Every platform now labels AI music differently, and the burden lands on artists

YouTube Music started putting labels on songs "created primarily using generative AI" on 14 Oct 2026, and says the labels will spread to the main YouTube app. The label sits on the player page, and tapping it opens a short explanation. YouTube says it decides which tracks get one by "teaming up with music partners" and by running its own detection tools.

That makes five large services, plus a recording-industry scheme, each with its own way of describing AI in music. For listeners that's a small improvement. For artists, especially the many who use AI for one part of a track and do the rest by hand, it's a new admin job, and the cost of getting it wrong differs from one service to the next.

How does YouTube's system work?

Music Business Worldwide went through YouTube's Help Center guidance. Partners (labels and distributors) declare each track through DDEX or CSV delivery templates as one of three things:

• Fully Gen AI, for example a track downloaded from an AI generator after someone typed a prompt.
• Partly Gen AI, for example an artist who generates a bassline or string part with AI and then adds live vocals and instruments.
• No Gen AI, which still covers pitch correction such as Auto-Tune and automated AI mastering.

Two details matter more than the categories. First, YouTube leaves it to partners to interpret the DDEX definitions for their own releases. Second, if a partner declares nothing, YouTube relies on its other signals alone and may label the track fully or partly Gen AI. Silence is no longer neutral: if you don't declare, the detector decides.

MBW also notes that neither the blog post nor the Help Center says whether a label changes recommendations or payouts, or whether "Partly Gen AI" tracks get a visible label at all. One of the Help Center's own "Partly Gen AI" examples is a song with entirely human-performed audio and an AI-generated music video.

How do the others do it?

Every service draws a different line and enforces it differently:

• Apple Music introduced AI Transparency Tags in March, which labels and distributors use to flag AI in a track, its composition, its artwork or its video. An August memo seen by MBW says these become visible "Made With AI" labels later this year where AI made a "material portion" of the content, including tracks "primarily derived from a generative AI service". Apple doesn't run its own detection for this. It relies on self-declaration, saying content providers are "best positioned to know how their content was created".
• Spotify began testing AI disclosures in song credits in April, supplied by labels or distributors (MBW). Separately, it badges AI Persona profiles, artists who present an AI-generated, photorealistic identity, and keeps their music out of editorial and personalised recommendations. Spotify's artist help page says the badge doesn't apply if you only use AI tools to make your music, and it is the one system here with a published appeal: an artist flagged by Spotify gets a notice in Spotify for Artists and can contest it.
• TIDAL said on 29 Jun that it would stop paying royalties on music it identifies as wholly AI-generated, effective immediately, and add a visible "AI" tag from 15 Jul. It plans to extend the tag to "substantially" AI-generated music as its detection improves. As The Verge noted, TIDAL hasn't said what tools it uses, and it says it will "begin to enforce" an expectation that distributors flag AI content before delivery.
• Deezer detects and tags AI music with its own technology. On 13 Oct it said listeners will be able to exclude fully AI-generated tracks from search, and that it receives nearly 90,000 of them every day (MBW).
• The RIAA, IFPI and partners including A2IM, IMPALA, WIN, the Grammys and SAG-AFTRA proposed a voluntary two-label scheme on 10 Jul. "AI-Generated" means AI produced all or the primary part of the creative elements, such as the lead vocal or a key instrumental performance. "AI-Assisted" means the recording was made substantially by humans, who performed the lead vocal and main instruments, with AI used for some elements. It covers sound recordings only, not lyrics, composition, videos or artwork.

Put one ordinary release through all of these. A songwriter writes and sings a song, plays guitar, and uses an AI tool to generate a string pad. On YouTube that should be "Partly Gen AI". Under the RIAA/IFPI scheme it reads as "AI-Assisted". On Spotify it's a disclosure against one instrument in the credits. On TIDAL it should earn normally, because it isn't wholly AI-generated, unless a detector disagrees. On Apple it depends on whether the provider judges the strings a "material portion". Deezer runs its own detector regardless. The artist usually doesn't type any of this directly: it passes through a distributor's upload form.

What artists are saying

Two posts we saw from working musicians on X in the days after the launch were practical rather than angry. One artist and co-founder of IMU urged musicians to tell their distributor exactly how they used AI before an algorithm labels them, and to treat metadata as something they own. A producer summed up YouTube's approach as disclose rather than ban. Both point the same way: under these rules, disclosure is the only protection an artist controls.

What we think

The industry is labelling at the wrong layer. Nearly every one of these systems asks the same question (how much of this recording came from a generative model?) and then stores the answer in a separate, platform-specific field, filled in by a middleman and checked by a private detector. The answer belongs with the recording, the way ISRCs and songwriter credits do: declared once by the artist, carried in the delivery metadata, and read the same way by every service.

The RIAA/IFPI split between AI-Generated and AI-Assisted is the right line for that standard, because it hinges on who performed the lead and main parts. YouTube's "Partly Gen AI" is too broad to be useful: a bucket that holds both a mostly generated track with a human topline and a fully human song with an AI music video tells listeners almost nothing, and lumps careful artists in with the people the labels were meant to catch. It's AI-assisted work that needs the precise wording, because that's where a wrong call costs a working musician. An artist who writes and performs, and uses an AI co-producer (Veena is one) to try arrangement ideas, is doing something different from someone who downloads a finished track from a prompt.

The stakes are no longer cosmetic. TIDAL already ties royalties to its "wholly AI" call and plans to widen that to "substantially", and YouTube will apply a label when partners stay silent. A false positive in that system costs real money. Spotify has shown an appeal route is possible for its identity badge, yet none of the track-labelling announcements we read, from YouTube, Apple or TIDAL, describes how an artist can contest a wrong call. Two commitments would go a long way: platforms should treat an artist's declared "AI-assisted" as binding unless a human reviewer overrides it with notice, and distributors should put the declaration field in front of the artist, in the shared RIAA/IFPI terms, instead of burying it in a dropdown.

Until then, our advice: write one plain sentence for every release saying what AI did and didn't do, give it to your distributor, and keep a copy with your stems and project files. It's the nearest thing to a paper trail you have.

Sources

• YouTube Music adds new labels for AI music, YouTube Official Blog
• YouTube Music starts labeling AI-generated songs, using partner disclosures and its own detection tools, Music Business Worldwide
• Apple Music to slap AI labels on Suno tracks, but pushes the onus onto record companies, Music Business Worldwide
• TIDAL to automatically tag AI-generated music and block it from earning royalties, Music Business Worldwide
• Tidal won't pay royalties on AI-generated music, but isn't banning it outright, The Verge
• Music community introduces new labelling programme to distinguish generative AI in sound recordings, IMPALA
• AI Persona badges on Spotify, Spotify for Artists Support
• Post on disclosing AI use on your own terms, X
• Post on YouTube Music's three-way labels, X