The Scoop
Last summer, YouTube ran an experiment: It had AI select segments of videos to answer users’ searches, like for the “best beaches in Hawaii,” and stitched them into slideshows.
In traditional YouTube searches, viewers would see a simple list of videos based on a wide variety of factors, sometimes including a customer’s prior preferences. The only way to get the information was to click on one of the videos.
With the AI slideshows, viewers were getting responses based on what content or information was in the videos, unconnected to a viewer’s habits or interests in specific creators. The move was controversial because it pushed the creators into the background and put the knowledge they imparted in their videos front and center. This version of YouTube search launched to US Premium subscribers in June 2025, expanded in August — and then quietly vanished.
“People just didn’t like it,” said Emily Moxley, YouTube’s vice president of product management. “Viewers want to hear from creators... the story and the information that creators can supply is something that we can’t replace,” she told Semafor.
YouTube, like most media companies, faces a critical challenge: Can it strike the perfect balance in the AI era and avoid overinnovating? If it adopts the latest technology wholesale, it risks disrupting its own bread and butter, but staying the course risks disruption from a new wave of content enabled or created by AI.
YouTube went back to the drawing board and recently launched a new AI-powered feature, called Ask YouTube, which enhances the search experience but doesn’t attempt to disintermediate viewers’ relationships with creators or their content. This time, the company says, people are treating it differently. “We’re seeing a lot more of these complex queries where people are clearly sharing the full extent of their underlying intent,” said Sandie Gong, the product lead for Ask YouTube. “We’re seeing a lot more learning queries, more how-to queries.”
Users can still scroll their feed and viral shorts, but they can also query the site for detailed information like how to fix a specific model of garage door opener, or to see the view from a specific seat inside the Las Vegas sphere. Users still watch creator content (rather than an AI-generated description), but they are taken to the specific part of the video they’re looking for.
“What we’ve always seen in search is that efficiency is at a premium,” Moxley said. “If we aren’t efficient enough, people will do that search elsewhere... they’ll just find it in text, they’ll find it on the web.”
AI could make YouTube’s creator content more valuable in other ways. The videos can be used to train models that could be crucial to the future of robotics and autonomous vehicles. And it could be used to generate new videos, made from scratch. But all of that threatens the pipeline of new content from humans — which, YouTube keeps finding, is what its viewers want.
When ChatGPT disrupted the entire tech ecosystem nearly four years ago, YouTube had hit its stride. It is now the single biggest distributor of TV time in the US. It took in roughly $62 billion in revenue last year — more than any other media company in the world, by MoffettNathanson’s estimate — and is coming off one of its biggest quarters, with ad revenue up 13% to $11.1 billion.
And it is sitting on a gold mine of content: 20 billion videos uploaded, almost all of it made by human creators, whom YouTube has paid more than $100 billion over the past four years.
In this article:
Know More
The idea for Ask YouTube came out of a different tool called YouTube.com/learning, curated content that acts as a kind of virtual classroom. Some people using the tool were ignoring its intended purpose. “We actually noticed that it was really good for things that weren’t learning-specific, like ‘plan me a 7-day trip to San Francisco,’” Gong said. “Or... queries that clearly weren’t learning, like ‘how to survive a zombie apocalypse.’”
A quick prototype became a hit with executives. One user, who was going to a visit to the Sphere in Las Vegas, typed: “I’m afraid of heights. What is it like in section 400 of the Sphere?” Up came a video shot from that exact section. “It can describe it to you in text, but you can’t see it and feel what it’s like to be there,” Gong said of chatbot competitors.
Ask YouTube launched at Google I/O in May for YouTube Premium subscribers and for US desktop users this month, with mobile and TV coming soon.
Under the hood, the feature breaks a complex question into simpler ones, fans them out across YouTube’s existing search systems, and organizes the results, jumping straight to the moment in a video that answers the question. The AI writes only the connective tissue. Merging clips into a single AI-generated answer video is now off the table. Ads are untouched: Moxley declined to comment on whether the feature might be integrated into Google search or Gemini in the future.
Users are already probing its limitations. Moxley’s favorite: someone hunting for glasses that could help them isolate the green pickleball while darkening everything else around them. “I was like, what, is this even a thing? Turns out it is a thing.”
Others ask it to resurface videos from their own watch history, a feature that currently doesn’t work but might be added to the product road map. “There’s no technical reason it can’t,” Moxley said.
For creators, the ground is shifting either way. Ask YouTube views count toward monetization, and a cottage industry of “Ask YouTube SEO” guides has already sprung up. AI-generated videos get no special treatment in results. “We want to give people the videos that they want to watch,” Gong said, “however it’s created.”
Reed’s view
The worst fear of content creators is that AI slop will replace what they do. But YouTube’s own experiments with AI should allay those fears, at least to some extent and for the foreseeable future.
Even using AI to simply edit and stitch creator content together in order to provide more cogent answers to search queries was rejected by YouTube’s human viewers.
That’s more evidence that, as the world becomes more automated, people increasingly crave authentic, human experiences. Just try to get tickets to see The Odyssey on 70mm IMAX in theaters. You can’t. A big reason is that it was created using enormous cameras with actual, physical film and almost no computer graphics. And the movie is being projected onto the screen through that actual, physical film.
The YouTube creator Hank Green figured this out early on: “My gut says that people want to connect with people,” he said. “A relationship with a creator is already artificial in some way. But I don’t think AI will be that good at building audiences.”
Plenty of people have made the case for human artists, perhaps no better than Nick Cave: “Algorithms don’t feel. Data doesn’t suffer. ChatGPT has no inner being, it has been nowhere, it has endured nothing,” he said.
YouTube is helping make the case for the human audience, and that’s why this is so interesting. Humans are often very bad judges of quality. (You might argue that is how YouTube became so huge). In fact, plenty of studies show they prefer AI “slop” to legitimately better, human-created content. But humans are very good judges of authenticity.