Every day, for every country, we look at what people read, search, watch, listen to, play and buy online, and find what stands out. Here's how, in plain words.
What we measure
We use public signals, published for each country, of what people:
- Read: the articles people look up most.
- Search for: what's trending in searches.
- Watch: what's topping streaming charts.
- Listen to and download: music and app charts.
- Play and buy: best-selling games.
- Expect: what prediction markets are trading on, shown as market odds.
Each source is credited next to the data it provides.
That's the attention side of a zeitgeist. For the opinion side we add what readers tell us in Pulse polls and debates, what public surveys found (below), how much people post about a topic on Bluesky, and the tone of the news. Each is labelled with what it measures.
One topic, many languages and sources
People in Brazil read about a series in Portuguese, people in Japan search for it in Japanese, and it also sits in a streaming chart. We recognise that these are the same topic, so you see it once, with every place it shows up.
Spotting what's new
Big topics are always big: a country's most famous football club or its president is read about every day. So we don't just rank by volume. We compare each topic with its own normal level in that country, and label the ones that stand out:
- New: it wasn't on anyone's radar until now.
- Breakout: a sudden jump far above its usual level.
- Rising: steadily climbing for several days.
- Back: returning to a chart it recently left.
- Falling: clearly fading.
- Weekly / Yearly: a repeat, like a weekly TV show or an annual holiday. These are shown but not counted as news.
Because every country is compared with itself, small countries aren't drowned out by big ones.
Ranking a country's topics
Each country's list combines how big a topic is with how fast it's changing. A topic that shows up in several kinds of sources (say, both searches and charts) counts for more than one seen in a single place. We put every topic in one of 20 categories, from politics to gaming, so you can see what's happening in each part of life.
Everyday apps
Some apps sit at the top of the app charts every single day: messaging apps, social networks, VPNs. They tell you what people use, not what's new. An app (other than a game) that was already on the previous day's chart is listed after every other topic and marked Everyday app, unless it really moves: it jumps into the top 10% of the chart, comes back onto it, or people start reading or searching about it (an outage, a ban, a new feature). Swapping places at the top of the chart doesn't count. Apps on the charts of many countries at once are only counted as moving when people read or search about them, because a handful of downloads can move them around a small country's chart. Nothing is removed: everyday apps keep their place in the charts and are named on each country page.
Issues, everything, entertainment
The world map can show three things for each country. Issues (the default) is the top topic in politics and world affairs, society and beliefs, business and the economy, money, tech and AI, science, health, climate and education that people in the country read or search about; app charts alone don't count here. Crime is left out of this view: on most days it is dominated by true-crime stories people read about because of a documentary or a series, not by the news. Everything is the country's top topic of any kind, and Entertainment its top film, show, song, game, sports or celebrity topic. Countries where too little is read or searched to tell are marked "No issue stood out".
Signed-in readers can add My interests: each country's top topic among the topics, issues and industries they follow. They can also hide or blur topics by category, by issue or by content (violence, war, sexual abuse, recent deaths, explicit songs, games for adults). Those content labels are set automatically from Wikidata and the sources, or from the title when a topic has no Wikidata entry. They will miss some things and catch others: they help you read, they are not parental controls. Hidden topics stay on the page, and each list says how many it hides.
Industries and careers
Every topic is tagged with the industries it belongs to: by its category (food and drink topics count for food and beverage), by its App Store genre, or by what Wikidata says it is or works in (a cryptocurrency, a bank, an airline, an insurance company). Each industry page shows, country by country, the industry's topics among the top 100 people pay attention to today, next to what households spend on its spending categories per person and how pricey they are against the world (World Bank, International Comparison Program 2021, the latest benchmark), how fast those prices rose over a year (Eurostat, then the OECD), and the sector's share of the economy and of jobs (World Bank, ILO estimates). Careers group the industries where a kind of work happens. We can't see sales, the price of single products, company figures or what people buy in shops.
Each career also shows its job postings on Indeed against 1 February 2020 (Indeed Hiring Lab, CC BY 4.0): Indeed publishes postings by occupational sector for six countries (Australia, Canada, France, Germany, the United Kingdom and the United States), and we average the sectors that fit each career. People in the news are tagged with careers by their occupation on Wikidata, through the occupation's ISCO-08 code. A person can hold several occupations, and an occupation is only as precise as its code, so a few people land in a neighbouring career.
