YouTube Music introduces AI labels: What production records should creators keep?

List human contributions and the stages involving AI

Follow your actual production sequence and record source materials, generated outputs, human edits and the parts used in the finished work. Using AI to organize ideas differs from putting generated vocals directly into a track. Describe the work you actually did.

Ask collaborators to add their contributions to the same record. A broad 'AI-made' label need not replace useful detail, and a tool's contribution should not be described as entirely human work. Keep the record useful for explaining and revising the track, without filling it with tool jargon.

Keep materials and versions connected to the final track

Link the final audio to its production record, retaining information about materials actually used and changes made. If the team replaces a section, voice or backing track, update the record too. Otherwise, the description may refer to an older version rather than the file uploaded.

Check the description listeners will see before publishing

Check the title, description and credits against the final audio. Do not describe a generated process as a real performance, or leave an abandoned production plan in the finished track's description. Record the reason for later revisions so the team can explain differences between versions.

Check results with a fixed input

For an original sample, record who composed the melody, whether vocals were generated, which parts were rerecorded and the source licences. Compare the final track with that record before uploading: vocal tuning alone should not automatically be described as a fully generated track, and manual editing should not conceal predominantly generated material. No music was uploaded; this is a recordkeeping example.