The YouTube Music AI Label: Who Actually Sets Your Gen AI Tag

YouTube Music started labelling AI-generated songs on 14 October 2026, and four days of coverage largely explained the three tags without answering the question an artist actually has. A youtube music ai label is not something you choose at mixdown. It arrives as a field in a delivery message, set by whoever ships your release — or, if that field is empty, set by YouTube.

That distinction decides what your listeners see. Get it wrong and you have a track tagged by a detector you cannot inspect, on a platform where your distributor had a field it did not use.

We read YouTube's announcement post and both of its music-partner help articles at source on 20 October 2026, including the DDEX contributor examples that nobody has quoted yet. What follows is the tag system, the four routes a declaration can travel, and the two things YouTube has conspicuously not said.

What YouTube Music shipped on 14 October 2026

The announcement is six sentences long. "Starting today, YouTube Music is rolling out labels to make it easier to see when a song was created primarily using generative AI," it says, giving as its example "using a generative model to create the composition and vocals". The labels "will also expand to the main YouTube app", with no date attached.

On placement, YouTube is specific: "When you listen to a track on YouTube Music where AI played a significant role, you'll see a label directly on the player page. Tapping the label opens a simple menu that shares more information." On how tracks are identified, it is deliberately two-sided: "We're teaming up with music partners and using our own detection tools to identify AI-generated tracks, applying the same labeling principles already used across YouTube."

The mechanics live in a separate place — a help-centre article for music partners titled Disclose Gen AI usage for music content. That page, not the blog post, is where the three designations, the delivery routes and the field names are. It also carries the sentence that should be the headline, and was not: "If partners do not include any Gen AI information for a piece of content, YouTube will rely solely on those signals and may designate content as fully or partly Gen AI."

Two documents, two audiences. The blog post tells listeners a label exists; the help page tells partners who sets it. Almost all of the launch coverage read the first.

Fully Gen AI, Partly Gen AI, No Gen AI: YouTube's own examples

The help page lists three "affirmative designations" a partner can choose: Fully Gen AI — "All content was created using Generative AI"; Partly Gen AI — "Part of the content was created using Generative AI"; No Gen AI — "No content was created using Generative AI".

Definitions that terse do not settle real cases, which is presumably why YouTube also publishes worked scenarios. Those are the useful part, and the part nobody has written up. Here they are against the workflows people in this niche actually use, with the CSV value each sends.

One line on that page governs all of it: "It is up to partners to interpret the Gen AI definitions provided by DDEX in relation to their content." YouTube is not adjudicating. Two distributors handed the same stems can file them differently, and both will have followed the documentation.

Auto-Tune and AI mastering are No Gen AI

The last two rows are the ones worth memorising, because the instinct runs the other way. An engineer who ran a vocal through Auto-Tune and a master through an automated AI service has used two things marketed as AI and has, by YouTube's own examples, nothing to declare.

The logic is consistent once you see it: the framework classifies whether content was generated, not whether software was involved. Pitch correction operates on a performance that already exists; AI mastering adjusts equalisation, compression and balance on a finished mix. Neither creates composition, performance or vocals.

That matters in both directions. A producer who over-declares out of caution puts a label on a human record that YouTube's own examples say is unlabelled — a reputational cost paid for nothing. The counter-intuitive case the other way is the music-video row above: audio recorded entirely by humans still lands at Partly Gen AI if the video shipping with it was generated.

So the question to ask is not "did I use an AI tool". It is: was any part of the released content generated rather than performed, written or played?

The four routes a Gen AI declaration can travel

This is where the launch coverage stops and where the practical answer lives. There is no single AI checkbox. There are four distinct delivery routes, each with its own field, and which one applies to you depends entirely on how your music reaches YouTube.

Three things in that table are worth drawing out, because they change what you can realistically do.

The DDEX route is per-role, not per-track. The declaration is not a flag bolted onto a release; YouTube's DDEX composites help article says the <Contributor> composite "enables differentiation between human creators and AI", and that "when AI is used, replace the individual party reference with <SpecialContributor>GenerativeAI</SpecialContributor>". Its published example files an AI lyricist in exactly that shape, with <AiContribution>All</AiContribution> beneath the role. So the generative declaration occupies the slot a human credit would have occupied — which is why a declaration can show up in a listener-facing credit rather than only in a label. That knock-on is a story of its own, and our sister site has mapped it in DistroKid AI credits.

The DDEX and CSV routes are both gated. Each requires Studio Content Manager, which is held by labels and distributors, not by individual artists. The declaration is therefore something your distributor does on your behalf, whether or not it exposes a field to you at upload.

The declaration covers more than the song. The CSV asset id accepts "any YouTube Asset ID where the Content Owner holds ownership, including Sound Recordings, Art Tracks, Music Videos, and Web Videos" — so a release can be declared inconsistently across its own assets.