Industry pages also show how much of each country's news is about the industry today, and how that compares with the past four weeks, from GDELT's news themes (17 industries whose subjects the themes describe: energy, finance, health, education, transport, tourism, food and others; films, music, games, fashion and sport have no themes of their own). Topics that belong to the same media franchise, brand, series or maker on Wikidata are grouped, so several topics from one franchise read as one story (on the world, country and industry pages, when at least two different topics share it).
Explore: where attention goes
Explore splits a place's Wikipedia reading over the last 7 days into 10 areas (industry and the economy, tech and science, politics and power, problems and concerns, society and beliefs, the environment, health, culture and entertainment, sport, lifestyle), then into sub-areas, and down to topics and companies. It counts every page read in a country that Wikimedia publishes, not only the top 100: about 180,000 pages a week. Wikimedia publishes a page's daily views in a country only above 90, 550 or 1,000 views, depending on the country, so small countries show their most-read pages only.
Each page is filed by what Wikidata says it is: its class, industry, occupation, sport, genre or field, walked up to the areas of the tree, and people by the role their description names. A page can sit in up to three areas, its views split between them. Adult content and lists count nowhere. We check the filing against a sample of 300 hand-labelled pages: on the 100 we never tuned the rules on, the area agreed for 87% (94% of their views) and the sub-area for 83%. Pages we can't file (about 7% of views) are left out of the shares.
A share is an area's views divided by the views of the area above it. An area shows once it has at least 5 pages and 2,000 views in the place, and a country from 5,000 filed views a week (small countries show only their largest areas; below that, Wikimedia's privacy thresholds leave a country only a handful of pages a week); the rest, and pages filed at the level above, show as "Other". The world counts everyone's reading together, so large countries weigh more, as they do in reality. "× the world" compares a country's share with the world's, pulled toward the world's when the country has little reading, so a few hundred views can't make an area look unusual. Where GDELT's news sample covers an area, a tick marks its share of the news. The most-read topics and companies need at least 5 Wikipedia editions and a clean name.
Pages read at least 5,000 times a week that no rule files are filed by an AI model, which picks one area from the tree's list (or none) from the page's name and description; an answer outside the list is thrown away, and a rule always wins over the model. With Pro, each area also shows its likely range (where its share would land in about 19 weeks out of 20 like this one), its share of the news, its weekly history, and its companies by heat: how each ranks among the area's companies in reading, growth on the week before, news headlines naming it and money raised (what it told the SEC it raised privately in the last 30 days, matched by its SEC filer number on Wikidata), averaged over the parts it has (a part counts once at least three of the area's companies have it). Each also shows its reading against the 12 weeks before.
In the news this week sorts a random sample of the week's headlines about an area (GDELT, about 70 for the world and 35 for a country) into the few stories they tell. An AI model names the stories and assigns each headline to one of them or to none; we count the headlines per story, so the shares are counts over the sample, not the model's estimates. A story needs at least two headlines, and its name and one-sentence summary must use only names and numbers found in the headlines, or they are dropped. Stories that continue from the week before keep their id. Marked AI-generated.
Top stories on the home map's card, and on each country's and US state's page, are the week's stories behind a country's leading theme. We take the week's GDELT headlines about the theme's areas that mention the country (a sample of them each day) and group the ones that share rare words and a phrase: a story needs two headlines from two outlets. The stories the country's own news sites carry come first, then those carried by the most outlets. Where one of the week's stories above matches, its name is shown. Each headline links to its outlet. A theme with no headlines that week has no stories, so the card shows the next of the country's top themes that has stories, and names it.
A theme's share of trending lists is a second lens: how many places it holds on each country's daily top 40, the ranking on its country page (trending searches, Wikipedia, the App Store, Apple Music and Podcasts, the Netflix Top 10, Steam), averaged over the same 7 days. Each topic counts toward the themes its explorer areas belong to, split like a page's reading when it sits in several; a topic with no area (an app, a song or a Steam game too little read on Wikipedia) counts toward its category's theme (games, screen culture, music, apps …), and categories that span several themes (politics, sport, business, health) count toward none. A list counts with at least 20 topics from at least 3 sources, a country with 4 days of lists; the world's figure is the plain average of those countries. It counts places on charts, not people or money, and some kinds of things have charts of their own: games hold a quarter of the US's, the UK's and Japan's top 40 through Steam while they are well under 1% of what people read and the news covers.
