YouTube announces AI creator agent, custom Gemini feeds and shopping search

At its annual Made on YouTube event, YouTube announced a wave of AI features spanning both sides of its platform: a background "agent" that optimizes creators' channels, prompt-based custom feeds that let viewers describe what they want recommended, and shopping upgrades to the Ask YouTube conversational search tool.

An AI agent that runs a channel in the background

The most aggressive creator-facing announcement is an AI agent that works continuously on a creator's behalf. According to The Verge, it monitors a channel's back catalog for older videos that have gained fresh relevance — because of news events or renewed interest — and suggests new titles or thumbnails to capitalize on the moment. It can also assemble sponsorship pitches by pulling demographic and audience data from a channel, according to Amjad Hanif, YouTube's vice president of creator products.

The agent expands on AI creator tools YouTube first showed in 2025, which included thumbnail A/B testing and a chatbot for querying channel performance. It can now generate complete thumbnails and titles from the content of an uploaded video, and creators can ask it for feedback on their own designs.

Testing whole videos, not just thumbnails

Testing is also scaling up. The Verge reports that a dynamic thumbnail tool lets creators upload up to three images per video, with YouTube's system assigning the best performer to different audience segments to lift watch time. Going further, creators can test three versions of a video — one with a different intro or hook, for instance — on small audience segments. The creator can promote the winning version to the permanent video, or YouTube will do so automatically after seven days. Hanif says the system ensures the variants are not wildly different from one another.

Viewers describe their own feeds

On the viewer side, TechCrunch reports a new "custom feeds" feature: users type a natural-language description of what they want to watch, and a feed built on Google's Gemini model appears in its own tab pinned to the top of the home page. Prompts can be long and specific, covering what to include, exclude and prioritize — a feed of video podcasts sized to a 30-minute train commute, for example, or one for a particular mood.

The feature follows moves by other platforms: Bluesky popularized user-defined feeds and added an AI builder called Attie this year, while Threads, Instagram, X and more recently Spotify have shipped comparable tools. YouTube's custom feeds will not replace the main recommendation feed, and support for creating multiple custom feeds rolls out on web and mobile starting next month, according to TechCrunch. Emily Moxley, YouTube's vice president of product management for viewer AI, framed the feature as a way to navigate a corpus of more than 20 billion videos.

Ask YouTube grows into a shopping companion

Ask YouTube, the platform's conversational search feature, is being repositioned as a shopping aid. TechCrunch reports that product searches will return organized recommendations, including comparison tables of product attributes based on a user's preferences, and viewers will be able to ask follow-up questions directly on a video's watch page. The company said the changes reflect how viewers already learn about products from creators they trust.

Usage is climbing, according to YouTube: over 140 million people used Ask YouTube to dig deeper into videos in June, an increase of more than 500% since December. Moxley told reporters that a custom Gemini model underpins the feature, pinpointing the exact second in a video where a viewer's question gets answered.

The reputational tightrope for creators

AI use among creators remains contentious. The Verge points to the backlash against Hank Green earlier this summer after he acknowledged using AI for research — a debate many observers read as being about creator overwork as much as the technology itself. Hanif told The Verge he distinguishes between AI that makes creators more efficient and AI that effectively does the job: writing scripts, shooting scenes or wholesale producing content is where, in his view, the tools start to cross a line. Asked whether YouTube has data showing the tools improve metrics for creators who use them, the company declined to share any; the benefit Hanif cited is time saved that flows back into creative work.

Why it matters

The announcements show YouTube extending AI beyond its recommendation engine into production, discovery and commerce at once. For viewers, custom feeds echo an industry-wide shift toward user-directed algorithms, while Ask YouTube's shopping features turn search into a product-research layer that ties video more tightly to purchasing. For creators, the agent and video-level testing promise time savings — but YouTube offered no evidence the tools actually move views or engagement, leaving the value proposition resting on efficiency alone. The open question is where creators and their audiences draw the line between AI that assists and AI that effectively makes the content, a boundary Hanif himself acknowledged is easy to cross.