How iTromsø Built Norway's Most AI-Ready Newsroom

Tromsø is small city in northern Norway and iTromsø is a small news outlet with just 25 total staff. According to their LinkedIn info it has about 48,000 daily readers in print and online. But it operates what may be the most sophisticated AI infrastructure in Nordic media. At the ONA conference in New Orleans last week, I got a chance to talk to Lars Adrian Giske, Head of AI at iTromsø, about the AI strategy at a small newsroom that is punching way above its weight.

Giske advocates a radical departure from typical newsroom AI adoption. “Find good hands-on projects. It doesn't have to be large or deep. It could be pretty much anything. Start small, find a project that you believe in, and just do it," Giske told me. He sees AI as a "creativity unlocker" that enables journalists to build their own tools and products.

At iTromsø such a small project was a real-time local real estate bot. But that was just a start. iTromsø’s AI implementation now goes way beyond little “dip your toes in” starter projects. The newsroom spent two and a half years consulting lawyers before launching a centralized data platform that aggregates personal-level information about residents, including income, property ownership, and corporate connections.  

Sponsored

Join over 4 million Americans who start their day with 1440 – your daily digest for unbiased, fact-centric news. From politics to sports, we cover it all by analyzing over 100 sources. Our concise, 5-minute read lands in your inbox each morning at no cost. Experience news without the noise; let 1440 help you make up your own mind. Sign up now and invite your friends and family to be part of the informed.

Subscribe to 1440 today.

This infrastructure enables discoveries impossible through traditional reporting. "If my end goal is having a system that is capable of saying that director A of company B had to sell their house because company C went bankrupt, that kind of news notification is incredibly powerful for a journalist," Giske told Nikita Roy in the Newsroom Robots podcast in March 2025.

iTromsø's breakthrough came with DJINN, an AI tool that scans municipal documents for newsworthy content. The system processes over 12,000 PDF documents monthly, ranking them by relevance and generating summaries that help journalists identify stories worth pursuing.

With tremendous success: DJINN now operates across 40 newsrooms in the Polaris Media group, covering over 130 municipalities. Polaris Media ASA, based in Trondheim, is one of Norway's largest media companies, owning 65 local and regional newspapers across Norway and Sweden.

Newsrooms using DJINN saw a 1,300% increase in traffic share while reducing journalist research time by 94% (IBM Think, May 2025). More importantly, journalists adopted the tool enthusiastically. It is intuitive to use and built for real use cases in the newsroom. 

iTromsø's approach to AI training differs markedly from conventional newsroom workshops. Instead of abstract lessons, journalists learn through live projects. The newsroom's ongoing school district redistricting coverage serves as a hands-on training ground where all 25 staff members use AI tools for real reporting tasks.

"We don't have the time, and journalists don't have the time in the daily work to use half a day for training," explained Rune Ytreberg, head of iTromsø's data journalism lab in the Newsroom Robots podcast: "They have to train while they are working.” 

The newsroom's five-person data unit includes developers who sit directly in the newsroom, participating in daily editorial meetings. This integration means that technical solutions emerge organically from newsroom needs rather than top-down technology mandates.

iTromsø pushes beyond simple document processing into sophisticated investigative applications. The newsroom used retrieval-augmented generation (RAG) systems for a hospital understaffing investigation that won a Data-SKUP award, one of Norway's most prestigious journalism honors.

"We used the RAG to do what we call smelling the data, and then we narrow it down as we go," Giske told Nieman Lab. The investigation uncovered that "a doctor from Denmark, who was working remotely, spent four seconds reviewing X-ray images." The project would have required "at least three months of manual research time" using traditional methods.

iTromsø recently adopted an open-source RAG platform called Kotaemon, designed specifically for advanced, industrial-scale document analysis with robust, accurate citation and source-highlighting capabilities. Kotaemon stands out because it enables complete local hosting and custom model selection, which is especially valuable for processing and safeguarding sensitive municipal data—a top requirement for organizations handling confidential government or civic records.

Moreover, the infrastructure reflects hard-learned lessons about AI implementation costs and capabilities. The newsroom initially tested OpenAI's enterprise solution but found it prohibitively expensive for organization-wide deployment. This cost barrier created internal divisions between journalists with AI access and those without.

The solution involved adopting Open-source models and self-hosted platforms like Kotaemon which reduce per-query costs while enabling universal access. 

iTromsø's AI capabilities extend beyond municipal document analysis to comprehensive community impact investigations. The newsroom uses its data infrastructure to examine how external forces reshape local housing markets, combining multiple data sources to reveal stories that affect residents' daily lives.

Giske described to me their approach to investigating AirBnB's impact on Tromsø's housing market: "We're data first. Our strategy is: We have an idea, a hypothesis, and then we ask, what kind of data do we need to prove it?" With this approach, the team discovered a dramatic growth in short-term rentals. The anecdotal evidence that AirBnBs in Tromsø were displacing long-term rental apartments for locals was real. The data showed "8,000% growth in the Airbnb market over two years”, according to Giske. 

These kinds of stories, based on automated data retrieval processes, scale across markets, enable hyperlocal focus on certain neighborhoods as well as follow-ups and story variations without requiring much extra effort. "We could do the story again in a year and see how the trend has developed. Therefore, it’s important to keep those data flows coming in," Giske explained. "One data source tells one story. But if you contextualize it and enrich it with other data, you get a totally different picture and you're able to tell totally different stories, like how does income affect it?" Giske noted, describing how the approach enables neighborhood-level analysis showing precisely how tourism trends impact different community and society segments.

Giske frames media's future challenges in stark terms: "Our competition is not within the industry. It is the big tech companies. They took control of our information stream, now the big AI companies are doing the same thing, they’re taking control of the information interface.” 

Giske predicts significant changes in how audiences consume news. "News media is about to change. The article as we know it may not be the preferred format of readers or listeners or viewers in the years to come. People are getting used to generative ecosystems, and that won't change”, he told Nieman Lab. This evolution requires newsrooms to think beyond traditional article formats toward data-driven experiences. "I think our role will not as much be providing a platform like a website, but it will be providing carefully curated and processed data," Giske said in the Newsroom Robots podcast. He describes scenarios where residents could ask specific questions while making dinner: "What's happening now with my school? My children, are eight and ten. What's happening? And you will get your answer." This conversational interface would provide personalized, accurate responses based on comprehensive local data.

According to iTromsø’s head of AI, deeply focusing on creating value for audiences will be essential. iTromsø's response involves building community connection through hyperlocal AI capabilities. The newsroom envisions conversational interfaces where residents can ask specific questions about their neighborhood, schools, or municipal services and receive accurate, personalized responses based on comprehensive local data.

At the same time, he thinks “it is extremely important that we use enough time to go out and get in contact with our communities and ask them, what do you need? How can we solve your problems?"

Beyond traffic metrics, iTromsø measures AI success through journalistic impact. The hospital investigation enabled deeper reporting that served the public interest. Similarly, the newsroom's municipal finance tracking system helps residents understand budget decisions that directly affect their lives.

"I think that before June or before summer, I think we're going to have the most AI-competent newsroom in Norway," Ytreberg predicted in March 2025 on the Newsroom Robots podcast. Judging by iTromsø’s standing among international AI leaders in news organizations, they have reached that goal.