A country's zeitgeist by theme, the first thing on its page and on its card on the home map, splits its attention this week so it adds to exactly 100%: a third its Google searches, a third its Wikipedia reading and a third its news, each of them split so it adds to 100% on its own. A search's count of searchers is split between its themes (most have one), and searches in no theme are "Everything else". A theme's news is its share of the country's news as above, and the news no theme covers is "Everything else". In the reading, each page's views go whole to its main theme: the theme its areas count toward most (a page filed under three areas, two of them screen culture's and one celebrity culture's, counts in screen culture). When two themes are reached equally, the one listed first in our catalogue of themes takes it. Pages in no theme are "Everything else"; themes under 1% are pooled. So unlike the shares of attention above, where an actor's page counts toward both screen culture and celebrity culture and the themes overlap, here every page counts once, and the ring's parts are a true breakdown; a theme's part of the ring can differ a little from its share of attention. Selecting a theme shows what's inside it in that country: its most searched on Google (with the headline Google lists), its most-read pages on Wikipedia and its top stories in the news; for sport and culture, which our news measure doesn't cover, the headlines Google lists with the searches. Countries whose zeitgeist comes from their news alone (China, where Wikipedia and Google are blocked, and the countries Wikimedia publishes no reading for), US states and China's provinces have no reading to split: their ring is each theme's share of their news, with the news no theme covers as everything else. A story can be about two themes, so where those shares add up to more than 100% they are scaled down to fit the ring.
A theme's breakdown is a ring of what its Wikipedia reading this week is made of, in the world or any country with reading: the areas it covers and, inside each, the sub-areas the explorer shows, plus what is filed at an area itself ("Other …"). A page's views are already split between its areas, so the parts are a partition: nothing counts twice, and the shares, rounded together to one decimal, always add to exactly 100%. Parts under 1% are pooled into "Everything else", which lists them. A part with sub-areas of its own opens into its own ring; any other part (films, TV series, video games) shows its five most-read pages, with a picture where a free one exists: a Wikimedia Commons file under CC0, public domain, CC BY or CC BY-SA, shown with its author and licence, or a store's own artwork with a link to its store; never a non-free film poster. A card's picture or name opens its topic page or, without one, a panel on this site: where else it is among the most read this week (its Wikipedia views in each country) and which countries' pages list it today, with its sources (Wikipedia, the store, the picture's Commons file) as links that leave the site. Next to it, the ring of the pillar the theme sits in (culture and entertainment, for screen culture) shows every area's share of that pillar's reading, the theme's areas in colour. A country's page shows what it reads about: its reading by pillar, each opening down to its most-read pages, and its leading theme's parts. Countries whose zeitgeist comes from their news alone show no reading ring. Most read on the map's card is the leading theme's three most-read pages in that country.Y Combinator–backed startups people read about lists, by their Wikipedia views in the theme this week, the companies Wikidata records as backed by Y Combinator (as investor or funder, or through one of its batches), refreshed weekly; a theme needs at least two. Names, founding years and websites come from Wikidata too, never from Y Combinator's own site.
New and growing on each theme's page lists phrases that are new or growing in the theme's news headlines. Every week we count the 2- and 3-word phrases and names in the same sample of GDELT headlines about the theme's areas (English-language headlines only, so a phrase and its translation don't split one thread) and compare the week with up to eight weeks before. A phrase needs at least 4 headlines from 3 outlets on 2 days this week, in at least 2 different stories, and must be at least twice as common in the theme's headlines as in the week's news overall; the theme's own words, its everyday phrases and place names are left out.New means it was under one headline a week before; Rising means it was there before and has doubled, or grown by a quarter two weeks in a row. Each phrase shows one of its headlines, linked to its outlet.
Headlines in other languages than the page's are machine-translated: an AI model translates each headline we show (only the headline) once into the site's languages. A translated headline is marked with the language it was written in ("Translated from Chinese"), and the original shows when you point at it. A headline already in the page's language, or one whose translation failed our checks, is shown as written.
