Map timeline
Counting, country by country and day by day, how often Chinese state text runs in the world's newspapers.
Scroll the grid sideways to see every theme.
Headlines appear in their original language. For anything not in English, use Translate for the headline and quote, or open the whole article in English.
State origin copies set beside the original state text they reproduce, where it can be found: how much of it outlets keep, cut or rewrite, and how soon they run it. Report only: no label or count changes.
How often China coverage uses the Chinese state's own set phrases, by kind of article. A match is a keyword match, not a reading of the article.
Aggregate engagement with state origin articles on public social media, looked up by hand. Counts only, never a post or an account.
State origin. The text was authored or commissioned by a Chinese state entity and published without substantive editorial alteration. Xinhua wire copy, CGTN promotional releases, China Daily and Global Times syndication, sponsored or branded placements commissioned by Chinese state entities, including those written to order by the publisher's own branded content studio, and signed opinion pieces by Chinese ambassadors and embassy officials. The defining property is that the words are the state's, written by it or to its order, and the receiving publication did not substantively alter them. Deciding to run a piece, or placing it on an opinion page, is not an alteration.
Unchecked state sourcing. Called unverified relay in the codebook, the model prompt and the data files. The article was written by the local outlet, and its central content is a factual claim attributed to a Chinese official or state source, passed on without independent confirmation. A factual claim is an assertion of fact about events, numbers or actions: what happened, how many, who did what. A ministry spokesperson or Xinhua is the only authority for such a fact and the article treats it as established, with no second source. A stated position, policy, warning, threat, opinion or denial is not a factual claim: an article that reports what Beijing said it wants, intends, condemns or rejects is reporting the statement itself, and that is independent journalism even when no outside voice comments on it. "Exports rose 8.7 percent in September, China's commerce ministry said", reported as the figure with no other source, is unchecked state sourcing; "China's foreign ministry warned Manila it would take resolute measures to defend its sovereignty", reported without outside comment, is independent journalism. This factual-claim rule entered the model's instructions on 2026-10-12; model labels made before it are judged again, unchecked state sourcing labels first and oldest first within each kind, with up to half of each day's model calls, and each keeps its old label until the new one arrives.
Independent journalism. The outlet's own reporting, including reporting that quotes Chinese officials but confirms, contextualizes or contests what they say, and reporting of a position, policy, warning, threat, opinion or denial attributed to a Chinese official or state source, with or without outside comment.
Not relevant. Does not concern China.
Unchecked state sourcing: provisional, then published or withheld. Until the model stage has run, unchecked state sourcing is shown as not yet measured, never as zero. Once it has run and until hand coding has checked it, the counts are shown as provisional: they are one model's judgement, no person has checked them, and every figure that contains them says so, on the map legend, in the tooltip and panel, and as relay_status in the CSV export. The check is a hand coding study. The owner and two research assistants code, blind to the model's label, a fixed random sample: about 200 articles in each language with at least 300 model labelled articles, stratified by language and, within each language, by the model's current category, so that unchecked state sourcing is present in numbers that can test it. If Cohen's kappa between their codes and the model on the unchecked sourcing versus independent journalism judgement is at least 0.6, with at least 10 articles on each side, the provisional marking is removed; if it is below, the counts are withheld, not merely annotated, and they stay withheld for any language whose own kappa falls below it. Provisional means not checked yet, never checked and failed. A model re-judging its own sample, run by itself as the budget allows, is kept as a stability check: a same model rerun with the same prompt measures whether the judgement is stable, not whether it is correct, so it never removes the provisional marking, and a rerun below 0.6 withholds the counts. A rerun made under an earlier version of the classifier's instructions is kept only as history and neither supports nor withholds any count; a new one runs once enough articles are labelled under the current instructions. That boundary is where classification error concentrates. Candidates still waiting for that judgement are in no measure on the map, since adding them to judged counts would mix a measured quantity with an unmeasured one. The country panel and the CSV exports show them as their own figure, which is not a finding.
Routes and arrival. Every state origin label records its route, the mechanism named by the signature that established it: a wire credit line or dateline, a press release distribution stamp, a sponsored or partner disclosure, or a byline by a Chinese diplomat. A label the model gave to an article sent as a near copy of a state origin item another outlet ran with its credit line has the route uncredited copy: state text republished without that credit, judged by the model. Any other label the model gave without a signature is unattributed. Every article also records where it was collected: an editorial feed, or the outlet's press release, sponsored or partner section. The route filter shows state origin for one route or one arrival at a time.
