As the landscape of AI content licencing evolves, the news media industry must look at new ways to create sustainable models.

During this week’s Webinar, presented by the INMA Product & Tech Initiative, members received a warning from Dr. Jonathan Roberts, chief innovation officer at People Inc.: “Solving the information economy — this world which we live on, how we get paid, how we pay our writers — is necessary for there to be an AI future. So this is a problem for all of us.”

The industry is entering a pivotal moment in which publishers have more leverage than they realise — and more responsibility to shape what comes next.

Roberts, a former theoretical physicist who has spent 12 years at the company formerly known as Meredith, said the shift from 2024 to 2025 marked a fundamental change in how AI companies view information.

For years, he said, the prevailing belief in Silicon Valley was that training large learning models on the entire Internet would eliminate the need for new information. That assumption collapsed as models failed at accuracy and hallucinations persisted.

“We know now that the best inputs to AI lead to the best outputs,” Roberts said. “Getting the best AI from here going forward is going to require quality inputs of quality information, not just a mass quantity of input.”

In fact, he added, most information found on the Internet is not factual — and that is great news for publishers: “In the future, we’re going to need a lot more content, not less.”

That content will need to be nuanced and specific, not a mass of information gathered from general questions.

“There is not currently today an economy that supports this need. And again, that’s obviously a problem for all of us. It’s an existential threat for the AI economy,” he said. 

How things changed in 2025

Two developments in 2025 accentuated demand for better AI content, beginning with the Chinese model DeepSeek, which replicated the performance of leading U.S. frontier models at a fraction of the cost. Then, within 48 hours, the open‑source community replicated it.

That, Roberts said, proved that “throwing hundreds of billions of dollars at a new model” could not keep other companies from copying it.

A second change was that major AI systems began citing their sources. OpenAI, Meta, and others adopted retrieval‑augmented generation (RAG) to ground answers in verifiable information. Roberts said the change was driven by the need to ensure that the answers aren’t hallucinated:

“If you don’t cite your sources, the answer is untrustworthy,” he said, adding that he realised this is something that’s obvious to journalists, but not to tech titans. “We are very early in teaching tech companies what good information is.”

The shift placed publishers at the centre of the AI ecosystem. High‑quality, rights‑cleared information became the only reliable way to improve model accuracy. That, Roberts argued, gives publishers leverage they have not seen in years.

Why AI needs publishers

Roberts said AI systems rely on three inputs: chips, power and information. Chips and power are commodities. Information is not.

“If you only have chips and you only have power, you have a fully powered data centre, but nothing going through it,” he said. “The thing that goes through it is words.”

However, whilst chips and power are well-funded, information is not: “Why is the only unique thing here the most undervalued of the three in this revolution?

He noted that open‑source models with Web access already outperform cutting‑edge models without it, which underscores the value of high‑quality inputs.

“If you don’t have access to the best inputs, you’re going to have risky and lesser quality AI,” he said. Regardless, AI companies continue to scrape publisher content without compensation.

It’s important for platforms to be reminded that “there is no magic content theory,” Roberts said. “They have lived in a world where they don’t need to pay for content. Google is not a content creator. Meta is not a content creator. They built a system. That’s where they got everyone else to pay for the asset they need for their systems to be good.”

Building a market for AI content

The next step is for the news media industry to build a market that works, Roberts said.

“It has to work for publishers, obviously, because we actually do have to pay our writers, but it also has to work for the buy side because if we just put a tax in place and it has no benefit, then unless the system is better, nobody’s going to buy it.” 

The market must reward quality, not quantity, and must be designed intentionally to avoid the pitfalls that plagued search and social platforms.

“How we design the market determines what people create,” he said. “Leaving it to chance is naive.”

He outlined several emerging models, including tollgate systems, in which AI companies pay to access content; revenue‑share models, similar to Apple News; collective rights organisations, modelled after the music industry’s ASCAP or BMI; and more.

No single model will dominate, Roberts said, but all must be built around clear rights, transparent pricing and fast retrieval.

The economics of AI content

Roberts pushed back on claims that AI companies cannot afford to pay for content, saying the argument is based on a flawed assumption that publishers must be compensated for the full cost of producing information every time an AI system reads it.

Instead, he said, the economics align more closely with how publishers monetise human visits. AI companies already charge developers US$10 to US$50 per thousand searches when those searches require Web access.

That pricing, he said, is in the same range as what publishers earn from human traffic.

“You should not end up in a world where we’re making less than 10 bucks per thousand human visits,” he said, noting AI companies are already making money from the content — “They’re just not paying you.”

The evolving information economy

The next phase of the information economy will require more structure, more metadata and clearer rights, Roberts said. AI systems need to know when content was created, who wrote it, and whether it is authoritative. Publishers, he said, must help define those standards.

“There’s a number of initiatives; you can all be involved in all of [them],” he said. “All of these initiatives need many, many people involved, and all of these initiatives are being built for everybody, not for the few.”

He pointed to ongoing work at the IAB Tech Lab, Microsoft’s content marketplace and the RSL collective as early efforts to build the infrastructure for a rights‑based ecosystem.

He suggested news companies get involved to make sure they — not tech companies — are determining what good content looks like: “This is a unique moment to get in the game and determine what the answer is.”

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .