YouTube Will Let You Prompt Your Own Feed. Creators Should Read the Fine Print.
YouTube's Gemini-powered Custom Feeds let viewers describe the videos they want, moving discovery out of the algorithm's hands and past the creator's.
At Made on YouTube 2026 on September 23, the platform announced a broad rollout of Custom Feeds, after a limited test that began in May: a feature that lets viewers type what they want into a prompt box and have Google’s Gemini assemble a homepage tab around it. Want 30-minute video podcasts for your commute, or relaxing commentary to wind down to? Describe it, and the feed appears as a pinned tab alongside the main home page. Viewers can build several. Crucially, these feeds supplement the standard recommendation feed rather than replacing it — the algorithm you already have is not going anywhere; you are just being handed a second set of controls.
Emily Moxley, YouTube’s VP of Product Management for Viewer AI, pitched the scale that makes it plausible. “There’s over 20 billion videos in the YouTube corpus, so you’re always one search away from a treasure trove of topics,” she said. Custom Feeds rolls out beginning next month — sources describe a US launch across web, mobile, and TV in the fall — making it one of the few announcements of the day with a stated rollout window.
It did not arrive alone. YouTube also previewed an expanded “Ask YouTube,” which answers conversational questions, including about products; Gemini-based chat editing for Shorts inside the Create app; live auto-dubbing for streams; expanded likeness detection for AI-generated voice and face; and serialization tools that let Shorts run as seasons and episodes. But Custom Feeds is the headline, because it changes the fundamental question of who decides what a viewer sees.
From “the algorithm chose” to “I asked for it”
For most of its history, YouTube discovery has been a black box that creators optimize against by proxy — thumbnails, retention curves, packaging, posting cadence. The implicit deal was that if you fed the recommendation engine the right signals, it would put you in front of people who had never searched for you. That serendipity — the algorithm surfacing a creator to a viewer who did not know to want them — has been the single most important growth engine on the platform.

Custom Feeds does not kill that engine. But it introduces a competing one that routes around it. When a viewer builds a feed by stating an intent — “calm woodworking, no talking, over 20 minutes” — Gemini is now matching that stated preference against the corpus. And stated preferences and revealed preferences are famously not the same thing. People say they want educational content and watch drama. The recommendation algorithm has always trusted what you do over what you claim. A promptable feed trusts what you claim.
What this does to discovery
Here is the tension worth flagging, and it is analysis, not something YouTube said from the stage: neither the platform nor its coverage has specified how Custom Feeds affect creator visibility or ranking. That silence is the story. If viewers increasingly curate by explicit prompt, discovery narrows toward what people already know to ask for — named genres, established formats, creators specific enough to describe. The channel that grows by ambush, by landing in a recommendation feed for someone who would never have typed its name, has fewer at-bats.
There is an upside case, too, and it is real: a viewer who prompts “independent climate documentaries” is expressing durable, high-intent demand that a mid-sized creator in that lane could never have surfaced against the pull of mega-channels in the generic feed. Prompting could be a leveler for niche work — if YouTube surfaces the long tail honestly rather than defaulting to the safe, big names Gemini already knows.
Which is exactly why creators should not celebrate or panic yet. The feature that lets viewers describe what they want is only as good for discovery as the model’s willingness to answer with something they had not heard of. YouTube has shipped the prompt box. It has not shown its work on who gets returned. Until it does, “build your own algorithm” is a promise to viewers whose bill, if there is one, gets paid by the creators who used to depend on not being asked for by name.
Sources
Researched and drafted with AI tools, fact-checked against primary sources and approved by a human editor before publishing. Prices and dates are as of publication. Our standards →
Sources- 9to5google.com — after a limited test that began in May
- techcrunch.com — Emily Moxley, YouTube's VP of Product Management for Viewer AI
- medianama.com — sources describe a US launch across web, mobile, and TV in the…
- socialmediatoday.com — YouTube also previewed an expanded "Ask YouTube," which answers conversational questions, including…
Latest from prompt/power
- Broadcom’s AI chip sales now outstrip its entire business from a year ago. What to watch on Dec. 9Nov 19
- NeurIPS 2026 is in three cities. Here’s how the split worksNov 19
- Pay yearly or monthly? What you save on AI subscriptions in CanadaNov 19
- IBM, NVIDIA and Fermilab plan a quantum test bed. Here’s what each brings, and what’s still missingNov 19
- GPT-6.2 Astra is OpenAI’s first released model rated ‘High’ for AI self-improvement. What that means, and what it costsNov 18