How the news sounds for an area compares the week's news about it (GDELT) with the four weeks before: its tone (positive minus negative words per 100) and how often it uses anxious, fearful, angry, sad and happy words, as on the country pages. An area needs 20 articles in the week, and a usual level 14 days of news in the four weeks before. How readers feel is a monthly Pulse question for each area (excited, hopeful, unsure, worried or angry), from readers who answer, not a representative sample. For the US, each area also shows what businesses in its sector told the Census Bureau's Business Trends and Outlook Survey every two weeks: whether they used AI in the last two weeks, whether they expect to within six months, and how their business is doing (an index from 0, all poor, to 100, all excellent; 50 is about average).
Gaps and opportunities (Pro, with one shown free): where a country's reading, its news, its new businesses and its households' spending don't line up. Wikipedia and the news file an area with different breadth, so each side is compared with the world first, and a gap is when the two comparisons differ by 1.5 times or more: read more than the news covers it, or covered more than read (the news side pulled toward the world's share when articles are few). Interest rising while new businesses lag compares an area's reading share over the last four weeks with four weeks about three months before, next to new business applications in its industries on a year ago (US Census Bureau, French Insee). Big in spending, small in reading compares the share of household spending on the area's categories (World Bank ICP 2021) with the median country, against how much the country reads about it compared with the world. Once a week, for the 30 countries with the most reading, an AI model reads these gaps and writes up to three opportunity cards for each country and each area with enough gaps; every card cites the gaps it comes from, and every name and number in it is checked against them. They are openings to look into, not forecasts.
US states: Wikipedia doesn't publish reading by state, so a state shows the news that names it (GDELT, through the places articles mention): each area's share of the week's articles naming the state, against the same share in the news about the US and over the four weeks before. A state shows from 200 articles in the week and an area from 10. Each state also shows the Census Bureau's business survey for it, its new business applications (Business Formation Statistics) and the private raises filed with the SEC (Form D). For an area tied to a few industries, it shows the state's private workplaces in them (the Bureau of Labor Statistics' quarterly census, by NAICS code) per 100,000 residents, against the US.
Countries without Wikipedia reading: Wikipedia is blocked in China, Wikimedia doesn't publish reading for some countries (Russia, Iran, Saudi Arabia, Vietnam and others) and small countries read too little. A country with no reading gets its zeitgeist from its news, as a US state does. Since 21 October 2026 that is its own news sites' news when they publish at least 1,000 articles in the week, 300 of them about the areas we measure (China's own sites publish about 15,000 a day, sport and entertainment included): what its own media covers, not what the world writes about it. A country whose own sites are too few in GDELT (Iran) keeps the news naming it, from the same minimums. Each theme's share of that news, against the same share in the world's news, how its news sounds, and the week's stories. Its pages say so ("What stands out in China's news is …"), and every number on them is labelled as news. A country with enough reading always keeps its reading.
China's provinces are counted like US states: the news that names a province (GDELT, through the places articles mention), each theme's share of it against the same share in the news about China, how it sounds, and the week's stories, from 200 articles in the week (a province's page from 300). The 31 mainland provinces, regions and municipalities count; Hong Kong, Macau and Taiwan have pages of their own. Province shapes come from Natural Earth.
What the news is talking about
Next to what people look up, country pages show what the news is talking about, from The GDELT Project, which reads news sites worldwide in more than 100 languages and machine-translates them. We sort each article into eight issues (AI, the cost of living, climate, immigration, democracy and elections, jobs and work, health, and war and security) by the themes GDELT tags it with and, for AI, by names and words in its headline. An article can count for several issues, or none.
For each country we measure how much of its news is about each issue, and how the news sounds: its tone (positive minus negative words) and how many words of anxiety, fear, anger, sadness and financial uncertainty it uses per 1,000 words, from standard English word lists. We read the country's own news sites when they published at least 100 articles that week, and otherwise all articles that mention the country. Every number is compared with the same country's past 4 weeks: the median day when there's plenty of news (at least 300 articles a day), or the last 7 days against the 4 weeks before when there's less. An issue is marked More than usual when its share of the news is at least 1.5 times its usual share, from at least 20 articles.
Word lists are English and run on machine translations, and newsrooms differ in style, so we compare each country's tone and emotions with its own past, not with another country's. Shares can be compared: they count articles. Some articles land in the wrong issue (a news-graphics label containing "AI", say); headlines are shown so you can check. Where a survey asked people about the same issue, its answer sits next to the issue, marked as a survey.
