Social Rocket
According to the New York Times, the market for AI content licences is becoming more publisher-friendly, but it is still far from a stable market model. Adam Greenberg, Vice President of Strategic Partnerships at the Times, named three conditions at the Digiday Publishing Summit that an AI licensing deal must meet from the publisher's perspective: a fair value exchange, a sustainable partnership and clear control over how the content is used.
These statements are interesting for publishers because they shift the debate away from the simple question "licence or no licence?". Increasingly, what matters is how content is used, paid for, linked and further processed in AI products.
Three conditions for an AI deal
In the Digiday report, Greenberg describes three principles.
First: value exchange. If an AI provider wants to use Times content, according to this logic there needs to be an agreement or explicit permission. Content should not simply be treated as free raw material.
Second: sustainability. The Times does not just want to collect a one-off lump sum. A deal should enable a longer-term collaboration and leave room for new products, better distribution and further revenue models.
Third: control and guardrails. The publisher wants to be able to track and limit where and how its content is used. Especially for training, inference, summaries and derived outputs, this can become contractually far more complex than classic syndication.
The Amazon deal is the current reference model
As an example, Greenberg cites the existing agreement with Amazon. The Times had already concluded the multi-year deal in May 2025. Reuters confirmed at the time that Amazon may use content from New York Times, NYT Cooking and The Athletic for, among other things, Alexa, short summaries and training its own foundation models. Financial terms were not officially disclosed.
For the Times, it is important that users can also be directed back to its own products when relevant Amazon experiences arise. This is precisely what makes AI licensing more interesting for publishers than a pure data sale: attribution, links and referral mechanics can be at least as relevant in the long run as the licence fee itself.
The economic counter-position is shown by the current dispute with OpenAI and Microsoft. Our article on OpenAI, paywalls and the New York Times lawsuit explains why the publisher argues much more firmly when use is unauthorised.
The AI licensing market remains hard to price
Greenberg names several unresolved questions: How is an individual piece of content valued? How do you prevent a race to the bottom on prices? What happens to outputs based on licensed content? And is the compensation high enough not to endanger a direct user relationship or a subscription business at all?
For subscription publishers in particular, this last point is central. A deal can bring money in the short term and still be problematic in the long term if the AI answer replaces the visit to one's own site.
Other publishers are therefore testing different models. Google is experimenting with pay-per-use compensation for publisher content in AI products. Axios is building Axios Direct, machine-readable news feeds for AI agents and models. And Wikipedia is trying to manage its enormous AI reach more economically and technically through enterprise APIs and direct platform partnerships.
What German publishers can take from this
The three Times rules also provide a useful review framework for smaller publishers. Before an AI deal, it should be clear which content a partner receives, for which products it may use it, how use is measured, what attribution is foreseen, and whether the deal strengthens or weakens the direct relationship with the audience.
The market is thus moving away from blanket training-data deals towards more differentiated contracts covering training, inference, distribution and referrals. That is where it will be decided whether AI licensing becomes a viable revenue model for publishers or merely another form of platform dependency.