News licensing for AI developers, explained

For most of the internet's history, news licensing was a quiet business. Aggregators, research databases, and media monitoring services paid publishers for the right to carry their articles, and few people outside the industry paid attention. Generative AI has changed that. Large language models need vast amounts of well-written, factual, current text, and news is one of the richest sources of it. As a result, licensing has moved from a back-office transaction to one of the central questions in the relationship between publishers and AI companies.

This post explains what news content licensing is, who pays whom, and how deals are typically structured, with a focus on AI developers who are considering licensing news content for their products.

What news content licensing is

A content license is a legal agreement in which a rights holder, usually a publisher, grants another party permission to use its content under defined terms. The license specifies what content is covered, how it can be used, for how long, in what territories, and at what price.

The key point is that a license grants specific rights, not blanket ownership. A publisher might permit a company to display headlines and summaries but not full text, or to use archived articles for model training but not to reproduce them in outputs. Everything that isn't granted remains with the publisher.

For AI developers, the uses that matter most generally fall into three categories:

• Training means using content to build or fine-tune a model. The content shapes the model's underlying capabilities but is not typically retrieved or shown to users directly.
• Grounding (often implemented through retrieval-augmented generation, or RAG) means pulling content into the model's context at the moment a user asks a question, so the answer reflects current, specific information. This is how AI search and assistant products deliver up-to-date answers.
• Display means showing content, excerpts, or attributed summaries to end users, often with links back to the source.

These rights are increasingly negotiated separately and priced differently. Training rights and grounding rights, in particular, reflect different kinds of value: One is a largely one-time contribution to a model, while the other is an ongoing, recurring use tied to the product’s day-to-day functioning.

Who pays, and who gets paid

In the AI context, the licensee, the AI developer or platform, pays the licensor, the publisher or rights holder. Publishers are compensated because their journalism has real costs to produce, and because using it without permission raises legal and commercial questions that are still being worked out in courts and legislatures in several countries.

The payment does not always flow directly between the two parties. There are several common arrangements:

Direct deals are negotiated one-to-one between an AI company and a publisher. These have attracted the most press attention, particularly deals between major AI developers and large media groups. They tend to involve large publishers with the leverage and legal resources to negotiate bespoke terms.

Intermediaries and aggregators hold licensing rights from many publishers and license that content onward to buyers. For an AI developer, this offers access to a broad range of sources through a single agreement, with rights already cleared and content delivered in a consistent format. For publishers, particularly smaller ones, it provides access to buyers they would struggle to reach alone. Newstex, which licenses content on behalf of its network of publishers, is one example of this model.

Marketplaces allow publishers to list content and set terms, and allow buyers to license what they need, sometimes on a per-use basis.

Collective licensing involves an organization negotiating on behalf of a group of rights holders, similar to how music rights have long been managed. This model is more established in some countries than others.

How Newstex can help

Newstex gives AI developers access to licensed news content at scale, drawn from more than 1,600 vetted publications in 52 countries and 12 languages. We process more than 1.5 million articles each month and deliver them in structured XML and JSON, with licensing details, unique article identifiers, and provenance metadata attached so every dataset can be traced back to its source. Developers can filter content by industry, geography, or language for training, fine-tuning, evaluation, or grounding. AI companies such as ProRata and AskNews already partner with Newstex to license content from publishers without having to negotiate hundreds of separate agreements. To learn how Newstex can supply rights-cleared news content for your AI products, explore our content licensing services for AI developers. 

How the deals work

Pricing structures vary widely. Common models include flat annual fees, per-article or per-use fees, fees based on volume of content accessed, and revenue-sharing arrangements tied to how the content performs in an AI product. Some agreements combine a fixed payment with a variable component. Direct deals can also include non-cash elements, such as technology credits, product collaboration, or commitments to attribution and referral traffic.

Terms are rarely public in full. Reported figures for individual deals should be treated with caution unless they come from the parties themselves or from court filings since many circulating numbers trace back to unnamed sources.One of the few figures disclosed by a party to a deal came from IAC, whose chief financial officer said on its Q3 2024 earnings call that OpenAI's license with Dotdash Meredith (now People Inc.) brings in roughly $16 million a year in fixed payments, with a variable component to be calculated later

Delivery matters as much as price. Licensed content usually arrives through a feed or API, often with structured metadata such as publication date, author, source, category, and update or correction history. For AI applications, that metadata is vital. It supports attribution, lets developers filter by recency or source, and establishes provenance.

Why licensing matters for AI developers

The case for licensing is not only about avoiding legal risk, though that is definitely a factor. Licensed content tends to be cleaner and better structured than scraped content, which reduces the engineering effort needed to process it. It can arrive in near real time which is essential for grounding. It comes with clear rights, which matters to enterprise customers who increasingly ask vendors how their models were built. And it supports a sustainable supply of the journalism that AI products depend on.

A checklist for AI developers looking to license news content

If you are preparing to license news content, these are the five questions worth answering before you sign:

• Do the rights match your use case? Decide whether you need content for training, grounding, display, or some combination, and make sure the agreement names each permitted use explicitly. Ambiguity about training versus grounding is a common source of later disputes.
• Does the licensor actually hold those rights? Particularly with intermediaries, ask how rights were obtained from the underlying publishers and whether those agreements cover AI uses. Review the source list for coverage and editorial quality as well as breadth.
• Will the content arrive in a form you can use? Check update frequency, delivery method, format, and metadata. Grounding use cases generally need low-latency delivery.
• What are your ongoing obligations? Understand attribution requirements, how corrections and takedown requests are handled, and what happens to content, and to models trained on it, when the term ends.
• Do the commercial terms hold up over time? Confirm that territory and duration fit your product, model costs under realistic usage growth, and keep records of what you licensed, from whom, and on what terms. You never know when you may need to vouch for something down the line.

‍