AI content creation tools like Google’s new Omnimodel threaten to make it even harder to separate AI fantasy from reality, but YouTube is taking an important step toward verifying the origins of videos. After debuting hopeful AI content labeling in 2024, Google will begin using more prominent labeling for AI videos, and the site will no longer rely entirely on uploaders to publish the videos they create using AI tools.
When YouTube first tried AI video identification in 2024, it was mostly free. At the time, AI video almost always defeated itself by looking weird or disjointed. In just a few years, AI models like Seedance, Runway, and Google’s own Veo have raised the bar for realism and consistency in AI video, making spaghetti more accurate than ever.
Recognizing that, YouTube is making AI labels more prominent and automating parts of the process. Creators are still required to indicate whether their videos were created using AI tools when they upload them. But uploaders previously had no incentive to be honest about it. Starting this month, YouTube will use “new internal signals” to flag AI content. This apparently applies to videos that show “significant use of photorealistic AI.”
Simplified AI Labels & Auto-Detection: What You Need to Know
Google is vague about what signals go into its AI detection system. We are asking for more information and will update if we hear anything. This blog post mentions two ironclad triggers: C2PA metadata to indicate pure AI sources and the use of watermarked Google tools like Veo. Creators who believe their videos have been incorrectly tagged as AI can appeal, but they can’t appeal if a site marks their upload as AI for any of these reasons. These labels are “permanent”.
