In-house AI tool at Amedia used by 51% of journalists

By Paula Felps

INMA

USA

When Amedia, Norway’s largest local news publisher, started looking at how to integrate generative AI into its journalism, it had a clear goal: make journalists more efficient and improve journalism.

During this week’s Webinar, INMA members heard from Mathias Høibakk Bergquist, head of editorial AI at Amedia, about the company’s AI journey.

It began in 2015, he said, by exploring possibilities along five specific tracks: language technology, personalisation, insights, format, and AI platforms. But all those tools and applications went into action after the stories were published: “It was analysation, not authorship,” he explained.

The entire landscape changed in November 2022, when generative AI burst onto the scene and “changed the mood overnight.” This new technology, Bergquist said, “could finally sit inside the writing process, not just comment on it later.”

And that began a journey to discover how to create something useful that would be easy for journalists to use and could fit into their existing systems.

The AI Hub at Amedia mapped out the possible uses for generative AI in its newsrooms.
The AI Hub at Amedia mapped out the possible uses for generative AI in its newsrooms.

Entering the Sandbox

In February 2024, Amedia launched its editorial AI hub. The multidisciplinary team included representatives from eight newsrooms across the countries and then added data scientists and developers who “tested every wild idea in real time.”

That mix of people and perspectives was crucial, Bergquist said, as “it kept us honest.”

Starting with pain points that were plotted on a whiteboard, the team mapped out the entire editorial journey from brainstorming through researching, interviewing, writing, fact-checking, and distributing.

“We stuck colour dots on every pain point and counted more than 30,” he said.

Then, the team used GenAI to see if it was a true pain point that “really hurt in the editorial process” and used the technology to create a solution. If a new feature couldn’t save a journalist 10 minutes, it was scrapped.

As the team experiemented with GenAI, it became apparent they needed a more secure working environment. Not only were there privacy considerations with using OpenAI, but cost was a factor as well.

That led to the creation of its AI Sandbox, which launched at the end of February 2024.

The secure, in-house platform is designed for journalists and is what Bergquist likens to a “Swiss Army knife,” featuring pre-programmed prompts like “improve my text,” SEO summaries, Facebook blurbs, and headline suggestions. Journalists can also upload PDFs or spreadsheets and even paste in URLs for quick content generation.

The interface is simple by design, and every element is built to save time.

“In 13 seconds, I get three distinctly different Facebook suggestions. And they are, let’s be honest, way better than any reporter would do,” said Bergquist.

The AI Sandbox provided a secure environment for experimentation and testing functionality.
The AI Sandbox provided a secure environment for experimentation and testing functionality.

Fostering adoption

As the idiom goes, “If you build it, they will come” — but Bergquist admitted there were concerns about getting journalists to embrace the AI Sandbox.

At first, adoption was slow. The Sandbox launched with only 150 active weekly users, which represents just 10% of the editorial workforce. A spike came in mid-summer, thanks to curious interns. But after they returned to school, usage dipped.

“We worked really hard through last fall to get to where we are right now, with 700 weekly users in our AI Sandbox,” Bergquist said. “That’s 51% of the total editorial workforce.”

While the tools are powerful and impressive, that isn’t what convinced the newsroom to give GenAI a try, he said: “What we did was to solve real newsroom irritation first and then brag about AI later. And that mindset beats any shiny interface.”

The process started by reaching out to everyone, finding out what they wanted to achieve — and then offering an AI tool to help them accomplish that. In October and November of last year, Amedia held 13 live sessions, from one-on-one desk visits to a meeting of 250 people.

“More than 500 journalists touched the Sandbox on their own laptops with their own stories just in two months,” Bergquist said.

Every question they posed was added to a spreadsheet and helped determine which problems were significant enough to warrant the creation of a tool.

“Reach out and talk to people,” he advised. “And when you do talk to them, you need to show, don’t tell. We had no glossy e-learning portal. We did physical hands-on training with real-world cases.”

For example, reporters were asked to bring large documents they hadn’t had the time to read through and discover how fast AI could do the job for them. In one case, it dived into a 100-page report about the school structure in a certain municipality. Whilst the local newspaper covering the story had been doing great work, Bergquist said, they uncovered more stories just by allowing AI to chat with the document.

Once reporters see the value of the tools, they’ll be more willing to use them.

Despite initial reluctance, today 51% of Amedia journalists use the AI Sandbox every week.
Despite initial reluctance, today 51% of Amedia journalists use the AI Sandbox every week.

The AI Sandbox in use

Bergquist offered compelling examples of how GenAI is streamlining and simplifying newsroom processes. In one case, a veteran reporter was tasked with transcribing the Constitution Day parade timetable.

“Normally this means hours of squinting at awful PDFs, but he thought, ‘I just need to drag this file into the AI Sandbox and hit extract and rewrite.’ And then 30 seconds later, he had clean text ready to publish,” Bergquist said.

“This is not a Pulitzer moment exactly, but overnight, he and every reporter around him understood that this was a tool that could save him from personal parade timetable hell.”

In another example, an experienced Iraqi journalist was struggling with the Norwegian syntax and used the “improve my text” button in the AI Sandbox. This, Bergquist noted, allowed her to focus on story angles while the AI fixed the grammar.

“So this is about empowerment, not copyediting, and instantly broadened the newsroom’s collective voice.”

Finally, he shared the story of Natalie Krisetensen, a journalist from Moss Avis who wanted to do something with the data released by the Norwegian government for Tax List Day, which cites all the tax details for Norwegians. The document is a treasure trove for local reporters, and she asked if they could use AI to mine the information when it was released.

“The road map said absolutely no, but my journalistic heart said, ‘fuck, yeah.’ So we did it,” Bergquist explained.

All 55 million tax records were entered into the database, and now reporters can access information such as how many millionaires the country has had in the last decade or can chart millionaire income growth. One newspaper, for example, found interview subjects for a story about women earning a medium income. Because of the AI assist, Bergquist said, “they picked up the phone and turned data into human stories.”

This shows the perfect blend of AI and the human touch, he said: “This is AI as a source machine, not a black box, but it came because we listened to a reporter having an idea on how to use AI.”

About Paula Felps

By continuing to browse or by clicking “ACCEPT,” you agree to the storing of cookies on your device to enhance your site experience. To learn more about how we use cookies, please see our privacy policy.
x

I ACCEPT