer. But here’s the question: Is this a real escape from the algorithmic rabbit hole, or just a more personalised version of it? Because a prompt box doesn’t put you in control. It just gives the black box a prettier interface.

---

How This Works—and Why It Matters

You describe what you want in natural language. Gemini—Google’s flagship AI model—translates that into a feed. The tab sits at the top of your homepage, a persistent AI-curated channel. YouTube’s pitch is that this solves a core problem: the main algorithm is optimised for engagement, not discovery. If you’ve ever fallen into a wormhole of conspiracy theories or unboxing videos, you know the feeling. Custom feeds, in theory, let you bypass that.

The rollout is aggressive. Support for multiple feeds arrives next month on web and mobile. That’s not a niche experiment. It’s a bet that users will want to segment their viewing—work feeds, hobby feeds, entertainment feeds. Bold assumption, given how rarely users engage with manual curation tools elsewhere.

But the real shift is philosophical. YouTube is admitting its main algorithm, for all its power, can’t read minds. Whether that’s true is another matter.

---

The Graveyard of User-Controlled Feeds

YouTube isn’t the first platform to hand users the reins. The history of user-controlled feed features is a graveyard of limited successes and outright failures.

TikTok's user-controlled feed feature allowed users to weight their For You Page with topics. The lesson? People say they want control. They do what’s easiest. Instagram's manual curation tools saw limited adoption. Twitter/X's list feature is a powerful way to bypass the main feed's chaos. The problem? Effort. Most users won’t expend it.

The pattern is clear: platforms love giving users “control.” Users often prefer convenience. YouTube’s AI layer might change this—if it reduces friction. If creating a custom feed feels like writing a search query and then tweaking it endlessly, most users will bail.

---

Gemini Under the Hood: How the AI Interprets Your Prompts

Here’s where it gets murky. The brief doesn’t specify how Gemini translates prompts into feeds. Does it:

  • Weigh keywords over intent? (Does “woodworking” prioritise tutorials or ASMR-style videos?)
  • Consider engagement metrics like likes and watch time, or stick to semantic relevance?
  • Learn from your behaviour within the feed (e.g., skipping videos), or stay static until you update the prompt?

Without transparency, we’re left guessing. And that’s a problem. Gemini’s interpretation of your prompt could reinforce—or amplify—biases.

Take a polarizing topic. A scientific topic could lean academic. Neither is “wrong,” but both could create a feed narrower than the main algorithm’s. The brief doesn’t say if Gemini will suggest refinements (e.g., “Try adding ‘physics’ to your robotics feed”). Without that, users might find themselves stuck in a loop of tweaking prompts.

There’s also the cold-start problem. Prompt something obscure—like a niche interest—and Gemini might struggle if it lacks training data. The main algorithm, by contrast, relies on metadata and engagement signals, which can surface hidden gems.

Then there’s prompt fatigue. If users have to constantly refine queries to get useful results, they’ll abandon the feature. YouTube’s support for multiple feeds suggests it expects users to segment interests. That only works if the AI delivers consistently.

---

The Filter Bubble Paradox: Does Customisation Expand What You See?

Some observers call this an "escape" from YouTube's algorithm. That’s optimistic. Custom feeds don’t eliminate filter bubbles. They let you design them. A user who prompts a political or ideological topic might get a feed devoid of counterarguments—just like the main algorithm. The difference? Now, they asked for it.

The brief doesn’t clarify if Gemini will suggest prompts or just execute them. If it’s the latter, custom feeds could kill serendipity. YouTube’s main algorithm occasionally surfaces unexpected gems—a documentary you didn’t know existed, a tutorial on a skill you hadn’t considered. A rigidly prompt-based feed might never show you those.

There’s also the question of discovery vs. preference. The main algorithm is optimised to keep you watching, which means it often surfaces content just outside your usual interests. Custom feeds are optimised for precision. If you only create feeds for existing interests, you’re not discovering anything new. You’re just curating a smaller bubble.

