YouTube: Improving User Satisfaction

It’s all about using our technology and research to help enrich people’s lives. Like platforms dedicated to giving everyone a voice and showing them the world. Our collaboration with YouTube’s product and engineering teams has helped optimize decision-making processes, increase safety and engagement, and enhance the experience for a wide range of users.

Making Shorts more searchable
YouTube Shorts — short-form videos less than a minute long — are viewed more than 50 billion times a day. Covering everything from emerging K-pop stars to local food guides, they’re quick to watch, quick to make — and gaining popularity all the time. However, because Shorts are created in just a few minutes, they often lack the detailed descriptions and titles that facilitate easy searchability. To address this, we introduced Flamingo, our visual language model to help generate descriptions. Flamingo analyzes the initial video frames and provides text to explain what’s depicted on screen (e.g., “a dog balancing a stack of crackers on its head”). This text is saved as metadata in YouTube, helping create clearer content categories and matching user searches to better results. YouTube is rolling out this technology across Shorts, with auto-generated video descriptions for all new uploads. Now, viewers can more readily find and enjoy a wider selection of relevant videos from diverse global creators.

Optimizing video compression
As video consumption has surged in recent years, the growing internet traffic has made video compression an increasingly critical issue. We collaborated with YouTube to test the potential of our AI model, MuZero, to enhance the VP9 codec, which helps compress and transmit video over the internet. We applied MuZero to some of YouTube’s live traffic and, at launch, recorded an average 4% bitrate reduction across a diverse set of videos. Bitrate influences the computing power and bandwidth needed for video playback and storage — affecting load times, resolution, buffering, and data usage. By refining the VP9 codec on YouTube, we’ve contributed to reducing internet traffic, data consumption, and loading times. This optimization allows millions of people worldwide to enjoy more videos while utilizing less data.

Protecting brand safety
Since 2018, our partnership with YouTube has helped educate creators about the types of videos that can earn ad revenue and ensure that appropriate ads appear in the correct context. We devised a label quality model (LQM) in conjunction with the YouTube team to label videos more accurately, in alignment with advertiser-friendly guidelines. This improvement enhances the accuracy of ads running on videos and ensures that advertisements appear alongside content adhering to YouTube’s standards. By better identifying and classifying videos, we’ve increased trust in the platform for viewers, creators, and advertisers alike.

Improving AutoChapters
As the methods of creating and consuming video evolve, creators are increasingly adding chapters to their videos. This feature assists their audience in finding desired content, though it can be a labor-intensive process. We collaborated with the YouTube Search team to develop an AI system that suggests chapter segments and titles for creators by automatically processing video transcripts, audio, and visual features. With AutoChapters, viewers can spend less time searching for content, and creators save time on chapter creation. Since this feature was introduced at Google I/O in 2022, auto-generated chapters have been applied to tens of millions of videos (and counting) across the platform.

Evolving technologies and products
We are continuously seeking ways to enhance Alphabet products through our AI research. Our collaboration with YouTube has already made a significant impact on people’s lives — and with more initiatives underway, we’re dedicated to further improving the user experience.