Confidence and limited data
Some signals are stronger than others. A topic seen in only one source, with no history to compare against yet, is marked Low confidence. Countries where few sources are available are marked Limited data, and we say so rather than guess.
How our calls hold up
Every week we check our own labels against what happened next, on every country and day in the archive: how many topics we labelled rising were still in their country's top 100 one, four and twelve weeks later, next to the same share for every top topic, and how well "where it's heading next" did against chance. The results are on the scorecard, good news or not.
Readers' calls
Signed-in readers can call whether a topic in a country (one rising there today), or an industry in a country, will get more or less attention after 1, 4 or 12 weeks. On that day we compare the topic's score in the country's ranking, or the industry's share of the country's attention, with the day of the call: higher is "more", the same or lower is "less". A topic that has left the country's list counts as lower. If the country has no data within 3 days of the due day, the call is not judged.
A right call earns 100 times the chance it was wrong, where the chance comes from the archive: how often topics in a country's top 100 (or industries with topics) had a higher score after the same time. Most rising topics fade, so a right "less" earns little and a right "more" earns a lot, and always calling the same way earns on average what calling at random would. Until the archive has enough history for a horizon, both ways earn 50. Readers appear under a name of two words and a number from a fixed list. Calls are a game and a record, not advice.
Where topics are heading next
Topics often travel: what rises in one country often reaches its neighbours or countries with shared language or culture a few days later. From our daily archive, we learn how topics have spread between countries in the past and estimate where a rising topic is likely to go next. These are estimates, not promises. Simultaneous worldwide releases are not counted as spreading.
The numbers behind the mood
Country pages also show official figures, such as inflation, unemployment, consumer and business confidence, interest rates and new businesses, from international statistics offices and central banks. They sit next to the trends, never inside them: they don't change what's ranked. Each figure shows the period it covers and who published it, because some are monthly and others only yearly. Where two publishers report the same thing, we show the one closest to the country (a European figure for European countries, for example).
What people say in surveys
Next to what people look up and what the news covers, some country pages show what people said when they were asked. From the European Commission's Standard Eurobarometer (twice a year, the EU and its neighbours): the issue most people name as the most important facing their country, and how many tend to trust the national government, the media and scientists. From the OECD Trust Survey(every two years, about 30 countries): trust in the national government. From the Climate Change Opinion Survey by Meta and the Yale Program on Climate Change Communication (2023, about 110 countries): worry about climate change, its causes, and how high a priority it should be. That one surveys Facebook users, so it describes them, not a whole country. Each figure names its survey and when it was asked; surveys are slow, so they change once or twice a year at most.
The Economic Policy Uncertainty index, a monthly count of newspaper articles about economic policy and uncertainty built by economists for 21 countries, sits with the news mood.
Pulse: what readers say
Pulse asks signed-in readers (accounts are for adults) short questions we write: how worried they are about each issue in the news where they live, whether they spend more or less on everyday things than a year ago, where they think each industry is heading, and whether they have heard of topics rising around the world. Answers are fixed choices, never free text. Each account answers each question once a month (once ever for "have you heard of it?") and can change its answer that month. Results show only after answering, so they can't sway the answer.
The answer's country is where the connection comes from, as Cloudflare reports it, and in the US its state too. Results across all readers show from 30 answers; a country's own result from 100 answers there, and a US state's from 100 answers there. Only totals leave the account database, and a country's counts only from 10 answers. A country's share of readers worried about an issue appears next to its news mood and in the issue's survey table, marked as Pulse. These are readers who chose to answer, not a representative sample of anyone: read them as a signal of what our readers think, not of what a country thinks.
Age and gender
None of the signals we measure records who is paying attention: Wikipedia, Google, the news and the charts count reading, searching and listening, not the age or gender of the people doing it. So we don't guess what women or 18–29-year-olds are looking at. Where the site shows age and gender, it comes from people who told us.
Pulse by age and gender (Pro). Readers can add their birth year and gender to their account, optionally, to break Pulse results down. They are kept apart from the answers and used only to count. Results show for four age bands (18–29, 30–49, 50–64 and 65+, plus 18–24 and 25–34 once those groups are big enough on their own) and for women and men; other answers count in the totals. Each country's answers are weighted to that country's adults by age and sex, using population figures from the World Bank's World Development Indicators (from the UN World Population Prospects, CC BY 4.0), and readers everywhere combine those weighted answers.
