Do You Have to Disclose AI in Your Music? A Producer's Guide to Stems, Mastering and Vocals

You made the track. You programmed the drums, played the bassline and built every breakdown by hand. But the vocal chop came out of a stem splitter, the master came from an AI service at 2am, and the cover art started life in an image generator. Now your distributor's upload form asks whether you used AI, and it's not obvious what an honest answer looks like.

This is one of the most common AI questions producers ask in 2026, and the answer is more reassuring than the headlines suggest. The platforms aren't trying to punish producers who use AI tools. They want to know when AI made something the listener actually hears, reads or sees. Once you understand that line, the form takes two minutes.

What changed in AI disclosure rules in 2026?

A lot, and quickly. Spotify announced support for the industry's DDEX standard for AI disclosures in September 2025, covering AI-generated vocals, instrumentation and post-production, alongside a stricter impersonation policy and a spam filter. It began rolling out AI Credits in beta in April 2026, shown in Song Credits, and in August announced AI Persona badges for artist profiles that represent AI-generated identities, rolling out from mid-September. Spotify says tens of thousands of AI credits are now submitted every day.

Apple Music introduced Transparency Tags in March 2026, with four categories: Artwork, Track, Composition and Music Video. They are optional for now, they apply when AI created a material portion of the work, and Apple has said consumer-facing labels will follow later this year.

Deezer went furthest on detection. It tags fully AI-generated tracks with its own tool and removes them from algorithmic recommendations and editorial playlists. By June 2026 it reported that more than half of its daily uploads were fully AI-generated, around 90,000 tracks a day. DistroKid rolled out its AI Credits disclosure broadly in May 2026, and Bandcamp banned music generated wholly or in substantial part by AI back in January.

Which AI tools actually count as AI use?

The practical test is simple: did AI generate part of the finished work, or did it help you process something you made? DistroKid's guidance says AI-assisted mixing, mastering, pitch correction and workflow assistance don't need AI Credits on their own. Here's how that plays out tool by tool.

• AI mastering and mix assistants: no disclosure needed. They shape audio you created.
• Pitch correction and vocal tuning: no disclosure needed.
• Stem separation of your own recordings: no disclosure needed. It's processing, like a very clever EQ.
• Stem separation of someone else's record: not a disclosure question at all. It's a clearance question, and you need the rights owners' permission.
• Loops or one-shots from a generative model that stay in the final audio: disclose them as AI instrumentals.
• AI vocals or voice models: disclose, and if the voice imitates a real singer, you need their authorisation before release.
• AI lyrics or a generated topline melody: disclose them under lyrics or composition.
• Generated cover art: disclose it wherever your distributor asks about artwork.

The grey area is AI as a sketchpad. If a model suggested a chord progression and you rewrote, replayed and rearranged it until it's genuinely yours, you're in AI-assisted territory. If the melody listeners hear is what the generator gave you, largely unchanged, declare it.

MIDI generators inside your DAW are the version of this that catches people out. A plugin that suggests ten bassline patterns is a tool. The pattern you keep note for note, untouched, is generated composition. The same logic applies to AI drum pattern generators: edit, replay and rebuild the part and it's yours; export it as delivered and it's the model's.

Ask one question per element: did AI make this, or did AI help me make this?

How do you declare AI use through your distributor?

You don't disclose on Spotify or Apple Music directly. You set the fields in your distributor's upload flow, and the distributor delivers them to the stores as metadata. The exact fields vary, but most split disclosure between the recording, the composition and lyrics, and the artwork, and some ask how much AI was involved.

Declare per track, not per release. On a four-track EP, one track might carry a generated vocal texture while the other three are fully human, and your metadata should say exactly that. Keep AI out of your artist credits too: an AI tool is never a primary or featured artist, and your producer and songwriter fields should list people.

Here's a typical example. A hard techno track with your own drums, synths and arrangement, a spoken vocal phrase created with an AI voice tool, a master from an AI mastering service and a cover made in an image generator. You declare the vocal as AI-generated, leave the composition as human if the words and music are yours, declare the artwork, and say nothing about the mastering.

Keep a short production log for every track: which tools you used, what each did, and which parts survive in the final bounce. It takes five minutes during the project and saves hours if a distributor, label or collecting society ever asks.

If you release through a label, the label's distributor sets these fields, not you. Apple has said content providers are best placed to know how their music was made, and the label can only pass on what you tell it. Send your log with the masters and nobody has to guess.

Will disclosing AI hurt your streams or royalties?

Disclosing AI-assisted work shouldn't hurt your reach. Spotify has said its disclosure system isn't about punishing artists who use AI responsibly or down-ranking tracks that disclose. Deezer's exclusions target tracks its tool detects as fully AI-generated, not human productions that went through a mastering service.

The bigger risk is getting it wrong the other way. Platforms are investing heavily in detection, and a generated vocal that turns up undeclared can put a whole release under review, or get it pulled from a store like Bandcamp that bans substantially AI-generated music. Honest metadata is cheap insurance.

Royalties are where disclosure has real consequences, and Dutch producers have unusually clear rules. Buma/Stemra registers AI-generated portions under a dedicated party, AI BUMASTEMRA with IPI number 01252737944, and collects no royalties for those parts. If the lyrics are AI-generated and the music is yours, you're paid for the music. Works where you made enough creative choices register normally with full royalties, and members can document prompts, edits and choices in MyMusicSafe inside MijnBumaStemra.

What do the EU AI Act rules mean for producers?

The transparency obligations in Article 50 of the EU AI Act apply from 2 August 2026. Most of the weight sits on the companies that build generative AI: their outputs must be marked in a machine-readable way, and a provisional agreement reached in May 2026 gives systems already on the market until 2 December 2026 to meet that marking rule.

For producers, the part to know is the deepfake rule. If you use AI to create audio that resembles a real person, such as a cloned voice of an existing singer, you're expected to disclose it. For evidently artistic work the obligation is lighter: disclose in an appropriate way that doesn't spoil the work, for example in the credits or release notes. AI mastering, separation and mix tools don't touch this rule at all. This isn't legal advice, and if a release uses a real person's voice or likeness, talk to a lawyer first.

What to do this week

• Start a one-line AI log for each track in progress: tool, what it did, what survives in the final mix.
• Open your distributor's dashboard and check the AI fields on every release scheduled for the next three months.
• Replace any AI vocal modelled on a real singer unless you have their written consent.
• Buma/Stemra members: re-register any partly AI-generated works using the AI BUMASTEMRA IPI number.
• Decide on your artwork now, and if it was generated, declare it at upload rather than after release.

Transparency is fast becoming normal in music metadata. Producers who get used to precise, honest disclosure now will have nothing to fix when the rules tighten.

Quick answers