About the Project

Every day, dozens of leading newspapers around the world publish hundreds of opinion and analysis pieces about the very same geopolitical events. No one can read them all - each of us reads one angle or two, usually the ones we already know, and misses the fuller conversation taking place between sources that see the same event through different lenses.

geopolitics-tracker tries to solve a focused problem: not to replace human reading, but to give it a tool. A system that follows leading newspapers in English and German, across a wide political range from left to right, identifies the geopolitical commentary among them, and builds a daily picture from it: which topics drew attention, and how each source chose to frame them. The goal isn't to decide "who's right" - but to make the full range of opinion accessible, not just the corner the reader already knows.

From a Daily Snapshot to a Trend

The immediate product is a daily report: a document that groups that day's commentary by real topic - not by mechanical tagging - and sets out a concise comparison between sources for each one. But a single daily report, however accurate, is only a snapshot. The real value only emerges over time: once weeks and months of reports accumulate, you can ask questions no single newspaper can answer - is a position being reshaped? Does a topic that disappeared suddenly return? That's why every daily report is built from the start to double as an archive entry - a building block for a wider view.

This past week, that wider view turned from an idea into a product. The daily report now runs fully automatically, with health checks and automatic stop on repeated failures. Alongside it came a new "Trends" section, with two components: a narrative trend report, published every two weeks, tracking how coverage of a topic shifts over the period - built from the daily reports themselves, and written like them in all three languages; and a graphical view of coverage by country, still at an early (Beta) stage.

The Method: Three Roles, Not Two

Alongside its substantive goal, the project has a second role: a place to actually learn how to build a complex system with advanced AI tools. The central insight: the simplistic model "person asks, machine executes" isn't accurate. Three parties are at work: a conversation-and-planning tool, which thinks through the problem before a line of code is written and translates a vague decision ("I want the system to detect trends") into a concrete task; a coding agent, which operates inside the development environment - writes, runs, reports back; and above both, the human operator, who watches both tools at once and, above all, holds the line and checks the work. Again and again, this was the point where mistakes got caught - a result that looked "almost good" but hadn't actually been verified. Every report's comparison content is produced using a language model (Anthropic's Claude API), by exactly the same method: human review before every publish, and a standing safety net in the code for any failure that repeats.

The Sources

The project started with four sources - two politically balanced pairs: Guardian vs. Telegraph in English, Süddeutsche Zeitung vs. Die Welt in German. After a few days, a clear gap emerged: the American arena wasn't represented at all. Four American papers were chosen to span a wide political range (The Washington Post, Wall Street Journal, Los Angeles Times, USA Today), alongside two weeklies (The Economist, Der Spiegel) that bring depth. Three more sources were added later (Frankfurter Allgemeine Zeitung, Neue Zürcher Zeitung, The Globe and Mail), bringing German-conservative, Swiss-liberal and Canadian perspectives - and reducing the US's relative weight in the mix. Today the system follows 12 sources across two languages, spanning a wide political range in every arena.

The Framework: Three Axes of Perception

Geopolitics is a combination of spatial, temporal, and textual perception. The homepage is built around three equal-standing "lenses" - an interactive map (including active conflict zones: Israel-Palestine, Iran-the West, Russia-Ukraine), a timeline, and a textual entry point - along with the ability to browse reports by topic, region, or date range.

The Third Language: German

The site was first published in Hebrew and English only. The expansion to German happened in two stages: every existing report in the archive was rewritten into German - not translated word-for-word - to keep it consistent; from that point on, every new report is written in German as original text, alongside Hebrew and English, from the same source articles - not as a translation added afterward.

Limits of the Project

A critical reader deserves to know a few boundaries. All the sources are Western quality press - not a representative sample of world opinion. Volume of publication isn't a measure of importance: a paper that runs many opinion pieces can look dominant without that reflecting real significance; the sorting mechanism by source-breadth only partly offsets this. There's no fact-checking layer - the system summarizes what each source claims, not whether it's true. Every daily report is written as a standalone unit; tracking how a topic develops has so far required manually reading several reports - which is what the biweekly report is meant to ease. The three language versions are written independently of each other, not as translations - so small differences in wording between them are expected.

Looking Ahead

The daily report, the biweekly report, and the graphical trends view all exist now - but not at the same level of maturity. The graphical view is the least mature piece: it shows raw numbers without explaining their meaning well enough, and it's due for a redesign. Beyond that, two questions remain open: a clearer experience for a reader unfamiliar with the project, and audience - whether to expand beyond researchers and policy people, for instance to a version accessible to teenagers.