A group shows only from 100 answers and an effective sample of 50 after weighting, and a share only when its 95% interval is narrow enough to read (the rule the US National Center for Health Statistics uses). Below an effective 100 answers, the margin of error is shown. Questions whose answers could hint at political opinions (worries about immigration, and about democracy and elections) aren't broken down. Weighting brings an opt-in poll closer to a country's adults but not all the way: in Pew Research Center's tests, weighted opt-in samples were still about 6 points off on average, and further for 18–29-year-olds. Read these as what our readers in each group said.
Pulse is for adults. Accounts are 18+, and a birth year under 18 ends Pulse for that account: its answers are deleted and the year isn't kept.
Who uses which platforms (US adults)
Some of what we measure comes from platforms whose users lean young or old, male or female. In the US, Pew Research Center asked adults which platforms they ever use. These are shares of people who use each platform, not of what they pay attention to.
| % of US adults who use it | All adults | 18–29 | 30–49 | 50–64 | 65+ | Men | Women |
|---|---|---|---|---|---|---|---|
| YouTube | 84 | 95 | 92 | 85 | 64 | 86 | 83 |
| 71 | 68 | 80 | 74 | 57 | 63 | 78 | |
| 50 | 80 | 62 | 40 | 19 | 44 | 55 | |
| TikTok | 37 | 63 | 44 | 30 | 12 | 30 | 42 |
| 32 | 37 | 40 | 30 | 20 | 30 | 34 | |
| 26 | 48 | 35 | 16 | 6 | 29 | 23 | |
| Snapchat | 25 | 58 | 31 | 13 | 4 | 22 | 28 |
| X (formerly Twitter) | 21 | 33 | 25 | 16 | 10 | 25 | 16 |
| Threads | 8 | 15 | 10 | 6 | 3 | 8 | 9 |
| Bluesky | 4 | 6 | 5 | 3 | 2 | 4 | 3 |
| Truth Social | 3 | 1 | 3 | 5 | 4 | 4 | 3 |
Source: “Social Media Fact Sheet.” Pew Research Center, Washington, D.C. (November 20, 2025) https://www.pewresearch.org/internet/fact-sheet/social-media/. Pew Research Center's National Public Opinion Reference Survey, 5,022 US adults, February 5 to June 18, 2025.
How people get news, by age (48 countries and territories)
The Reuters Institute's Digital News Report asks about 97,000 people online in 48 markets each year how they get news. Younger people get it from social media, video networks and, increasingly, AI chatbots; older people from TV and news sites. These are shares of all markets together, by age.
| 18–24 | 25–34 | 35–44 | 45–54 | 55+ | |
|---|---|---|---|---|---|
| Social media, video networks and AI | 52 | 44 | 36 | 28 | 18 |
| News websites and apps | 20 | 24 | 28 | 27 | 24 |
| TV | 18 | 21 | 26 | 34 | 47 |
People who used a source of news in the last week, all markets: 18–24 = 10,252; 25–34 = 15,771; 35–44 = 16,575; 45–54 = 15,729; 55+ = 33,394 (Q4).
| Under 35 | 35 and over | |
|---|---|---|
| Social media | 43 | 30 |
| Search | 22 | 23 |
| Going direct to a news site or app | 15 | 25 |
| Aggregators | 8 | 8 |
| Mobile alerts | 7 | 7 |
| 3 | 4 |
People who used an online news gateway in the last week, all markets: under 35 = 25,185; 35+ = 62,626 (Q10a).
| 2025 | 2026 | |
|---|---|---|
| 32 | 42 | |
| TikTok | 28 | 36 |
| YouTube | 30 | 33 |
| 27 | 29 | |
| 19 | 26 | |
| X | 18 | 15 |
18–24-year-olds, all markets (Q12B). 2026 from a repoll in March 2026. TikTok is banned in India and doesn't operate in Hong Kong.
| 18–24 | 25–34 | 35–44 | 45–54 | 55+ | |
|---|---|---|---|---|---|
| AI chatbots | 17 | 15 | 11 | 8 | 5 |
Total sample, all markets: 18–24 = 11,041; 25–34 = 16,944; 35–44 = 17,822; 45–54 = 16,764; 55+ = 34,949 (Q3).
Source: Reuters Institute Digital News Report 2026 (doi), licensed CC BY 4.0; figures from the report's charts, shown here as tables.
