Do you have to disclose that you used AI in a song?

It depends on where you are releasing it and what you used AI for. Distributors and metadata standards increasingly include AI declaration fields, some client and library contracts require disclosure, and some territories are introducing labelling rules for AI content. Outside those, disclosure to listeners is generally a choice.

Does using AI mixing or mastering count as AI music?

In most disclosure frameworks the concern is generated content rather than assistive processing. Automated mastering, noise reduction, pitch correction, and stem separation are usually treated as production tools. Where a form asks specifically about generated audio or generated composition, that is what it is asking about.

Should you tell listeners you used AI?

There is no consensus. Disclosure builds trust and protects you if it comes out later, and some audiences respond well to hearing how a track was made. Non-disclosure avoids a label that some listeners apply reflexively regardless of how much human work went in. The stakes are highest where a listener would feel misled.

Disclosing AI Use in Music: When It Is Required and When It Is a Choice

Disclosure splits into two questions with different answers. Where it is required — distributor forms, metadata fields, client contracts, library submissions, some territories' labelling rules — disclose accurately, always. Where it is voluntary — telling your audience — it is a genuine judgment call, and thoughtful people land on both sides. Requirements vary by platform and territory and are changing, so check what applies to your release rather than relying on a general article.

Where disclosure is required or close to it

Distributor and metadata declarations

Industry metadata standards have added fields for indicating AI involvement, and distributors increasingly ask at upload. Answer honestly. An inaccurate declaration is a policy breach in a way that an honest one is not, regardless of what the platform does with the information.

Client and commercial contracts

Sync briefs, library submissions, and commissioned work frequently include warranties about how the music was made and what rights you hold. Some now address AI specifically. Read them, and if a contract requires human authorship, do not sign it for a generated track.

Territorial labelling rules

Some jurisdictions have introduced or proposed requirements to label AI-generated content. These are new, differ substantially, and are still being implemented. If you release into a territory with such rules, check the current position there.

Competitions and grants

Most now have an AI clause. They vary from outright prohibition to a disclosure requirement.

Where it is genuinely your call

Telling your listeners is different. Here are the real arguments, without a thumb on the scale.

The case for disclosing

It survives discovery. If AI use comes out later — and detection is imperfect but people talk — having said it up front costs you nothing. Having concealed it is the story.

Some audiences are interested. Process is content. Producers who explain their workflow build a following on it, and AI tooling is part of a lot of workflows now.

It normalises honest use. The quieter people are about legitimate AI-assisted work, the more the conversation is defined by the worst examples.

It protects the relationship. Listeners who feel misled do not usually forgive it, and the line between assistance and generation is one many audiences care about even if they cannot define it.

The case against disclosing

The label flattens everything. A track where AI generated a drum pattern and a track generated entirely from a prompt get the same reaction from a lot of listeners, which is not a useful distinction for anyone.

Nobody discloses their other tools. No one lists their compressor, their pitch correction, their sample packs, or their drum library. Singling out one category of tool implies it is different in a way that is not always defensible.

It can invite bad-faith responses. Some audiences respond to any AI mention with hostility regardless of context, and a musician is not obliged to hand that to them.

Where is the line? Pitch correction, quantisation, drum replacement, and automatic mastering have been normal for decades and are all algorithmic. A disclosure norm needs a coherent boundary and does not obviously have one.

A workable position

Situation	What to do
A form asks	Answer accurately
A contract requires it	Comply, or do not take the work
AI generated the composition or the vocal	Lean strongly towards disclosure
AI generated backing parts you then arranged	Reasonable either way; disclosure is low cost
AI was used for mixing, mastering, separation, or repair	Generally treated as production tooling
You used an identifiable voice	Disclosure does not fix this — you need permission

The underlying test that most people can agree on: would a listener feel misled if they found out? A vocal that sounds like a human singer and is not is on one side of that line. A generated hi-hat pattern is on the other.

The one thing not to do

Do not claim human performance that did not happen. Presenting synthetic vocals as your own singing, or a generated performance as a session musician's, is the case where disclosure stops being a judgment call — it is a misrepresentation, and in commercial contexts it may breach warranties you have signed.

Nothing here is legal advice. Requirements differ by platform, contract, and country, and they are changing quickly enough that the current terms are the only reliable source.