What does “made with AI” actually mean?

Whenever I release a track, I spend a decent amount of time filling in boxes.

Title. Composer. Producer. Performer. Master owner. Release date. ISRC.

Now another question is entering that process:

Was AI used?

In July 2026, several music organisations proposed two voluntary labels for sound recordings: AI-Generated and AI-Assisted.

The distinction is useful. It is also broad.

I use AI as one of many tools in my increasingly larger creative toolbox. Most often, it helps me test an idea or think through possible approaches to a mix. Those conversations may shape a decision, but they do not necessarily become part of the finished audio.

Still Here is a fairly clear example.

I wrote and produced the track, including the final vocal lines. I am not a singer, so I used generative AI to realise those lines as a vocal performance. The vocals form a smaller part of the overall work.

The proposed system lists an AI-generated lead vocal as an example of an AI-Generated recording. Applied to Still Here, the vocal method could determine the label for the whole track, while saying nothing about the scale or placement of that contribution.

Then there is a less obvious example.

I used Claude Code to help build Chaos, a custom synth, and Chaos Effect, an audio effect for Ableton. An upcoming track contains sounds I made with Chaos. AI helped create the code. I designed the sounds, played the notes and produced the track.

Does that make the finished recording AI-assisted too? The proposed labels do not give a clear answer.

AI can generate an element heard directly in a recording. It can also help build an instrument, suggest a mixing approach or support a decision that the artist later carries out. These uses enter the creative process at different points.

That is the problem with broad AI labels. They tell us that AI was involved, but not where it entered the process, how much it contributed, or what the artist actually made.

Of course, if the track was made entirely with AI, a clear “AI label” makes perfect sense. But very often, producers only use AI as one tool in a larger workflow.

I’m not trying to downplay or exaggerate the use of AI. What I want is clear, accurate language that actually informs the listener.

Music already has a useful model. We credit songwriters, performers, producers, mixing engineers and mastering engineers.

AI disclosure can work in the same way. A visible label can provide a quick overview, while detailed credits explain whether AI contributed vocals, lyrics, composition, instruments, sound design, artwork or video.

Another important point is how independent artists often are the composer, producer, performer, label and metadata department at the same time.

We will frequently decide which box to tick. Vague categories will produce inconsistent answers, even from artists trying to disclose their process honestly.

The EU adds a separate legal layer. Article 50 of the AI Act applies from 2 August 2026 and requires machine-readable marking for certain AI-generated or manipulated outputs. Its guidelines exclude source code from that obligation.

Music services are also using different systems. Spotify displays contribution-level credits. Apple uses metadata tags for tracks, compositions, artwork and music videos. Deezer detects and labels fully AI-generated music and excludes it from recommendations.

The same recording may therefore be described differently across platforms, leaving artists to navigate several systems and listeners to work out what each label means.

Without clear credits, the listener is left with a label, but no real understanding of how the music was made.