YouTube AI label: three auto-detection signals and Sume outputs

YouTube's help page lists three signals that can apply an AI label automatically: content made with YouTube's GenAI tools, content that contains C2PA metadata, and content its internal systems detect as AI generated or altered. A Sume timeline or trim render uses only ffmpeg, so it adds none of its own generation, but the sources you feed it carry their own history. Check each source before you upload.

What YouTube says

YouTube's blog post of May 27, 2026 says the label for photorealistic or meaningfully altered content moves below the player for long videos and onto the video as an overlay for Shorts. It also says new internal signals began rolling out in May 2026.

What a Sume render adds

Sume's timeline and trim pages say the jobs compile ffmpeg and do no provider inference (Sume docs, read 2026-10-08). They cut, join and re-encode pixels that already exist. The docs for these jobs do not say whether C2PA metadata in a source is kept or removed, so do not assume either. Treat the output as carrying whatever the inputs carried until you test it.

• Timeline render: assembly only; no generative step in the job.
• Video trim: exact re-encodes with libx264; keyframe is a stream copy.
• Video inspect probe fields are duration, size, codec, color and audio facts; C2PA is not among the documented fields.

A practical sorting step

Sort your inputs into recorded footage, footage from an AI generator, and footage that was recorded then altered. The first group needs no AI disclosure from the pipeline alone. The second and third groups fit YouTube's description of realistic content that is generated or meaningfully altered with AI, and for those you should answer Yes in the AI use setting at upload.

If a label is applied automatically and you believe it is wrong, YouTube says the Studio notification stays until you update the survey, but it also says some labels cannot be adjusted. Do not plan around reversing one.

What to keep as a record

For every upload, keep a note of where each source clip came from: recorded, generated, or recorded then altered. This is a record you hold yourself. YouTube's page says creators who consistently choose not to disclose may face a manual label or penalties, including removal of content or suspension from the YouTube Partner Program, so an accurate note protects you when a label is questioned.

Remember that labels are about the content a viewer sees, not about the tool that cut it. A cut, a join and a re-encode do not change that, but a generated clip inside the join does.