Methodology

WorldBrief turns the world's headlines into a picture of the standing conflicts of world politics. This page explains how, step by step, and says plainly which parts are decided by a person, which by fixed rules, and which by AI.

From a headline to a conflict page

1. 01

   Collecting headlines

   Around the clock, WorldBrief reads the headlines of 247 news outlets from 73 countries, in 14 languages. It works from headlines and their links, not full articles. Every headline keeps its link, so you can always go back to the original report.

2. 02

   Placing them on the map

   Each headline is matched to the countries and regions it is about, using curated lists of names in each language. A headline can belong to several places at once, and is never forced into one.

3. 03

   Grouping them into events

   An AI model labels each headline: what kind of story it is and who is involved. Headlines that report the same development are then grouped into one event by fixed rules over those labels. An AI model checks whether two events are in fact the same story, and writes each event its title and summary.

4. 04

   Attaching events to conflicts

   WorldBrief follows a fixed list of conflicts and their fronts, each with its place and its parties. An event is attached to a front when it happens in that place and its headlines speak the vocabulary of that front. When an event fits two fronts of the same conflict, an AI model reads it and picks the one it belongs to.

5. 05

   Writing the conflict pages

   Every conflict page answers the same questions: where it stands, who is in it, how each side explains it and what that account leaves out, and what has changed its course. The pages are written with AI assistance against a fixed set of rules: each side's account in its strongest honest form, no verdicts in our own voice, contested names attributed. A person reads every page before it goes live.

6. 06

   Reading the press

   Outlet pages show what each outlet covers and how often, and an AI model scores the tone of its headlines toward the countries and people it writes about. Everything on the site is published in English and German; the German version is translated by AI.

Who decides what

The guiding rule is simple: structure is decided by people and fixed rules, and AI is used only where reading and writing are needed.

Why we show comparisons, not counts

How many headlines or events a conflict has depends on which outlets we read, how headlines are grouped and how models label them. Those numbers mean something only in comparison. So activity is shown as a comparison: whether a conflict is very active, active or quiet compared with the others, and whether it is rising.

Limitations

WorldBrief is made for orientation, not completeness. Like any large automated system, it has limits:

• it reads headlines, not full articles, so summaries can miss what only the body of a report says,
• it reflects the reach, priorities and blind spots of the outlets it reads; large and English-language outlets are overrepresented,
• the list of conflicts is an editorial choice; a new conflict appears only once it is added,
• an event can be attached to the wrong conflict, or missed by the right one,
• AI-written summaries and translations can contain errors,
• conflict pages are written at a point in time and can lag behind the latest events, which are listed beneath them as they happen.

The known issues page describes these in detail and what is being done about them.

How to use it

Last updated: October 2026