YouTube will let you build your own algorithm with AI
Sep 23, 2026, 7:30 AM · TechCrunch

Custom feeds let viewers prompt Gemini for a pinned YouTube tab — a user-shaped algorithm that still sits beside, not instead of, the main For You machine.
Why it matters
Also from Made on YouTube on September 23, 2026: custom feeds. Viewers describe what they want in a prompt — commute-length podcasts, “relaxing commentary videos to unwind with,” detailed include/exclude rules — and Gemini builds a feed pinned as its own home-page tab. Multiple custom feeds arrive on web and mobile starting next month; the main recommendation feed stays.
This is the industry’s answer to “the algorithm did this to me.” Bluesky, Threads, Instagram, X, and Spotify have all shipped feed-building tools. YouTube joining with Gemini and a 20-billion-video corpus claim is the mass-market version of that idea.
From the desk
We’re cautiously for this. Giving people language to steer discovery is useful AI when the default ranking feels opaque or addictive. Emily Moxley, YouTube’s VP of Product Management for Viewer AI, framed the pitch around exploration across a huge corpus. A pinned tab you asked for is a healthier mental model than an endless undirected stream — if people actually use it.
The hedge in YouTube’s own framing matters: custom feeds do not replace the main feed. That’s product honesty and a soft admission that engagement still lives in the primary recommender. Prompted feeds can be a pressure valve while the core machine keeps optimizing watch time. I’m watching whether custom feeds get comparable inventory quality or become a sandbox of leftovers.
Gemini in the loop means long, natural-language constraints — prioritize, exclude, mood, format. That’s powerful and fragile. Vague prompts will produce mush; adversarial or extremist prompts will test safety filters; “build me a rabbit hole” prompts will show whether user agency includes self-harmful curation. Platforms that hand users a steering wheel still choose the roads.
Compared with Bluesky’s chronological and community-driven feeds, YouTube’s version is AI-mediated and catalog-scale. The upside is real control for parents, students, professionals, and anyone tired of one blended identity. The downside is fragmentation: creators may chase both the main algorithm and a swarm of niche prompted feeds they can’t see.
Useful AI earns the win if custom feeds reduce doomscroll regret without becoming a second black box. Name the trajectory: every major consumer surface is racing to let users “describe an algorithm.” The winners will be the ones that keep the prompt honest about what it can’t override.
Context
TechCrunch by Sarah Perez, 7:30 AM PDT September 23, 2026. Feature announced ahead of / at Made on YouTube; Gemini powers feed construction; multi-feed support begins rolling out the following month on web and mobile.
Who feels it
- Viewers
- Prompted tabs can match intent and mood better than a single home feed — only if inventory quality matches the main recommender.
- Creators
- Discovery may split across opaque main ranking and invisible custom-feed niches; analytics transparency will matter.
- YouTube / Google
- Gemini-driven feeds deepen AI in the viewer loop while preserving the engagement core of the primary feed.
- Peer platforms
- Reinforces the 2026 pattern of natural-language feed builders across social and media apps.
What to watch
- October+ rollout quality: latency, prompt fidelity, and whether excludes actually stick.
- Whether custom feeds receive comparable freshness and diversity to the main home feed.
- Creator analytics for traffic from custom feeds vs. default recommendations.
- Safety handling of harmful or extreme prompted feed descriptions.
Companies: Google