Music

YouTube Labels AI-Generated Songs, but the Label Doesn’t Touch the Money

Rivals are cutting recommendations and royalties for machine-made music. YouTube's answer is a tap-to-learn-more badge, with the definitions left to the partners who deliver the songs.
Noah Brandt
Add us on Google

The most important word in YouTube’s new policy for AI-generated music is “primarily.” YouTube Music has started putting a label on songs created primarily with generative AI, the kind of track where a model wrote the composition and produced the vocals, and the same label is heading to the main YouTube app. It reads like accountability. Look closely at what the label attaches to, and at what it leaves alone, and it looks more like a company paying the smallest available price for a problem its rivals are already charging for.

The pressure has been building from two sides. Listeners have complained that the service kept feeding them machine-made tracks they had tried to dismiss, and competing platforms have moved from disclosure to consequence: fewer recommendations, withheld royalties, filters that let users switch AI music off. YouTube’s answer is information. Tap the label on the player page and a menu explains the role AI played. Nothing in the announcement says the song will be recommended less, or paid less.

A label that leaves payouts alone

TechCrunch put the core of the rollout plainly: AI involvement alone does not make a track ineligible for monetization. Music Business Worldwide read both YouTube’s blog post and the Help Center page for music partners and found that neither says whether labeled tracks are treated any differently in recommendations or royalty payouts. The nearest thing to a statement comes from YouTube’s video labels, updated in May, where the company wrote that “a disclosure label alone does not change how a video is recommended or whether it’s eligible to earn money.”

That position is consistent. It is also convenient. YouTube’s post frames the move warmly, “Generative AI is opening up new ways for artists to create, experiment, and share music,” and the company sits on both sides of the trade. Its own Dream Track tool lets selected creators generate soundtracks with AI versions of participating artists’ voices, and under YouTube’s wider rules, content made with its own generative tools carries a disclosure that cannot be removed. The platform labels its own product and keeps the shelf open for everyone else’s.

The sellers decide what counts as AI

The mechanics hand the hardest judgment to the people selling the songs. Music partners, the labels and distributors that deliver the catalog, declare each track as “Fully Gen AI,” “Partly Gen AI” or “No Gen AI” through the industry’s DDEX delivery feeds or a CSV template. YouTube’s examples sketch the lines: a song downloaded straight from a prompt-based generator is fully generated; an artist who uses AI for a bassline or a string section and then adds live vocals and instruments is partly generated; Auto-Tune fixing pitch and automated AI mastering count as no generative AI at all. In the help page’s own words, “it is up to partners to interpret the Gen AI definitions provided by DDEX.”

YouTube keeps a backstop. When a partner supplies nothing, the track is treated as unknown unless other signals allow a call, and YouTube says it may then label content as fully or partly AI on the strength of its own detection. How those tools work, and how often they miss, the company has not said. TechCrunch flagged the same gap, along with the question of whether anyone checks a “Partly Gen AI” declaration.

And “partly” may be where the label goes quiet. The blog post promises the player-page badge for songs made primarily with AI. Music Business Worldwide notes that it does not say whether partly generated tracks get one, and that a help-page example in that tier is a fully human-performed song paired with an AI-made music video. The hybrid record, a real voice over a machine-built arrangement, is the production choice a listener is least able to hear and the one a badge could explain best. YouTube has left it the blurriest.

What rivals already charge

The comparison is what makes this a story about YouTube rather than about AI. Deezer, which said in July it receives nearly 90,000 fully AI-generated tracks a day, labels the music its detection catches, keeps it out of algorithmic recommendations and editorial playlists, withholds royalties from fraudulent streams, and said on October 13 that it will let users exclude fully AI-generated music from search. TIDAL said in June it would tag fully AI-generated music and stop paying royalties on it. Spotify announced in August an “AI Persona” badge for AI-generated artist identities, with their music kept out of editorial and algorithmic recommendations by default. Apple Music plans “Made With AI” labels later this year, relying on labels and distributors to declare. In July, the RIAA and IFPI proposed two voluntary labels of their own, “AI-generated” and “AI-assisted.”

Listeners have been clear about what they want. A Deezer and Ipsos survey across eight countries, cited by MediaNama, found that 97% of respondents could not reliably tell AI music apart and 80% said platforms should label it. MediaNama’s January report collected YouTube Music users who said AI tracks kept coming back after downvotes; one counted six of ten new recommendations as “AI slop.” At the time, the outlet noted that YouTube had published no policy on how it identifies, labels or de-prioritises AI music in recommendations. The new labels answer two of those three verbs.

A listener who taps the badge will learn what the machine did on the track. Whether that track was pushed into their queue, and who gets paid each time it plays, is a question the label was not built to answer.

Tags: , , , ,

Add us on Google

Discussion

There are 0 comments.