Rising on Google
For 42 countries, country pages also list the searches that rose most on Google in the last complete week, from Google Trends' public dataset. Google publishes them region by region; in practice every region of a country lists the same searches, so we show those most of its regions share, linked to each search on Google Trends. Adult searches, and the long "team vs team standings" entries Google generates, are left out. Like the other figures next to the ranking, they never change it.
Trending on Bluesky
The world page lists the topics trending on Bluesky that day, with their category and how many posts mention them. The list is Bluesky's own: one list for the whole world, in practice mostly about the United States and in English, and its names are written by Bluesky. We link to Bluesky's page for each topic and keep only the topic, its name and its counts, never posts or accounts. It never affects the ranking.
On social media
The home page, the theme pages and some country pages show the TikTok videos, Instagram posts and YouTube videos that the week's most-liked Bluesky posts shared. Each day we ask Bluesky's public search for the most-liked posts linking to TikTok, Instagram and YouTube, once for the whole world and once for each language that stands for a country (Japanese for Japan, German for Germany …), and rank each video or post by the likes and reposts of the posts that shared it. The English-speaking countries show the world's items shared in English; other countries show none. Bluesky's posts in Chinese come mostly from Taiwan, Hong Kong and abroad, and China's own platforms (Weibo, Douyin, Xiaohongshu, Bilibili) can't be read from outside China within their terms, so a country's page also lists what's going around online there, as its own news sites report it: the week's headlines from the country's own outlets saying a story trended, went viral or got people talking online ("冲上热搜", "网友热议", "в Сети", "went viral"), found by their words in GDELT's headlines, at most two from one site, each linking to its article. We keep only those counts, the posts' languages and the links, never the posts' text or anything about who wrote them, and leave out posts with a content warning. We never read or copy TikTok's or Instagram's own data: each item plays only in its platform's own player, loaded when you press play. Titles and pictures are the platform's own link preview, as a Bluesky post's link card carries it (TikTok's caption and cover, Instagram's caption and picture), or YouTube's title and thumbnail. A TikTok video or Instagram post without a link preview isn't shown. An AI model files the items under the zeitgeist's themes for the theme pages. It never affects the ranking.
What's trending on social media: our ranking
Readers share posts from TikTok, Instagram, YouTube, X, Threads, Bluesky, Facebook, Reddit, Pinterest and LinkedIn by pasting a link, or post pictures of their own. Nothing else to fill in: an AI model (Cloudflare's Llama 4 Scout) reads the post's text and picture and files it under up to 3 themes and 3 subjects (industries among them), the countries it's mainly about (or the world) and the feeling it's most likely to give; without a title from the platform, it writes a short neutral one. A post's own text is never changed. The Social page and the "On social media" sections rank posts every hour with our own formula, from what we're allowed to count:
- Readers here: how many people shared a post (weight 4), its likes (2), comments (3) and plays or opens on the site (1). Each fades with a 36-hour half-life over 7 days, and each person counts once.
- Bluesky: how many of the week's most-liked Bluesky posts shared it (1.5) and their likes, reposts and quotes (1, counted per 5). A Bluesky post shared here counts its own likes and reposts the same way.
- Growth: a post with more activity in the last 24 hours than in the 3 days before rises, by up to 75%.
Subject and industry pages also get the past week's most-liked Bluesky posts with a picture, a video or a link for the subject's hashtags (#uidesign, #gamedev …, listed inmeta/social-subject-queries.json), each checked before it shows and ranked there by its own likes and reposts. Every signal counts on a log scale (the 100th share adds less than the 2nd), so no single source decides. A post counts in the world's ranking and in each country, theme and subject the model filed it under; Bluesky's posts also count in the country whose language they were in (the world's English posts for the English-speaking countries). Platforms' own counts (YouTube's views and likes, Bluesky's likes) are shown next to a post where their terms allow, but never ranked: YouTube's terms forbid building our own metrics from them. We never read TikTok's, Instagram's or other platforms' data ourselves; each post plays in its platform's own embed.
What deserves attention
Signed-in readers can pick up to 3 issues they think deserve more attention, for the world or their country, across all issues or within a theme or an industry, and suggest issues of their own. We show the share of people who picked each one, next to that issue's share of the news today (the 8 issues we measure in the news, see Issues). The gap between the two is what people feel is missing from the coverage. It's not a representative poll: the people who pick are our readers, and only totals are ever shown. The totals count the picks of the last 90 days, so they say what matters now; older picks are kept as history (until the person withdraws), and so are weekly totals (counts of 3 or more).
