# How Loudest works ## In short - Loudest measures how widely the world's press is covering each story, not how many people read it. - Every ten minutes it reads the public feeds of news outlets around the world and groups articles about the same event into stories. - A story grows with the number of countries and outlets covering it. A new country counts for much more than one more outlet from a country already counted. Each country and outlet counts from when it joined the story and halves every 4 hours, so the map shows what is loud now. - Only hard news is shown on the map. An AI model reads each story's headlines and decides whether it is hard news. - Nobody picks or ranks stories by hand, and no one can pay for a place. The one thing done by hand is merging, when the system has split one event into several stories. - Every change to the method is dated in the [changelog](/methodology/changelog), and each archived day keeps the ranking it was captured with. ## 1. Sources Loudest reads one public RSS or Atom feed per outlet. The full list, with each outlet's country, language and ownership, is on the [sources page](/sources), together with the current counts. - Each outlet is filed under its home country. A few outlets that cover a region from abroad are filed under the region they cover. - Larger countries have more feeds than smaller ones, within a floor and a ceiling, so that no country is left with a single voice where more could be found and none drowns out the rest. - Outlets are tagged as private, public-service or state-owned, as well as that can be established. - Trade press and official sources are read as a separate class. - A small number of outlets with no usable feed of their own are read through Google News. They are marked on the sources page. - The reader identifies itself as LoudestBot, obeys each publisher's robots.txt and signs its requests. Details are on the [bot page](/bot). Loudest uses only the headline and a short excerpt of each feed item. It never reads or stores the article itself. ## 2. From articles to stories Articles about the same event are grouped into one story automatically, by how similar they are in meaning. This works across languages without translation. - A story lasts 48 hours from its first article. Coverage that continues after that starts a new story, linked to the one it continues, so a long-running subject is a chain of stories. - When the system creates two stories for one event, they are merged once it is clear they are the same. Most merges are automatic. Occasionally, when the system has split one event into several stories and does not join them itself, they are merged by hand. A merge only joins stories about the same event; it never changes which stories are shown or how they rank beyond that. - A story is shown only when at least two outlets cover it. - The same wire report published by several outlets counts once for each outlet that ran it. An outlet that publishes many articles on one story still counts once. Grouping is approximate. Section 6 says how it goes wrong. ## 3. How a story is sized A story's score is its coverage: - Every country whose general press covers the story counts once, however many of its outlets cover it. This is the main term. Twenty outlets in one country give one country vote, the same as a single outlet elsewhere. Each of those outlets still adds a smaller amount, as the next point says. - Every outlet adds a smaller amount on top, so breadth inside a country still shows. - Each country and each outlet counts in full from the moment it joined the story and halves every 4 hours from then on, on its own clock. A late article from an outlet already counted adds nothing. - Trade press can add only a little, however many trade outlets cover a story, and their countries do not count. The moment an outlet joined is the moment Loudest first saw its article, not the time the publisher gave. ## 4. Hard news Loudest shows hard news: politics and government, the economy, markets, business and trade, energy, conflict and security, disasters, and public policy, including climate, technology and health policy. Sport, entertainment, culture, lifestyle and celebrity stories appear only when they have consequences for money, power, security or large numbers of people. Where the line falls: - Technology and AI count when they involve regulation, chips and supply chains, big-tech earnings or deals, security breaches, or use by states. Product launches and reviews do not. - Crime counts when it is terrorism, a mass-casualty attack, political violence, or organised crime with state or economic impact. Individual and local crime does not. - Sport counts when it involves corruption, state doping, public money for hosting, or state ownership of clubs. Results, transfers and personalities do not. - Health counts when it involves outbreaks, drug approvals and pricing, or public-health policy. Lifestyle research does not. How it is applied: - Articles that their publisher clearly marks as sport, entertainment, arts or lifestyle are set aside before processing, and counted. - Every other story with enough coverage is judged by an AI model, from its headlines only. - A story appears on the map only once it has been judged. - A story judged outside the rule is left off the map but kept in the archive and on its own page, with the judgement shown. - When the judgement is unsure, similar stories already judged are consulted, and a widely covered story is shown rather than hidden. - The number of articles set aside and stories left off the map is published every day in the open dataset. ## 5. Headlines, places and topics The headline on a story is written by an AI model from the outlets' own headlines: one short, neutral line in English. It is rewritten as the story grows. The model also names where the event happened and picks one topic. - A story with no single country goes to International. - Headlines are not reviewed by a person before they appear. To check a story, open it and read the outlets' own headlines. ## 6. What you see - **Size:** the story's score. Bigger means more countries and outlets. - **Colour:** the story's region. - **Freshness:** how recently the story's newest article arrived. The fresher the story, the more its tile stands out: brighter on the dark theme, stronger in colour on the light theme. - **Triangles:** whether the story is louder or quieter than an hour ago. - **Views:** Loudest is the default. Recent shows the most recently seen stories, Rising the ones gaining coverage, One-sided the ones covered overwhelmingly from one region. The [glossary](/glossary) explains every term on the site. ## 7. Archive and data After each day ends (midnight UTC), Loudest records that day's most covered stories. A captured day never changes: a mistake found later is recorded as a dated note on the day's page, never corrected in place. The only exception is removal on legal grounds, which is logged. A day in the archive is ranked by the highest score each story reached that day, by the same rule as the live map. Each day is a page, a JSON file and a CSV file, with a checksum for each file. Every field is documented on the [data page](/data), and the [cite page](/cite) says how to quote it. The data is published under Creative Commons Attribution 4.0. The licence covers the measurements, not the articles linked, which belong to their publishers. ## 8. Limits and biases Read this before quoting the site. - **Loudest measures the outlets it reads, not the world's press.** A story is "covered in 35 countries" when at least one listed outlet in each of 35 countries has an article on it. - **The map leans toward English.** Most feeds are in English, and headlines are always written in English. Coverage in languages that are thinly read can rank low or not appear. - **Country votes are equal, the outlets behind them are not.** Some countries have one listed outlet, others many. One desk's choices can be a country's whole vote. - **State and public-service media count like any other.** A country count does not tell independent coverage from official coverage. - **Grouping makes mistakes in both directions.** Different events are sometimes joined, and one event is sometimes split, most often across languages. A story's figures can be too high or too low. - **Time is when Loudest saw it.** A slow feed makes its outlet look late. - **Rank is a snapshot.** Coverage halves every 4 hours, so the biggest story of the morning can be small by evening. - **Headlines and hard-news judgements are made by a machine, from headlines.** Some are wrong. Every judgement is published, and nothing is hidden without a number saying so. - **Some outlets are missing because they refuse automated readers.** Absence from the map is not absence from the news. - **The archive ranks each story by its highest point in the day.** A story that peaked at noon ranks where it peaked, even if it was quiet by midnight. Its figures are those at the end of the day. The highest point is worked out when the day is archived, so a story that was merged with another during the day can stand higher there than either did on the map. What is never shown: stories covered by a single outlet, articles older than 48 hours when first read, sponsored content, and stories outside the hard-news rule.