Yahoo becomes first licensee of ‘Spotify-like’ collective AI content licensing model

Digiday spoke with ILC co-founder Doug Leeds about how this model works and how publishers can earn revenue.

How the model works 

A licensee like Yahoo pays a blanket licensing fee for access to content in the collective. Leeds declined to say how much Yahoo was paying, but the fee is tied to the “demand-side revenue.”  Yahoo declined to be interviewed for this story.

As a licensee’s business scales, the amount available for publisher compensation can grow as well.

“It allows businesses that are making a lot of money off of the content to keep that content in place and sustainable. But it also allows for entrants,” Leeds said. “As the revenue scales, the percentage scales. So more money comes out. But that also means that entrants can come in, and there can be new use cases and new players and new products, and all of that’s important to the evolution of what we’re doing to the market.”

A new report out today from publishing analytics firm Mather Economics estimated that collective licensing for AI could generate publisher compensation equal to 15% to 35% of AI company revenue, a potential market that in three years could surpass the U.S. display advertising market. The analysis was based on the current annual revenue from the two biggest frontier models, OpenAI and Anthropic, and an economic model created by Mather Economics incorporating AI training-data scaling laws.

Leeds likened the licensing collective to a Spotify-like model.

“It’s kind of like Spotify, but also a song is a song is a song, and content is very different. And so we had to come up with these rules that made sense specifically for digital content,” Leeds said.

How is that fee paid out to publishers?

Leeds said ILC worked closely with its publisher collective to figure out those rules. The first step was to define a “segmented royalty pool.” This separates licensing revenue by AI company, rather than combining payments into a single fund shared by all participating publishers. That means publishers only receive a share of the revenue generated by AI products that actually use their content. 

More AI and tech companies are testing pay-per-value licensing models. Digiday has previously reported that Google and Microsoft are testing and developing these models, which compensate publishers when their content meaningfully contributes to AI-generated responses, rather than paying based on how frequently their content is used. It’s a shift from flat-fee licensing deals to more dynamic pricing, based on the value publishers’ content delivers to AI companies.

When an AI company signs a licensing agreement with ILC, that money goes into its own royalty pool. Rather than combining all these payments into one enormous fund and distributing them across participating publishers, ILC keeps the royalty pools separate.

This protects specialized publishers from having their revenue diluted, Leeds explained. A general-purpose AI search engine like Yahoo Scout might use large quantities of content from hundreds or thousands of publishers, but a specialized healthcare AI product might use a much smaller amount of content from a handful of medical-focused publishers, and that specialized content could be considerably more valuable to the healthcare company, he said.

ILC wanted to ensure that a publisher’s compensation reflects the value of the AI product using its content, not simply how much content it contributes to the overall licensing ecosystem.

“An AI vertical in health might be very low volume, but much higher specific value. So if your content gets used in that, you should split that high value, not get diluted by all the content that got used by Yahoo,” Leeds said.

Over 3,000 publishers are now part of the ILC, he noted.

How does ILC determine how much money each publisher should receive from the relevant pool?

The payments are determined by what Leeds called “allocation rules.” These take into account how much content is used, like grounding, retrieval or displayed in an AI response, and whether the publisher is credited or cited as a source.

For example, an AI product that uses a publisher’s content to check a fact might receive a lower allocation than one that displays detailed election results and explicitly credits the Associated Press, Leeds said. He did not disclose the exact formula used to calculate each publisher’s share.

The allocation rules are intended to discourage AI companies from choosing content simply because it’s cheaper, and instead incentivize them to use what they consider the highest-quality sources, he said. 

Publishers can also decide who they want to license to. Right now, their only option is Yahoo. But as more licensees sign up, they can go into their ILC portal and select which AI companies they do and don’t want to license their content to. For example, if a publisher signs a separate content licensing deal with an AI company, they can take their content out of the collective. In the portal, they can also see what their content is being used for, such as RAG or training. 

Behind the name change

RSL, which stood for the Really Simple Licensing (RSL), was two things: a group of publishers working on a collective licensing agreement, and the organization behind the RSL Protocol, a standard that publishers can use in robots.txt to describe in machine-readable language what their terms of service are. Publishers can ask AI companies for a pay-per-crawl or pay-per-inference fee through the RSL Protocol, for example. “Thousands” are currently using this standard, Leeds said.

But Leeds wanted a new name for the non-profit organization that reflected the publisher collective, to make it clear that this is a licensing model that any publisher can join, he said.

Yahoo is the first, now what?

Leeds said ILC is currently in talks with “some” of the major AI companies interested in the licensing collective. He declined to name which ones. 

“You can do a deal with a one-off, and that’s fantastic. But that doesn’t invest in the infrastructure of content in general,” he said.