Samuel Smith’s AI demos: trace the workflow before labeling the final song

The reported workflow

AP’s May 30 reporting described London Americana musician Samuel Smith using AI demos to communicate arrangements to musicians. The report distinguishes that preparatory work from the studio recording of Horizon, saying the generated demo audio was not mixed into that final track. This concerns Samuel Smith, not the pop singer Sam Smith.

Sources: Associated Press: AP reporting on Smith’s demo-to-studio workflow

Follow the audio through production

Instead of asking only whether AI was used, ask what artifact it produced and whether that artifact became part of the release. The answer can be mixed: a system might influence an arrangement without its generated sound appearing in the master. Keep that distinction visible rather than forcing the whole project into one origin label.

1. Match the artist name to the track title, album and official credits. Similar names should not cause one musician’s workflow to be attributed to another.
2. Identify the demo’s function. Was it a guide for musicians, a source of retained stems, a temporary vocal, or something else? Record only the role the evidence supports.
3. Map the handoff to the recording session. Look for an account of which parts were performed again and whether generated audio was retained, replaced or left undocumented.
4. Separate the people credited for composition, arrangement, performance and production. A human performance credit does not answer every question about how the arrangement developed.
5. Write the conclusion at track level. Do not infer the same workflow for every song on an album or every release by the artist without further evidence.

Report influence and retained material separately

A precise note can say that a generated guide informed a workflow while leaving the provenance of other stages unverified. If you cannot establish whether a stem entered the final mix, mark that as unknown instead of claiming either a fully synthetic or AI-free recording. This approach is more informative than inferring origin from polish, genre or sound quality. isGenAI’s image tools cannot audit session files or classify the finished audio.

Questions about this case