Whether your distributor surfaces any of this is a separate question, and one the industry has not answered in public. Spotify said in September 2025 that it had "commitments from 15 labels and distributors" to adopt the DDEX AI disclosure standard but never named them. DistroKid moved its own AI self-disclosure out of beta on 14 May 2026, and Digital Music News reported at the time that artists "will only encounter the expanded questions while uploading new material" — new deliveries, not the back catalogue. For the two distributors most people in this niche use, our DistroKid AI music policy and TuneCore AI music policy pages track what each one publishes.

What happens if your distributor declares nothing

An absent declaration is not a neutral state, and this is the single most consequential sentence YouTube has published on the subject: "YouTube may use additional signals to augment partner declarations. If partners do not include any Gen AI information for a piece of content, YouTube will rely solely on those signals and may designate content as fully or partly Gen AI."

Read the sequence carefully. The declaration is an input to YouTube's judgement, not a substitute for it — detection runs alongside a declaration, not only in its absence. And where there is no declaration, the two outcomes YouTube names are fully and partly. No Gen AI is not among them. Both documented fallbacks are AI designations.

There is a separate internal state for silence. "If you do not provide a value, YouTube will interpret it as unknown unless we are able to make a determination based on other signals." So an undeclared track sits at unknown until the detector reaches a view, and then it does not.

YouTube has published nothing about what those signals are, how accurate they are, or whether a detection-applied designation can be appealed. The closest precedent is its May 2026 video update, where it said it would apply labels automatically on detecting "significant photorealistic AI use" even where creators had not disclosed.

Practical consequence: if you have made a human record and your distributor exposes no declaration field, you have no documented way to assert No Gen AI short of delivering through Studio yourself. Silence is only safe where you would have declared all anyway.

Does the label show on a Partly Gen AI track?

Nobody has said, and the absence is not an oversight in the reading — it is genuinely open in both documents.

The announcement describes labels for songs "created primarily using generative AI" and for tracks "where AI played a significant role". Neither phrase lands cleanly on Partly Gen AI, which by YouTube's own definition covers a track with a generated bassline under live vocals. "Primarily" suggests not. "A significant role" suggests perhaps. The help page, which is the document that defines the three values, says nothing at all about display — its subject is declaration.

Music Business Worldwide flagged the same hole on launch day, noting that the blog post "does not specify whether tracks designated 'Partly Gen AI' will carry the player label". Jack Righteous, whose launch-day guide is the most thorough page on this topic anywhere, does not resolve it either — there is nothing published to resolve it with.

Until YouTube says otherwise, the working assumption for anyone declaring Partly should be that the label displays. It is the cheaper error: a Partly declaration that turns out to be invisible costs you nothing, while a No Gen AI declaration chosen to dodge a label you were not going to get is a false statement in a delivery message.

Does an AI label change recommendations or payouts?

YouTube has not said, and the gap is specific enough to state precisely rather than hedge.

In its May 2026 update on AI labels for video, YouTube said "a disclosure label alone does not change how a video is recommended or whether it's eligible to earn money". That is a clear commitment — about video. Neither music document repeats it. The help page says partners should disclose "to help YouTube accurately treat and label media that contains Gen AI", and as Music Business Worldwide observed, neither document spells out what "treat" involves.

The fair reading: the label is a disclosure, there is a nearby precedent for disclosure being consequence-free, and it has not been extended in writing to music. That is weaker than "AI music is safe here" and much stronger than the alternative, because three other services have gone further the other way.

Deezer's number is the context for all of it: it reports receiving nearly 90,000 fully AI-generated tracks every day. Our Deezer AI music policy and Spotify AI music policy pages go through each regime in detail.

A note on the earnings question underneath this, because labelling is not the hard part — reach is. If a platform-side tag can shape who hears a track and nobody will commit in writing about how, income that depends on algorithmic discovery is exposed in a way that a direct sale or a paid sync placement in TV, film, games and ads is not. Neither of those runs through a recommendation engine.

What to check before your next delivery

None of this needs a change to your music. It needs five minutes on your delivery, once.

The last row matters even if you never need it. Across the five platforms above, YouTube's unknown value is the only published way to undo an AI declaration filed incorrectly. It is a reset rather than a correction — it clears the value and returns the asset to the undeclared state, where detection applies — and it lives behind Content Manager, so in practice it is a request to your distributor. But it is documented, which is more than anyone else offers.

The label sits at the end of a metadata pipeline, and every decision in that pipeline is made by someone other than the person who made the record. Knowing which field carries it, and who holds that field, is the whole of the practical advice. For the monetisation side, our AI music on YouTube and YouTube Content ID pages cover claiming and earning — separate systems from labelling, and routinely confused with it.