In May 2026, YouTube said it would enforce its labelling system for AI-generated content. So, we tested YouTube’s system using two AI-generated videos

This is what we found:

• YouTube asks creators to self-disclose the use of AI during the upload process.
• When AI metadata is present, YouTube automatically detects AI-generated videos. 
• YouTube can automatically label AI-generated videos without creator disclosure if the videos contain relevant AI metadata (if they have AI metadata). 
• Sometimes, YouTube does not apply an AI-generated label if the uploader falsely denies using AI, especially when the file lacks AI metadata.

I uploaded three AI-generated videos to YouTube. One was generated using Google’s Gemini AI Model, and the other was uploaded twice, downloaded directly from the Melty ASMR channel using yt-dlp, a command-line YouTube video downloader.

This is how YouTube’s system responded when I uploaded the AI-generated videos: 

• YouTube automatically detected the video generated with Google’s Gemini, even when I did not self-disclose the use of AI. (disclosure not mandatory)  —  [ Video ] 
• But YouTube did not “detect” the AI-generated video downloaded from a third-party channel. — [ Video ] 
• However, when I self-disclose that a video is AI-generated, YouTube labels it. it.  — [ Video ]

Quick note: Surprisingly, YouTube did not flag Melty ASMR’s video for a copyright violation.

What happens if you falsely disclose? Those users may face “penalties from YouTube, including removal of content or suspension from the YouTube Partner Program.”

YouTube is using C2PA metadata to detect AI-generated content, especially when it is produced with “YouTube’s GenAI tools.” Explore C2PA’s FAQ for more information.

Important read: According to the European Union’s Code of Practice for AI-generated content, the AI-labels should appear at the beginning of the video as well as after a regular interval, like ads. Read more about it here: [ PDF | EU Website| MediaNama lowdown ]

Once a video is labelled, YouTube shows a link under the description. YouTube lets users visit the page titled ‘Understanding ‘How this content was made’ disclosures on YouTube’.

Non-realistic AI content doesn’t need to be disclosed: Content creators are generally not required to disclose AI-generated non-realistic content, minor aesthetic edits that don’t mislead viewers about actual events, or routine production assistance, such as beauty filters, color adjustments, special effects, captioning, video/audio repair, idea generation, voice cloning for dubbing, gameplay footage, or AI-extended backdrops. This non-exhaustive list excludes alterations that could materially deceive audiences about reality.

Realistic AI-generated or altered content should be disclosed:  Content creators must disclose AI-generated or meaningfully altered content that appears realistic and could mislead viewers, including AI-generated music, fabricated extra footage of real places, realistic simulations of events that didn’t occur, or false depictions of real people giving advice, stealing, confessing to crimes, or being arrested.

Also Read: 

• MeitY Scientist Calls for Limits on Deepfake Virality
• YouTube Introduces Automatic Detection for AI-Generated Videos 
• GitHub Tool Bypasses Google SynthID Watermark.