GlobalGist puts the world's news on a map by where it happens. When you open it, the map turns until a place lands under the cross in the middle. Drag the map to choose another place. The panel lists what was reported there, newest first.
An AI model rates each story's importance from 1 to 5. A place's mark shows the rating of its top story. An empty mark means 1, a filled mark means 2 or 3, and a filled mark with an outer ring means 4 or 5. A bigger mark means more reports. Open the Key on the map to see each mark in the design you chose.
Zoomed out, the map shows only places with a story rated 4 or 5, or reported by outlets in four or more cities. Each step in adds the next level, down to every story. Places close together share one mark until you zoom in far enough to separate them.
The reports come from a hand-picked list of outlets in every region. Each story sits at the city where it happened. The AI model reads where the reporting says it took place. A program then checks that city against a fixed list of world cities before placing it. A story that names no single city stays at the city where its outlet is based. Every report also says where its outlet is based. The map shows no borders and names no countries. The outlets' reports, the events and the word update once a day.
Our outlets don't reach every town. The map adds the newest local stories from towns our outlets didn't report from that day, up to two from each town and every town's newest first, found through GDELT, an open index of news sites in more than 65 languages. GDELT tags each article with the towns it names. A program checks that town against the same list of world cities and a list of towns from GeoNames, down to places of 1,000 people and every seat of local government, and places the story there. No AI model reads or rates these stories. They are marked "via GDELT", load and show only when you zoom in close, and link to the outlet that published them. They update every three hours.
Each day an AI model sorts the day's reports into events and rates each event from 1 to 5 for importance. For each event rated 3 or higher, up to 25 a day, the model writes an explanation that a program checks against the sources. The model then scores each of these events from −2 to +2 by what happened to people. Loss of life or harm on a large scale is −2. Lives saved or a lasting agreement is +2. The model scores what happened to people, never which side gained. It scores the day three times, and each event keeps its middle score, so one odd answer can't change the word. Every score quotes the checked sentence it rests on.
A fixed formula, not the model, then places the day on a five-step scale: Severe harm, Harm, Mixed, Relief, Resolution. The steps describe what happened to people in the scored events. They do not judge the day. If any scored event is below zero, the lowest of those scores sets the day, so good news never cancels out a tragedy. Otherwise the day is the average score, and more important events count for more. The model picks the word from a fixed list for that step, and a program rejects any other word. Open the word to see every score and its reason.
The same model writes an explanation for some events. Every sentence must quote a passage from a source article. A program checks that the passage is in the article, word for word, and removes any sentence that fails. Each sentence links to its source and the quoted passage.
Translate uses the translator built into your browser, which runs on your device. Nothing is sent anywhere. Browsers without a built-in translator can't translate.
No cookies, no analytics and no accounts. Your design, view, topics, Translate setting and pinned places are saved in this browser's local storage only. A story's picture, and an article you open inside the site, load from the outlet that published it.
Map and city data: Natural Earth (public domain). Town list: GeoNames, CC BY 4.0. Local stories: The GDELT Project. The Course of Empire design shows Thomas Cole's The Course of Empire: Destruction (1836, New-York Historical Society, public domain) in black and white. Fonts: open source, under the SIL Open Font License 1.1 or the Apache License 2.0. Headlines, summaries and articles belong to their publishers.