The same story, told by newspapers in 53 countries, side by side. Sources and method in plain view →
World coverage in focus
The news of the last 24 hours told by the most newspapers, politics first, with each one's headline side by side.
How these stories are grouped
Automatic grouping by uncommon words the headlines share, with no artificial intelligence. It may wrongly join or split some stories, and a story told in Portuguese and in English may still appear separately, despite a fixed list of equivalents between the two languages. Leading stories may include a summary written automatically with the help of artificial intelligence, only from the headlines, with the source of each sentence. The first 3 stories include at most one per leader, so the top is not all about the same person. Sports stories, and celebrity stories that are not about politics, are left out.
Political Press Observatory
Now
How the press tells the politics between countries, right now.
Open the ObservatoryLoading the map…
Mostrar detalhesOcultar detalhes
Each dot is a country; its size is the number of political headlines naming it in the last 24 hours, in newspapers of other countriesCountries named in the same headline: intensity of the relationship in the press (being named together is not friction)Dashed: a leader and the country named in the same headlineCounts only, no tone, until 7 December 2026
<50 %≥90 % same votes at the UN as the chosen countryFormal ally (dashed: suspended)No dataExternal data, no tone
How to read this map
- Countries: each dot is a country. The bigger the dot, the more political headlines named it in the last 24 hours in newspapers from other countries. Newspapers talk a lot about their own government, so headlines from the country itself do not count towards the size.
- Leaders: the same, but only with headlines that name a leader from a fixed list (heads of state and government, foreign ministers and a few other figures). The most named leader appears next to the dot.
- Lines: they link countries, or leaders, named in the same headline. A thicker line means more headlines together, not conflict or agreement.
- Alliances: choose a country to see, in colour, how often it voted like each of the others in the UN General Assembly, and with an outline its allies by mutual defence treaty. This layer does not come from the headlines.
- No tone until 7 December 2026: the map only counts who is named. Measuring whether the press tells a relationship with more friction or more agreement only appears after that day, and only if the published tests pass.
- The lists of countries, political terms, leaders and treaties are sealed and published.
Coverage observatory
Choose a region to see the newspapers we follow and recent headlines, each linking to its source.
Tone measured (closed until the end of the first analysis period)Headlines only (tone not measured yet)Selected regionNo sources followed yet
Search the headlines
Searches every headline since 11 October 2026, across every newspaper we follow: the original headline, its translation and the story summary, with or without accents. It finds words that start with what you type. You can then filter by topic, country and period.
Collection status
The lexicon is sealed. Tone values remain closed until 7 December 2026 at 02:00 UTC. These numbers describe collection, not tension.
Sources in view. A verifiable method.
The method comes before the numbers. Everything that goes into the measure is described here, and each version is sealed with a SHA-256 fingerprint.
Read the method and public documents →
Sources, coverage and presentation details
Sources
Extended coverage
Region
For newspapers, the country of origin. For Wikipedia, the language edition. Each newspaper is also measured separately, to show whether one is consistently more charged than the others.
Stories across newspapers
Two headlines count as the same story when they share at least two uncommon words (names, places, numbers) and the weight of those words is similar enough. To join the Portuguese and English versions, a fixed list of equivalents counts words such as papa and pope, or Arábia Saudita and Saudi Arabia, as the same. Subjects every country has (minimum wage, budget, pensions) only join newspapers from the same country: across countries other words must be shared. Words from company results (profit, quarter, revenue) are not enough on their own either: another word must be shared, usually the company’s name. The rule is fixed and public, uses no artificial intelligence and goes into no measure: it only arranges the headlines. In comparisons, words used by only one newspaper among the headlines in the same language are highlighted (same word cleaning, no AI). Political stories come first: a story is political when at least a third of its headlines name one of the political leaders on the list (the Pope does not count) or contain a word from a fixed list about governments, parliaments, elections, diplomacy, war and conflict (src/temas.js); within each group, the stories told by the most newspapers come first. The first 3 stories include at most one per leader (the leader whose face goes on the card); a story that would repeat a leader moves down to just after them, so the top is not all about the same person. Sports stories are left out of the featured stories, even when they name a leader: a story is left out when at least half of its headlines contain a word or name from a fixed list (football, NBA, Formula 1, for example; src/temas.js). Celebrity stories (Hollywood, singer, for example) are left out too, unless the headline is about politics: it names a leader or has a word from the politics list. Those headlines remain in the archive and on the map. Since 1 November 2026, headlines from Portuguese and Brazilian newspapers also get a second opinion from an artificial intelligence model that only classifies (Jev, by TypeSafe AI; it does not write text): when it gives at least an 80% probability of sport or celebrities, the headline is left out as if the rule had caught it; in the Brazil edition, when it gives at least 80% of Brazilian politics, the headline gets in even without the words on the list. The fixed rules remain the basis, and the model never goes into the measured number. The top 30 stories have a 2 to 4 sentence summary written automatically with the help of artificial intelligence, only from the headlines, with sealed instructions (sinteses/instrucoes-historias-v2.md), the source of each sentence and a transparency notice in the footer. The summary may point out where newspapers stress different things, but does not say who is right. It is redone when the story gains new newspapers and never goes into the measured number. The Portugal button shows only the Portuguese newspapers: two telling the same story are enough, at least half of the headlines must name a place, an institution, a party or a person in Portuguese politics, and only politics and defence stories stay, with the same rule for sport and celebrities (src/portugal.js).
Measure
A public lexicon of tension and calm words, in Portuguese and English, applied the same way every day and only to each headline, for every newspaper. It measures the tone of the text, not the leaning of whoever wrote it.
No artificial intelligence in the number
No language model goes into the measure. Each region's summary is written automatically with the help of artificial intelligence, from the headlines only, citing the source of each sentence and marked «AI». The instructions the model receives are published and sealed (SHA-256 7954ec075da5… for the region summaries, a58fb553151f… for the story summaries).
Everything in view
Every region and every hour, including the quiet ones. Where there are few newspapers we show "insufficient data" instead of pretending it is calm. Each day of data is sealed once it closes.
How this differs from GDELT
We do fewer things, but with a sealed, public method, and we show the gaps instead of always giving a number.
Map
Borders from Natural Earth 1:110m (public domain), which follows de facto control of territories. Veramundi takes no position on disputed borders. The map and fonts are served by the site itself.