Can Publishers Flip the Split on $42B AI Music Licensing Opportunity?
AI is changing music licensing, giving publishers a chance to claim a bigger share of revenue. Learn how attribution technology can help songwriters and publishers take control and get paid.
Can Publishers Flip the Split on $42B AI Music Licensing Opportunity?
by Benji Rogers of Lark42 from Substack
Remember:
Our human creations are what AI companies call their “training data.”
Our work is being taken and used without permission, and most often without remuneration.
Petitions won’t work. Creators must show AI companies what their work is, without sending it to them, in order for them to do the right thing.
A Rare Moment, Soon to Pass
The rise of AI in music creation presents a unique licensing opportunity for publishers to potentially reverse the traditional Market Share revenue split that record labels previously negotiated with platforms like Apple Music and Spotify.
This shift is made possible by the critical role that publishers’ assets play in AI music generation coupled with a unique quirk that attribution technology enables.
The Original Digital Sin (As a Publisher Might See It)
When the labels licensed their sound recordings to streaming platforms like Apple and Spotify, they landed on a ~70/30 revenue split:
- ~70% goes to sound recordings (controlled by labels)
- ~30% goes to compositions (controlled by publishers)
However, it is possible that AI-generated music, when paired with accurate and persistent attribution, fundamentally reshapes this dynamic in favor of songwriters and publishers.
Why Publishers Have the Upper Hand (for Now!)
- AI Prompts Focus on Songs, Not Recordings: Users who interact with AI music platforms typically request “songs” rather than sound recordings. (“Make me a song that sounds like Taylor Swift.”) To me, this fundamentally aligns more closely with the songwriter/publisher. As we have seen from our attribution technology, during generation, AI models draw from composition elements, such as melody and, since we rolled it out, the song’s lyrics and, crucially, its MIDI profile.
- Core Copyright Elements: As we have scanned and audited various non-compliant AI platform song outputs for our clients, it has become clear that the fundamental components that AI models use to generate music – melodies, lyrics, and composition structures – would seem to fall heavily under the publisher and songwriters side of the equation.
- Attribution Technology: Even with non-compliant or unlicensed AI platforms, a sufficiently powerful attribution system can demonstrate how much a composition might have influenced an AI-generated song, providing a strong basis for negotiation at scale and a real-time view of where works are being used.
Flipping the Split
Given these factors, publishers could negotiate with AI companies the key split in their favor. But only if they act quickly to define their assets digitally and demand attribution at the level of the composition’s influence. In a best-case scenario, this could simply reverse the current market share licensing model that has long been a source of contention:
- ~70% to publishers
- ~30% to labels
No Attribution, No License. Period!
Attribution technology should be table stakes in any licensing discussion. And no, it can’t be offered by the AI company itself. That’s not a thing. And if it were, could we ever trust them to audit themselves after scraping and training on our songs without permission? By providing robust attribution for lyrics, melodies, and MIDI data, songwriters and publishers now have a powerful technical solution to:
- Protect their intellectual property by opting into or out of AI training. (hint – opt-in is better – but you do you)
- Control how and which songs AI systems use with granular and machine-readable rules.
- Monetize their contributions to AI-generated music and get a real-time view of when AI models are using their works and for what. (Trust but verify)
What this means…
- Increased Revenue: A larger share of the AI-generated music market (estimated to reach a cumulative $42 billion by 2028) is better than whatever is left over from the label and distributor deals already being executed and announced.
- Negotiation Power: Publishers who can show AI models what their works are via APIs can set precedents and rules in AI deals, which would lead to advantages in future negotiations, including when it comes to synthetic data.
- Value Recognition: This shift in technological power acknowledges the fundamental importance of songs and songwriting to the AI music creation process.
- New Licensing Models: Publishers can pursue innovative attribution sharelicensing tailored explicitly for AI usage, which will be far more lucrative than traditional DSP deals.
The Biggest Risk
These next months are likely the biggest potential windfall for the songwriting and publisher community or the biggest fumble in the music industry’s history. I fear that they will wait and see, sue, or simply allow the sound recording owners to beat them to the party.
This is a singular opportunity to rewrite the rules and put the song in its rightful place in the AI-generated music market. A market that has not just arrived but has already spawned multiple multi-billion dollar companies with more announcing music models all the time.
Publishers can lead the charge and make their mark for a long time by leveraging their ownership of the fundamental building blocks of their creations and embracing cutting-edge attribution technology, or they can accept what is left over by the first movers in the space. Today that seems to be the case with labels and distributors.
We shouldn’t fear users typing “make me a song that sounds like Adele” into platforms like Suno or Udio if those AI companies agree to neutral attribution technology. Knowing whose work went into creating the resulting song ensures that the songwriters and publishers can get paid accordingly.
It’s Attribution Share for the win.
One final point…So if I can get an AI to create a song, and I can see from my attribution dashboard that it was made from 5-10 other songs (to keep it simple), who would own the composition of the newly generated song? I know that an AI can’t create a copyright, but what if it’s merely assembling a new song from 5-10 existing songs? Who should/could own the publishing on that?
Something to think about, no?
