AP makes its archive AI-ready to tap the enterprise RAG boom

The Economist and The Financial Times have already clocked the revenue opportunity presented by the boom in enterprise retrieval-augmented generation (RAG) demand, and have been licensing their content archives accordingly. Meanwhile, Dow Jones’ Factiva unit — which has a network of 30,000 publishers — now operates an AI-licensed content marketplace for enterprises.

It’s a strategy that should secure AP’s future as an information data repository for the AI era, and widen its customer base (historically news organizations) to include more enterprise clients by meeting their AI needs, AP global chief revenue officer Kristin Heitmann told Digiday.

Heitmann said AP intelligence offering is aimed primarily at enterprises developing their own LLMs or proprietary AI systems that require structured, verified data. Earlier this year AP launched on the Snowflake Marketplace – a cloud-based exchange where companies like AP can license structured, rights-cleared data directly to enterprises building AI systems.

Since doing so, AP has licensed its content to Snowflake’s finance clients, many of whom are building internal AI tools and need access to trustworthy, rights-cleared information, added Heitmann. 

“This offering really covers not just finance companies, but companies looking to monitor their supply chain, looking at crisis and operations management, environmental and regulatory awareness. I feel like there are almost unlimited use cases,” she said. 

Turning the AP’s entire archive into AI-readable fuel is a quiet moonshot: it means five decades of verified reporting are now cleanly digitized, tagged, rights-cleared and served via APIs so enterprise RAG systems can trust, cite and pay for it.

There is scant publicly available data to show evidence of the scale of the demand from the enterprise sector, but to publishers it’s clear there is an enterprise RAG boom that’s opening up a new revenue lane for them.

A Deloitte report published last week showed a large share of big companies aren’t just using off-the-shelf AI tools (like ChatGPT or Gemini) – they’re actively building or customizing their own LLM models and RAG systems. The AI ROI study of nearly 2,000 large private-sector companies across 14 countries in Europe and the Middle East found that most businesses are mixing internal and external AI tools. About 38% take a hybrid approach, combining in-house systems with third-party tools, while 32% rely mainly on vendor-built solutions for speed and scale and 24% plan to focus on developing their own in-house AI capabilities, per the study.

With 230 locations in nearly 100 countries, AP produces 5,000 individual pieces of distinct content every day. “The value of that eyewitness journalism in today’s AI information age is more important, and frankly more valuable, than ever before,” said Heitmann. “Because the AP has produced content in multiple languages, formats – text, photos, video and audio – and has the relevant metadata associated with it, means that our ability to succeed in the space is very significant and that’s why we’re investing here to ensure that we can thrive and survive in the new AI information age.” 

AP began its AI licensing journey in 2023 when it struck a licensing deal with OpenAI. It signed an agreement with Google to license AP content to Gemini at the start of 2025. And this week it revealed it’s also one of a handful of Microsoft’s publisher partners of the tech company’s AI content marketplace, along with People Inc. and USA Today Co. (rebranded from Gannett this week).

Heitmann said the content marketplace is still at the experimental phase but that it’s critical that the AP has a “seat at the table” in its development. Her comments are in sync with People Inc.’s chief innovation officer Jon Roberts, who has said that it will help Microsoft build from the “ground floor.’

Speaking at The IAB Tech Lab’s International Summit in London yesterday, People Inc.’s Roberts likened the current scenario where AI crawlers steal creator content without permission to the ad tech tax typically taken by vendors in the digital advertising supply chain. Except in this case, AI companies that scrape content without providing remuneration to publishers are like DSPs taking a “100 percent take rate” instead of the “rational” 5 percent that a DSP should take.

Roberts said that in 2024, investors and technologists believed AI would make publishers obsolete — that machines could absorb the world’s information and churn out content without ever needing journalists. That illusion cracked in 2025, when China’s DeepSeek model stunned the U.S. by matching leading systems at a fraction of the cost, and the open-source community replicated it within hours. It became clear that building powerful AI required reliable, high-quality data, he stressed. In that light, the value of verified journalism looks far less replaceable than Silicon Valley once assumed.

“The best AI has the best inputs and generates the best outputs. This is the inverse of garbage in, garbage out,” said Roberts at the summit. “This means that in the future, we need more good quality, high quality content, not less. We’re going to transform the entire multi-trillion-dollar global economy based on AI, and AI fundamentally needs great information to be great. Now this is interesting because it actually puts the incentives of the AI companies and the publishers on the same side of the table, whereas, I think, for the last couple of years…these were two separate groups standing on opposite ends of the table.”