How we work at ExtraFokus

Last summer we tested the municipalities’ shared chatbot in 41 municipalities. We asked the same kind of questions a citizen would ask. About a decision, a complaint, a deadline. In several cases the chatbot sent the citizen on to the election office. The failure belonged to the platform, not to any single municipality.

Stories like that are not found by waiting for a tip. They are found by testing yourself, systematically, and documenting every step so others can check it. That is what we call OSINT (open source intelligence), and it is the backbone of everything we do.

Open source does not mean easily accessible

A source is open when it was created for a public audience or is lawfully accessible, free or for a fee, such as a freedom of information response or an academic article. That covers public registers, official documents, social media, satellite images and archived versions of pages that have since been changed or deleted.

It excludes leaked databases, hacked systems and private accounts. We have seen other outlets balance on that line. We do not. If a piece of information can only be obtained by breaking the law or invading someone’s privacy, we do not use it. No matter how good a story it would make.

How we have used the method in practice

The Muni chatbot case: We tested the system across municipalities rather than basing the story on one citizen’s experience. That made the difference between “a municipality has an error” and “the platform fails systematically”. The Parliamentary Ombudsman subsequently opened a formal investigation and cited ExtraFokus directly in the case against seven municipalities and the ministry.

The shadow fleet series: We cross-checked oil tanker movements in European waters against the Paris MoU’s Thetis database and compare them with national port-state inspection practice. It is slow work (the database requires precise filter combinations to give accurate figures for individual ports), but it gives us our own data, not a retelling of other people’s.

The DUP series on the police complaints system: Built on freedom of information requests to the National Police and a systematic comparison of complaint cases over time, not individual cases taken out of context.

Verification, step by step

Before anything is published, it is cross-checked:

• Images and video are reviewed for metadata and recognisable landscape features, and compared with satellite images where necessary.
• Social media accounts are assessed on language use and posting patterns if they serve as a source for a claim.
• Company networks are mapped via the CVR register and similar public registers.
• Everything is archived with screenshots and timestamps, so a trail does not disappear if a page is changed or deleted afterwards.

Larger investigations take weeks, sometimes months, because we do not publish until several sources independently confirm the same picture.

Anonymity protects sources, not ourselves

We follow the Danish Union of Journalists’ press ethics guidelines: no access to anything other than what is publicly available. That is a line we do not negotiate, no matter how good a story could become by crossing it.

When a source risks losing their job or facing reprisals by coming forward, we use VPNs and separate accounts to shield the trail back to the source. That protects the person who speaks; it is not a way of hiding ourselves.

Most of our section journalists write under pseudonyms, while Kenn Hawk appears under his full name as editor-in-chief responsible for all content on the site. It is an editorial model, not a hidden identity: responsibility sits in one place, and it is public.

Why it works faster

Traditional research can take months, because it often waits for someone to call back. OSINT works in parallel with what is happening: a satellite update, a new document in a register, a change on a website can be followed while the story develops, not only afterwards.

Bellingcat has shown the strength of the method at scale by verifying war crimes through satellite images. We use the same basic logic on a smaller scale: documenting the trail while it still exists, rather than reconstructing it from memory and loose quotes afterwards.

The limits of the method

OSINT is not infallible. Large volumes of digital material also mean large volumes of noise and outright false information, and it takes time to separate them. When we are in doubt about a source’s reliability, we say so in the article instead of smoothing it over.

If you have questions about how we arrived at something specific in an article, you are welcome to write to us.

ExtraFokus.eu: A deeper look.