Since 2023, AI developers have signed a growing number of agreements with news publishers and online platforms. These deals sit at the center of one of the most important questions in AI: who gets paid for the data that models learn from?

What the deals usually include

  • Archive access for training or fine-tuning.
  • Real-time content so assistants can answer questions about current events.
  • Attribution and links back to the publisher in AI answers.
  • Technology access for the publisher’s own products.

Early examples include the agreement between the Associated Press and OpenAI in July 2023 and OpenAI’s partnership with Axel Springer in December 2023. Many other publishers and platforms have followed with deals of their own.

Why AI companies sign them

  1. Legal risk. Several publishers have sued AI companies over the use of their content, most prominently The New York Times in December 2023. Licensing reduces exposure.
  2. Freshness. Assistants need reliable, current information.
  3. Quality. Professionally edited text is valuable training data.

Why publishers sign them

New revenue at a time of declining advertising, and a say in how their journalism appears inside AI products. Critics in the industry worry that deals undervalue archives and may accelerate the loss of direct traffic to news sites.

The bigger picture

Courts are still deciding how copyright law applies to training. Until there is clearer law, licensing deals and lawsuits will continue side by side, and their terms — which are rarely published in full — will shape the economics of both journalism and AI.

Examples of the deal types

Since 2023, many large publishers and platforms have announced agreements with AI companies — including news agencies, newspaper groups, magazine publishers, stock image libraries and online communities. Reported terms vary, but common elements include:

  • access to archives for training;
  • real-time access to new content so AI assistants can answer questions about current events with links;
  • attribution and links back to the publisher in AI answers;
  • technology cooperation, such as using the AI company’s tools in newsrooms;
  • payments that may be fixed fees, usage-based or a combination.

Financial terms are usually confidential, which makes it difficult to judge whether deals are fair.

The alternative: litigation

Not all publishers have chosen deals. Some, including The New York Times, have sued AI companies for copyright infringement over training and outputs. Outcomes of these lawsuits will influence the value of future licensing deals: if courts find training on copyrighted content without permission is lawful as fair use, publishers’ bargaining power falls; if not, licensing becomes essential. See our explainer on AI copyright and training data lawsuits.

What it means for readers

For readers, licensing deals can mean AI assistants that answer news questions with links to reputable sources, rather than unattributed summaries. They may also influence which publications AI assistants cite most often — so it’s worth noticing where answers come from and checking original articles for important topics.

What creators can do

Independent writers, photographers and artists rarely negotiate deals individually. Options include using platform settings and robots.txt to signal preferences about AI crawling, joining collective licensing organizations or marketplaces that negotiate on behalf of many creators, registering works where useful, and following developments in copyright law and transparency rules that could give creators more information and leverage. Unions and professional associations are also negotiating AI terms in some creative sectors.

What to watch next

  • Court rulings on fair use and AI training.
  • Collective licensing schemes for smaller publishers and creators.
  • Traffic and revenue that AI answers actually send to publishers.
  • Regulation requiring transparency about training data.

Key takeaways

  • Licensing deals give AI companies content access and give publishers payment and attribution.
  • Deals often cover archives, real-time content, links and technology cooperation.
  • Terms are usually confidential, making fairness hard to assess.
  • Some publishers sue instead; court outcomes will shape deal values.

Frequently asked questions

Why do AI companies pay for content they could scrape?

To reduce legal risk, get high-quality and current content, secure ongoing access and build relationships with publishers.

Do small publishers get deals?

Most announced deals involve large organizations. Collective licensing and marketplaces aim to include smaller creators.

Many deals include attribution and links, but how prominently they appear varies by product.