How to disclose AI use properly: DDEX ContainsAI, AiContribution and Spotify AI Credits

Most advice on AI disclosure stops at "be honest". That is the right instinct and the wrong level of detail. The system Spotify now reads is not a yes-or-no box. It records AI use role by role and file by file, and it only reaches a store through a distributor that has built support for it. If you understand the structure underneath, you can disclose exactly what you did, no more and no less.

This is a practical guide. The launch itself, a beta that Spotify started on 16 April and that Music Business Worldwide reported began with DistroKid users, is background here.

What is actually in the standard?

DDEX's ERN standard defines the new-release message that carries release metadata, and version 4.3.2 of its data dictionary includes two AI fields. They work at different levels.

• ContainsAI, per file. It sits on a sound recording, text, image, video and the release itself. Its allowed values are None (no content created with generative AI), Partly (part of it was) and All (all of it was).
• AiContribution, per contributor. It sits on each contributor entry, next to that person's roles. It defaults to All when the contributor is the special type GenerativeAI; for anyone else it can only be None or Partly.

GenerativeAI is one of four special contributor types, alongside Traditional, UserDefined and VariousArtists. So in the standard, an AI system can appear in the credits as a contributor in its own right, with a role.

Two consistency rules tie the levels together. If a recording is marked All, only GenerativeAI contributors should be listed. If it is marked None, no contributor should carry Partly or All. One oddity: the same dictionary page still describes ContainsAI as a true/false flag in its header, while its data type uses the three values. The three-value set is what the rules are written against, so that is what we follow.

What does Spotify show?

Spotify's help page says AI credits apply to individual roles, not the whole track. Its common examples are lyrics, vocals, instrumental performances and production. The credits are in beta, visible only on mobile, in Song credits and the Now Playing view, and not on artist profiles or as a track label. Disclosure is optional. The key line for anyone with a mixed session: if AI and human contributions exist for the same role, credit them separately.

You cannot add these credits in Spotify for Artists. Spotify says you add them through your distributor at upload, and only certain distributors support them. Its newsroom post admits that not all distributors enable artists to disclose yet, and lists the partners working on the standard, from Amuse and CD Baby to DistroKid, EMPIRE, FUGA and RouteNote. Being on that list is not the same as offering the form, so check your own distributor.

How do you disclose, step by step?

1. List every element of the track. Lyrics, melody, each vocal part, each instrument part, mix, master, artwork. Write who or what made each one.
2. Sort generated from assisted. DistroKid's explainer says credits are for generated material: AI audio, AI lyrics, AI melody or arrangement. It says pitch correction, AI-assisted mixing or mastering and AI-assisted workflows don't need a credit. Other distributors may draw the line differently, so read theirs.
3. Give each generated element a role. Use Spotify's buckets: lyrics, vocals, instrumental, production.
4. Split shared roles. If you sang the lead and a model generated the harmonies, that is two credits under vocals: you, and the AI part. Don't merge them into one "AI vocals" line.
5. Pick the whole-file value. None if nothing generated survives in the final file, Partly if some does, All only if everything the listener hears is generated.
6. Find your distributor's form. On DistroKid, the upload flow asks whether any of the track was generated by AI. If you say yes, you tick at least one of four boxes: the lyrics, the music, all of the audio, or part of the audio. Ticking "all of the audio" brings up a follow-up asking whether the artist is a human or an AI persona.
7. Fix it later if needed. DistroKid says you can add or update credits after upload, from the Credits link next to each song on your album page.
8. Check the result. Open the track's Song credits in the Spotify mobile app after release and confirm what appears matches what you meant.

Notice the mismatch in step 6. DistroKid's four boxes don't map one-to-one onto Spotify's four roles, or onto the per-contributor detail DDEX can carry. A form with fewer fields than the standard loses detail on the way. That is our reading of the two pages, but it's why your own written record matters.

A template for honest disclosure

Keep this with your session files, and use it to fill in whatever your distributor's form asks.

Track: [title]   Version: [final master date]
Whole recording: None / Partly / All

Lyrics:       [Name], human. AI: none.
Melody:       [Name], human. AI: none.
Lead vocal:   [Name], human. AI: none.
Backing vox:  generated with [tool], arranged and edited by [Name]. AI: Partly.
Instruments:  drums [Name]; pads generated with [tool]. AI: Partly.
Production:   mixed by [Name] using AI-assisted tools (assistive, no credit needed per [distributor] guidance).
Artwork:      [Name], human. AI: none.

Tools used:   [tool, version, what it generated]
Kept as proof: project file, stems, dated bounces, licences.

If you write liner notes, the same lines read well in plain prose. The point is that each role carries its own answer, and the human names stay visible next to the AI ones. Tools that blur the line make this worth writing down: Veena, for example, has a CoProducer that can generate beats, chords and melody as editable tracks while you make the calls, so a part it generated is a credit and a mix you made yourself in it is not.

Quick checklist before you submit:

• Every generated element has a role.
• Shared roles are split into a human line and an AI line.
• The whole-file value agrees with the role lines.
• Assistive-only tools are recorded but not credited.
• Your distributor actually passes AI credits to Spotify.

This is general guidance, not legal advice.

What we think

The standard is better than the debate around it. "AI or not" is a crude question; None, Partly or All per file, plus a contribution value per person, is close to how music actually gets made. Artists who used a model for one stem get a way to say exactly that, rather than being lumped with fully generated uploads.

The weak link is the middle of the chain. Spotify has promised that the system is not about punishing or down-ranking artists who disclose. But artists only reach the standard through distributor forms, which can be coarser than the standard, and Spotify itself says a missing credit doesn't mean AI wasn't used. So the honest artist carries the label while anyone who stays silent carries nothing. Distributors should expose the full role-by-role detail. Until they do, the best protection is a precise record of your own. For what happens when a detector disagrees with you, see our guide to how distributors decide a song is AI, and for the technical marks that travel inside files, our explainer on audio watermarks and content credentials.