The rise of AI is uprooting news publishers’ understanding of content, how it’s created, and how it’s distributed.

During the INMA Agentic AI for News Media Master Class, hosted as part of the Product & Tech Initiative, media leaders from Associated Press, Schibsted, Scroll, and IAB Tech Lab shared how agentic AI is prompting publishers to reconsider how they are creating content — and how they are leveraging it on the Web.

News is data at Associate Press

AP Intelligence, a strategic initiative launched in late 2025, is a massive effort that turns Associated Press’s extensive archive and real‑time reporting into structured, machine‑readable data designed for a variety of industries. It’s not just a change in practice; it’s a shift in thinking.

“We are transforming how we think about our news as data and unlocking that new potential to ensure that our news is reaching new industries and new audiences,” Sara Trohanis, vice president of America revenue and head of strategic partnerships, said, explaining that it is “creating the world’s leading source of verified news data.”

AP Intelligence is fueled by the extensive amount of data points every story contains. Regardless of the size of the story, it contains “data points that resonate in a lot of different industry segments beyond who we’ve traditionally reached with this information.”

A story on wildfires, for example, includes location, time, casualty numbers, economic impact, weather conditions, and historical context. Traditionally, those details lived inside narrative journalism. But AI systems don’t read stories the way humans do, Trohanis said: “They ingest entities, timestamps, geolocations, and metadata relationships.”

If journalism could be structured for machines, she noted, it could unlock enormous value — not just for media companies, but for industries that rely on real‑time, verified information to make decisions.

AP Intelligence provides AP’s text, photos, video, audio, and archival content with metadata, structure, and context, making it relevant beyond its traditional uses.

“We are making it easy to integrate a factual understanding of current and historical events into any data stack, user flow, and business process,” Trohanis said. “And doing it in real time.”

Agentic AI and content creation at Schibsted

Aagentic AI is reshaping products at Schibsted. Juan Carlos Lopez Calvet, director of data and AI, shared a live demo of a new tool that aggregates content from within the brand’s asset library to produce video stories. That tool, Videofy, is currently in its 10th iteration.

“We have a lot of assets to ingest, you know,” Lopez Calvet said. “It can be articles, it can be texts, it can be videos, audio. Then we transform it, and then we make an output.” For the purpose of the demo, the output was a video.

Schibsted’s AI agents actively collate and aggregate stories with similarities across all assets. For example, a major sporting event about a particular athlete or rival teams at the Winter Olympics might draw several articles, photos, and videos. Agentic AI is automatically curating these assets in alignment with media organisation’s assets.

Between the resources that Schibsted already owns and the assets the AI agents collected, the development of stories begins.

“Once I get to start, I can choose the template,” Lopez Calvet said. “In this case, I think’s a VG template. I can choose the language. We have many languages.”

For the purposes of the demo, the selected output was a video format gathering dozens of articles about the Epstein case. The prompt read: “A short hook-driven video about the recent revelations of Crown Princesse Mette-Marit’s close contact with Jeffrey Epstein, the consequences for the Norwegian Royal Family, and the subsequent drop in public opinion.”

The video concept is generated with a pre-selection of audio and images from the asset library that comes via a content API. A human user can override the selected options, from the hook to the number of scenes to the voice narration. The agentic AI allows for an initial concept and assets, all from scratch, that are then editable and customisable. The final steps involve the file download and integration with a CMS.

Scroll leverages granular units of content

Today’s online experience, with the advancement of AI, has created an ecosystem where every single person has the ability to scrape the internet and create their own personal news digest.

“If we are entering that ecosystem as a news industry, what does it look like?” Sannuta Raghu, lead of AI Lab News & Journalism at Scroll in India, asked.

Raghu discussed three levers that can be utilised to navigate this transformation, one of which is focused on expanding the flexibility of content.

“We’ve figured out that you can do granular units of journalistic knowledge,” Raghu said. “You’re able to design your archive in a way where this on-demand transformation is much easier.”

This opens up a situation where a one-article container no longer contains just one story. The team can pull in the entire contextually relevant archive into this one container.

To do this, they built a tool called Factivo, which takes a verified news story and converts it into multiple formats such as video, timeline, calculator, mindmap, FAQs, and so on. Again, fidelity is an important component to source.

An example of this in use was the reporting on devastating floods. There were about 20 stories published in all by Scroll on the floods, and using Factivo, the team was able to generate a timeline of the event and all its related stories.

“So you have the granular units of atoms, you have the granular units of information, and you structure it in a certain way in your archive,” Raghu explained. “That creates the potential to really do fun things for the user on the outside.”

Combatting AI traffic with tokenised access

For decades, publishers built their businesses around a simple assumption: people visit Web sites. That assumption no longer holds.

Today, most traffic on news Web sites doesn’t come from humans — it comes from machines. AI agents, crawlers, and automated systems now read, interpret, and redistribute journalism at scale.

Shailley Singh, executive vice president of product and chief operating officer of IAB Tech Lab, said that it’s not just a technical footnote but a fundamental change in who the Web’s audience actually is.

“More than 51% of Web traffic is not human,” Singh said.

As AI agents proliferate, publishers need curated, frequently updated lists. Singh emphasised that reactive filtering alone is not enough: “In an environment where spoofing is possible, and new agents appear regularly, a more robust method of verification is required.”

One solution Singh discussed is tokenised access. Instead of guessing whether a request is from a known crawler, publishers can issue a digital token based on a licensing agreement.

“Once you have an agreement with an agent that you want to allow access to your content, you can issue them a token based on the licensing deal,” Singh said. “When their agent comes, it presents that token as a digital signature.”

If the token is valid, access is granted. If not, the request is blocked. Singh called this “a very foolproof way of managing who gets to see your content.”

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