Top 200 games on Steam
Once a week we read each country's Steam top sellers (Valve's own list, 100 games per country) with each game's 20 strongest player tags. The world list gives a game 101 minus its rank in every country's list, over 100, and ranks the most points first: every country counts the same, so it shows what sells widely, not only in the biggest markets. DLCs, hardware such as the Steam Deck, adult-only games and countries whose list is a copy of the US list are left out. Art style, viewpoint, genre, theme, mood, mechanics and players come from the players' tags, grouped our way; a game seen in first or third person whose tags don't say 2D or 3D counts as 3D.
The same week we read each country's App Store charts for games (Apple's own lists of the top-grossing and top free games on iPhone and iPad, 100 each) and rank the world's top 100 of each the same way. A game's page adds its photos, description and rating from Apple's own lookup, and Steam's for PC games, with each store's news where it has some. Android and the consoles publish no charts we may use, so they aren't here yet.
Most AI-exploited art styles
For each art style and sub-style, once a week: how often Bluesky posts naming it are tagged #aiart against #art (the newest 100 of each, over the days they span; artists on Bluesky tag their own work #art), and how many text-to-image models on Hugging Face are named for it (most are models made to copy one style). Its AI share is #aiart's share of the two. Its saturation, from 0 to 100, ranks it against every other style: the AI share counts 60%, the number of models 40%. A style needs 10 posts in all to be measured. "Stand out with" names a style's least saturated sub-styles that artists still post at least every few days. Tags are what people choose to write, so this is a measure of what is posted and built, not of every image made.
AI briefs
Each country's short brief is written by an AI model. To keep it honest:
- It sees only that country's data on the page, nothing else.
- Every sentence must point to the topics it's based on.
- It may not invent reasons. It can only say why something is trending when a linked news headline says so.
- It may not copy headlines, mention things from other countries' data, or use loaded language.
- Every brief is checked automatically before it's published. A brief that fails gets no place on the page: no brief is better than a wrong one.
Topic pages can carry a short "Why it's rising" note written the same way. It starts with the news: what happened, as a cited headline reports it, and who or what the topic is. The model sees only that topic's figures (where and when it rose, which sources show it), up to eight headlines from countries where it rose, those that name the topic first, and the opening of its English Wikipedia article. It may give a reason only when it cites a headline that states it, may not name anything that isn't in that input, and is checked automatically before it's published.
Briefs and notes are always labelled AI-generated. Rankings, labels and charts are calculated, not written by AI.
Subject pages
Each subject page (UI, AI, Marketing, Human relations, every industry) shows what the field argues about, where people say it is heading and what those ahead of the curve are doing. It is built from two things only:
- Quotes by named people, word for word, each linked to the page where it can be read. Every quote is checked automatically against that page, or against the episode's transcript for podcasts, and is shown only if it is found exactly. Quotes are never shortened or edited.
- Published survey figures, entered as the publisher printed them, with who answered, how many and when. These are the only numbers on the page about what people think or feel.
Once a week an AI model reads a subject's quotes and figures and names the questions the field argues about, the positions on each side, the direction the quotes point, common confusions and what practitioners ahead of the curve do. It must cite a quote or a figure for every point, may not name anything that isn't in its input, and may not say what a group thinks or feels unless a survey that asked that group is cited. A position gets a size only from such a survey. Everything it writes is checked automatically before it's published and labelled AI-generated.
Forecasts keep the speaker's own words for when, and a date they can be checked. Once that date passes they are marked due and graded by hand, against published evidence. The grade is about the prediction, not the person. To correct or remove a quote, use Report a problem next to it.
What we can't see
- Each platform reaches some people more than others: phones, apps and game stores differ a lot from country to country.
- A few countries block or don't use these platforms, so their picture is thin or missing.
- Very small countries have less data, which makes their signals noisier.
- Who reads and searches online isn't everyone: surveys of Wikipedia readers, for example, find more men than women. We describe what people do online, not what everyone thinks.
The same method, every day
The archive is only useful if every day is measured the same way. When we improve the method, we note it in the corrections log.
Questions about the data? data@currentzeitgeist.com