What share of what Spotify pays out makes it's way into the pockets of song writers and musicians is a more complicated story, generally more if the artists are with a good indie label, generally less if they are with a major. At the same time, majors have had to offer less abusive deals than they used to, because DIY and indie distribution more viable.

The other big shift is that in the retail days new releases drove most purcahse, but with streaming catalog is a source of reliable recurring revenue, and the majors own a lot of catalogue, especially stuff they acquired outright in an era when artists often had their work basically stolen from them.

The key difference between Spotify and LLMs scraping the open internet is provenance. Music on Spotify does not just appear there out of nowhere. It arrives through an accountable chain: a label, a distributor, an aggregator, a publisher, a rights holder. Sometimes this chain is thin, like with self-serve, pay to publish distribution through companies like CD Baby. Most of what is actually streamed has a provenance that reflects serious editorial and financial commitment by an organisation in the form of money spent recording, developing, and promoting an artist. This provenance chain is critical contextual information about who vouched for the work, who invested in it, who holds rights to it, and when it entered the culture. Art, music, writing do not exist in a vacuum. They are part of an ongoing cultural conversation, and who said what, when, and under what institutional backing is integral to its meaning.

So I share OP's hope the long-run equilibrium for LLMs looks more like licensed media than scraping and open web search. I want a world where models license published content from rights holders, not for training, though that would be nice, but to surface answers with links to identifiable sources in a verifiable published database, and let part of my subscription pay for access to the underlying referenced material. Information is valuable, and it's reasonable to pay for it. Aligning incentives around truth is the challenge.

Putting ink on paper and moving books around is the least important part of what a publisher does. The important part is selection, investment, positioning, promotion, and accountability. This curatorial function has always been important, and it can only become more important the tsunmai of ai slop and misinformation grows. I hope that chatbot manufacturers partner responsibly with rights holders and lean into the value that publishers have created instead of potentially destroying it.