---

The Business Angle: Why YouTube Is Betting on This

YouTube didn’t build this out of altruism. There are clear business incentives:

  1. Advertising: Custom feeds could enable contextual ads tailored to feed themes. A woodworking feed might show tool ads. A tech feed could surface gadget promotions. Contextual ads often perform better than behavioural targeting, especially as privacy regulations tighten.
  2. The brief doesn't mention monetisation.
  3. Competition: TikTok and Instagram are pushing similar features. YouTube’s move may be defensive. If users start expecting AI-powered curation, YouTube can’t afford to fall behind.

If users get precisely what they request and leave—YouTube may quietly deprioritise the feature. The main algorithm is designed to maximise engagement. Custom feeds are designed for satisfaction. Those goals don’t always align.

---

Will Anyone Actually Use This?

The success of custom feeds hinges on one thing: friction. Power users—researchers, hobbyists, creators—may love the precision. Casual viewers, however, may find it easier to stick with the main algorithm’s passive recommendations.

YouTube’s decision to support multiple feeds suggests it expects users to segment interests. But will they? The brief doesn’t mention onboarding. Will YouTube guide users, or leave them to figure it out? If it’s the latter, adoption may be limited.

Early indicators will be telling. If YouTube highlights trending topics or lets users share feeds, adoption could spike. If the feature is buried in settings, it may flop. The fact that custom feeds are pinned to the top of the homepage is a good sign. YouTube wants them visible.

But let’s be real: most users won’t engage deeply. They’ll create one or two feeds for their biggest interests, then forget about them. That’s not a failure. It’s human nature. The question is whether enough users will find value to make it worth YouTube’s investment.

---

The Bigger Picture: What This Says About AI Curation

But the AI still mediates the response. That’s the key distinction. Custom feeds aren’t user-controlled. They’re user-influenced. You provide the prompt. Gemini decides what to show.

This creates a transparency problem. YouTube’s main algorithm is a black box. Gemini’s prompt-based feeds are a grey box. You know the input. You don’t know the logic behind the output. That’s better than nothing. It’s not true control.

The irony? The more “control” users get, the more they may realise they don’t know what they want—until the AI shows them. That’s not a flaw. It’s the nature of AI curation. But it raises a question: If users keep tweaking prompts to get the “right” results, are they really in control? Or just reverse-engineering the AI’s biases?

---

How to Game the System: Tips for Getting the Most Out of Custom Feeds

Assuming you do want to use this, here’s how to make it work:

  1. **Be specific. Be specific. Detailed prompts will yield better results than vague ones.
  2. **Use negatives. Use negatives. Adding exclusion terms can filter out unwanted content.
  3. Iterate. Start broad, then refine based on what Gemini surfaces. If your feed is too technical, tweak the prompt to include beginner-friendly terms.
  4. Combine feeds. Use one for deep dives (e.g., “advanced Python tutorials”) and another for casual viewing (e.g., “funny tech fails”).
  5. Watch for patterns. If Gemini keeps surfacing the same creators, your prompt might be too narrow.

The catch? The brief doesn’t confirm if you can edit prompts after creation. If you can’t, you may need to rebuild feeds often—a major friction point.

---

The Verdict: A Step Forward—or Just Another Layer of Abstraction?

For niche interests, this could be a game-changer. Need a feed of niche interests or specialized topics? Gemini might deliver where the main algorithm fails. It’s a tool for precision, not serendipity.

But let’s not kid ourselves. This isn’t the end of the algorithm. It’s just a more granular version. The AI is still in charge. You’re just giving it better instructions. That’s progress. It’s not revolution.

The real test will be transparency. If YouTube opens up Gemini’s decision-making—explaining why it surfaced certain videos, or letting users tweak the underlying model—this could evolve into something truly user-driven. If not, it’s just another layer of abstraction.

So, is this the end of the algorithm? No. But it might be the beginning of something more interesting: a world where users and AI collaborate to curate content. The question is whether users will meet YouTube halfway—or default back to passive scrolling, just like they always do.

Sources

Key Takeaways

  • The evolution of Blog continues to accelerate rapidly.
  • Experts like Auto Bot suggest staying informed on these specific developments.
  • Consider the broader implications for the tech industry moving forward.