This AI Helps Local News Report Faster, Smarter

A city can change without anyone noticing. Not because something happens in secret, but because information arrives in a tidal wave of paperwork that nobody has time to read. And in the newsrooms of a community newspaper, the space between “public” and “understood” is where accountability can slip.

A case in point was iTromsø, a provincial newspaper in Northern Norway owned by the Polaris Media group. They learned about the problem the hard way. Planning documents, building files, and other official records contained leads for stories, if someone could dig them up. The newsroom answered the problem with a system named Djinn. Short for Data Journalism Interface for Newsgathering & Notifications, the system was developed by iTromsø with help of IBM and the Norwegian company Visito.

Djinn came from a very practical starting point. Journalists spent hours searching through building-case files, sometimes hundreds at once, mainly to find the few that seemed connected to real civic unrest. According to the Visito account, the original application even had a nickname, “Byggebot,” before being developed into a platform.

The scale of the challenge is easy to miss until you look at the numbers. Polaris rollout statistics show that the number of planning and building documents that Djinn downloads and processes each month from more than 130 Norwegian municipalities is about 12,000. That volume was impossible to handle manually in a newsroom.

Djinn uses a pipeline built around triage. It employs customized web scrapers that pull new documents from municipal archives. It converts the documents into machine-readable text, then applies analysis and summarization. This approach uses language models tailored to Norwegian for understanding, and cost-effective tools like Meta’s Llama for summarization.

The key part is how Djinn decides what makes the cut. Rather than some vague “magic” ranking tool, Djinn appears to combine journalist-trained scoring with search keywords, named actors (people, locations, developers), and anomalies in document patterns that can hint at bigger events. The ONA case study article also explains how the tool helps detect critical actors and hidden connections to generate story leads. Overall, the setup seems closer to an editor’s instincts than a simple search filter.

This matters because the tool isn’t trying to replace a reporter’s judgment. It’s trying to protect it. According to interviews conducted by WAN-IFRA, journalists who previously took two to three hours to search the archives may now spend just a few minutes before they start contacting sources. Another story says tasks that once took one hour may be cut to ten minutes.

In smaller markets, that time saved becomes a competitive advantage. When a newsroom can spot the right document faster, they can publish sooner. They can ask sharper questions and spend more time on shoe-leather reporting instead of research.

Djinn’s impact has been big enough to scale across the company. The service launched as a project at iTromsø has been extended to around 35 Polaris newspapers. The Newsroom Robots interviews also indicate that roughly 36 papers in Norway have adopted the service since its start in 2023. Small differences in the count may reflect different adoption timelines.

Even IBM’s internal metrics highlight major research-time savings and meaningful increases in traffic share for the published titles. Those figures reflect partner-reported results but reinforce the larger point: this was a change in newsroom habits, not a gimmick.

Why Djinn matters, even though AI in the media isn’t the tech itself. It’s the thinking behind it. The tool sets a narrow, civic goal: highlight the most public-interest documents before they disappear in the pile. Put another way, the aim isn’t to outsource journalism to a model, but to add a layer of intelligence that supports journalism. Local democracy fails not only when information is suppressed, but also when information is technically available yet functionally unreachable.

Djinn’s real value is shrinking that gap, turning municipal paperwork into a prioritized trail of leads a human journalist can follow. And that’s what sustains the watchdog role: timely insight into the decisions that shape a community.