YouTube shifts to automated AI labelling for photorealistic content
The platform’s policy update follows the release of Google’s Gemini Omni models, aiming to increase transparency for significant AI-generated material while leaving monetisation and recommendations unaffected.
YouTube has announced a significant overhaul of its content transparency policies, moving away from a model that relied solely on creator disclosures to one utilising internal detection systems for artificial intelligence-generated material. Effective from May 2026, the platform will automatically label videos containing significant photorealistic AI content, marking a shift towards more active policing of generative media on the site.
The new system employs internal signals to identify when AI has been used to create or alter content to a photorealistic degree. Under the updated policy, these labels will be displayed prominently below the video player for long-form content and overlaid directly on YouTube Shorts. This change is designed to ensure that viewers can easily identify AI-altered or AI-generated material, addressing concerns as generative video models become increasingly sophisticated.
The policy adjustment coincides with the recent release of Google’s Gemini Omni multimodal AI models at the Google I/O developer conference. These models are capable of producing high-quality video content that reflects an understanding of physics, culture, history, and science. YouTube stated that while creators are still encouraged to disclose AI usage, the platform will apply labels automatically if such disclosures are omitted or if internal detection identifies significant photorealistic AI.
Creators retain the ability to update disclosure status if their content is misidentified by the new system. However, labels cannot be removed if the content was created using YouTube’s own AI tools, such as Veo or Dream Screen. Additionally, labels will be permanently attached to videos that contain C2PA metadata indicating full AI generation. This move aligns with broader industry trends, as OpenAI recently committed to the C2PA standard, joining other major technology firms including Nvidia, Kakao, and Eleven Labs.
YouTube’s approach distinguishes between photorealistic content and clearly animated or unrealistic scenarios, such as fictional creatures. For the latter, labels will remain in the expanded description rather than appearing prominently on the video itself. The company also confirmed that the introduction of automatic AI labels will not impact video recommendations or monetisation eligibility, ensuring that creators are not penalised financially for using AI tools within the platform's guidelines.
This update follows the expansion of YouTube’s AI deepfake detection capabilities, which now allow any adult user to scan the platform for face matches. Initial tests of this feature were conducted with public figures, politicians, and celebrities. Alongside its content policing efforts, YouTube continues to invest in generative AI features for users, including Ask YouTube, a playlist generator for YouTube Music, and AI video summaries.
The transition to automated labelling represents a maturation of YouTube’s AI policy, which has required creator disclosures for over two years. By integrating detection tools directly into the user experience, the platform aims to provide greater clarity for viewers navigating an increasingly complex digital media landscape. The changes take effect in May 2026, giving the platform time to refine its internal signals and ensure consistent application across its vast library of content.
