WTF is pay per ‘demonstrated’ value in AI content licensing?

If it works, it could be a way for publishers to reassert their pricing power with LLMs.

Where does the term come from?

It’s surfaced as one of the challenges and goals being defined in Microsoft’s publisher content marketplace. Its publisher pilot partners have spent the last six months trying to determine how they can plug into the marketplace and the pricing structure around it. For example, the difference in price for a deeply investigated journalistic report versus football scores. The publisher needs to set their pricing floors, and of course, the AI buyers need to think it’s worth buying. But that’s just one part of it — arguably the simplest.

How does this fit into other pricing structures?

There are currently four main ways publishers can get paid by LLMs for content licensing: lump sum deals, pay-per-crawl, pay-per-query (sometimes called pay-per-inference), and pay-per-value. 

With “pay-per-value,” AI platforms pay publishers based on the relative value that piece of content contributed to a query, and the value of that query, according to Paul Bannister, chief strategy officer at Raptive.

“This is the most complicated but by far the best model, and is somewhat similar to existing licensing models like ASCAP [and] BMI for music royalties,” Bannister said. “It aligns interest the most. It’s almost like a revenue share. Depending on how much money the platform makes, the publisher makes a percentage of that.”

What are some of the factors that influence pricing?

Factors that influence pricing include whether the content is being used to train an LLM, how often a publisher’s content is cited or referenced in AI-generated answers, whether the usage of the content will drive referrals or clicks back to a publisher’s site, and the relevance and timeliness of that content.

Types of content can also be priced differently, such as breaking news, evergreen explainers or product reviews. This could fluctuate with different events too, such as elections, sports seasons and breaking events.

For example, publishers could be paid differently for their content archive access versus real-time answers.

“One of the things we learned early on at TollBit is that a flat fee for access or crawling does not work at all. Content is unique, and each piece has a different value. This is why on the TollBit platform we have built a variety of vectors by which that content can be priced for RAG [and] agentic use, such as by publish time, pages, directories [and] keywords,” said Toshit Panigrahi, co-founder and CEO of TollBit.

On TollBit’s platform, premium publishers, breaking news stories, paywalled content, and unique content like local news or academic papers are priced higher, for example, according to Panigrahi.

Sounds really complex, especially without knowing why LLMs need specific pieces of content.

Right. Other factors for pay-per-value pricing could include the monetary value of a query to the AI platform, as it relates to timeliness — a query about making investment choices is more valuable than one about ranking the most popular Pokemon, Bannister said, for example.

One publishing exec — who asked to speak under the condition of anonymity — gave the example of an annual editorial franchise it publishes in in the spring. For the first 24 hours of that package being published, AI agents would try to scrape the original source — especially because the publisher blocks AI bots from accessing the story (unless they pay to license it).

“[We] would be able to get a good price for it right then. Maybe 24 hours later, [we] probably wouldn’t. There’s also all kinds of variables in a marketplace, like, is this [AI] agent looking to use this information one time? Are they looking to use it for a period of time? Are they looking to index this information forever? Are they looking to train a model on it? There’s all kinds of use cases that theoretically should factor into that,” the exec said.

Because of that, it’s a complicated process to determine what the demand side (in this case, LLMs) want the content for.

“What’s the value of what they’re putting it in, and how it’s going to be presented, and the value to the end user on their end? And then how do you accurately price it, so that there’s a fair value to the source of the information?” the exec said.

How far along are these pay-per-value markets?

Not very far along at all. Microsoft AI’s vp Nikhil Kolar told Digiday the pricing structure is still “extremely complex,” and that Microsoft was trying to figure out how to simplify it. He also stressed that every piece of content, call, or usage instance is priced individually, it becomes operationally complex and unpredictable. “That means we need to keep it consistent in some way for the demand [side] to know how much they will end up having to pay for it, but at the same time, it is currently extremely complex, so some simplicity has to come in,” he said.

The idea behind pay for demonstrated value is that in some cases, the value will be associated with a particular publisher source, but in other cases it might just be the raw information that an LLM needs to generate a response to a user’s prompt.

Really Simple Licensing (RSL Collective) recently hired Matt Lindsay, CEO and founder of Mather Economics, a firm that helps publishers develop consumer revenue strategies, to develop a pricing structure for the value of content licensed by LLMs, but it remains a work in progress.

Why does pay-per-value matter?

It all goes back to publishers pushing to get compensation from AI companies that is proportional to the value they believe is being extracted by LLMs.

It’s also a potential to recoup losses on the revenue publishers make from search referral traffic, which has eroded as AI tools answer queries with summarized responses. Publishers want compensation tied to how their journalism is used, cited and monetized inside AI systems. A value-based model offers recurring revenue, transparency into usage and a share of the upside if AI platforms profit from their reporting — helping ensure publishers remain economic stakeholders in the AI layer rather than invisible suppliers feeding it.

What complicates this?

There are still a lot of questions around how this will work: who controls the market valuing content? Is it the publisher, or the AI or tech company?

Also, how will “value” be defined, if there is a chance it can be subjective? While there are already some factors emerging that affect pricing, a compensation structure needs to be agreed upon by both publishers and AI and tech companies to build a functioning marketplace. 

For now, each deal is a bespoke one — and publishers can value their content as they see fit. It still remains to be seen if AI companies are willing to pay the price. Publishers are continuing to block AI bots to try to limit the scraping of their content without compensation — but it’s not very effective, meaning publishers still have limited leverage if AI and tech companies don’t come to the table to work out the development of these pay structures.

Will enough data be shared from the AI and tech companies with publishers for them to understand these metrics, and how it may impact the price of their content? If demonstrated value is based on how often publishers’ content gets cited, could payment skew toward larger brands that are already dominating in their visibility and citations in LLMs? These are all things that still need to be worked out.