Color scale. Exactly zero is white. Any positive value takes one of seven tinted steps from pale pink to deep red. The step edges follow a square law between zero and a cap, the 95th percentile of every country's value on every date of the timeline for the chosen measure and window (once the timeline passes 60 days, on 60 evenly spaced dates), so the scale does not move when the day changes and a shade means the same number on every date. The cap is not frozen for good: it is computed in the browser from the published data, and each data update, about once a day, adds the newest days and recomputes it over the timeline as it then stands, so a shade can stand for a different number from one week to the next. Every measure, basis and window has its own cap, so the key changes when any of them does, and its heading names all three. The edges are rounded to what the key prints, and the map colors with the rounded edges, so a printed tier is exactly the set of values that takes its shade: whole numbers for amounts, where a step too narrow to hold a whole number is left out rather than printed twice, and two significant figures for everything else. A share of monitored output is read per 1,000 published items, since it is a small fraction of one percent almost everywhere. Values above the cap sit in the darkest, overflow step, and the legend gives the highest value. The theme counter shades its cells by the number of articles in them, on the same steps and colors, with its own cap pooled over every day of its timeline for the chosen measure and window, recomputed at each data update in the same way. Countries withheld, unreadable, not monitored or with too small a denominator have their own patterns and are never on the scale.
Stance toward government. A third grouping records whether each outlet supports its own government, is independent of it, or sides with the opposition, taken from a dated third-party source: Reporters Without Borders country pages, Freedom House country reports, the Media Ownership Monitor and Media Landscapes. An outlet's description of itself does not count. A stance is toward the government of the day, so each carries the year its source describes and a source from before a change of government, or older than 2023, is not used. State owned outlets without editorial independence are state controlled, from the sourced ownership. Outlets no such source describes are not assessed.
Political leaning and ownership. An outlet is given a political leaning and an ownership type only where a cited source supports it: Wikipedia media lists with alignment columns, Media Bias/Fact Check, RSF's Media Ownership Monitor, the Reuters Institute Digital News Report, academic studies or the outlet's own editorial statement. Leaning is judged against each country's own political spectrum, so a centre right paper in one country and in another need not hold the same positions. Not assessed means no reliable source was found; it is never inferred from an outlet's name, language or country. A state owned outlet without editorial independence is shown as state controlled: its line is the government's, not a left or right one, and the ownership source stands for it. Many outlets are still not assessed, especially small ones, so the chart describes the assessed outlets, not every outlet. Ownership separates private outlets, domestic public service broadcasters with statutory editorial independence, state funded international broadcasters, state outlets without editorial independence, and party owned outlets. State funded international broadcasters are funded by a government to reach audiences abroad, with editorial independence set by statute, charter or grant agreement: Radio Free Asia and RFE/RL (grantees of the U.S. Agency for Global Media), Deutsche Welle, RFI and France 24, the BBC World Service language services, KBS World and RNZ Pacific. They are kept apart from domestic public service broadcasters such as the BBC at home, ABC, SBS or CBC, because their funder's foreign policy is part of their remit; outlets a state runs abroad without editorial independence, such as RT and Sputnik, stay under state. Media Bias/Fact Check is a US based rater that places outlets on a United States spectrum, so a leaning outside the United States that rests on it alone is kept but shown with a caution in the chart's table, and the note under the chart counts how many leanings come from each kind of source. The chart is all time, counts active outlets only, and leaves out the global press release distribution services, which are not national newsrooms. A share of articles partly counts outlets, because a leaning with more monitored outlets publishes more, so the table beside the chart also gives the rate per outlet.
Themes. Every article that concerns China is tagged with each theme whose terms appear in its headline, its feed summary or anywhere in its body. Until themes version 2026.09.4 only the first 800 characters of the body were read, which favored whatever an article names early. An article can carry several themes, so theme counts do not add up to article counts, and one with none is Other. The theme counter carries the same three measure buttons as the map, and the two move together, so it can show themes for state origin, unchecked state sourcing or all China coverage. It has its own window and timeline, separate from the map. The One country toggle, either Country search box, or a click on a country opens the single country view in both places, and Whole world returns to every country. Its tags are keyword matches, not a reading of the article, so they should be checked by hand before a theme count is cited.
Denominators. Per outlet we monitor divides by the country's active monitored outlets. % of all its news divides by every item the country's monitored outlets put in their feeds during the window, whether or not it concerns China, and the tooltip gives the same figure per 1,000 published items; until audience ranks are recorded in the registry that is every active outlet, not the largest publications, and no share is shown for a country with fewer than five outlets, where it would describe a handful of feeds. No per capita figure is shown, because population is a poor denominator for what outlets publish; each country's resident population is still given in the population column of the CSV of the current view, and no figure is divided by it. Paywalled articles are never classified, so they are missing from every count, and from the share of China coverage, where an unread article is in neither half. They do stay in the share of monitored output denominator, which counts every item the monitored outlets published whether or not it could be read, so in a country with many paywalls that share is a lower bound. A country where paywalls pass a third of retrieved articles carries a warning on every figure shown for it.
Placements and underlying items. One wire item syndicated to thirty outlets and thirty separate items are different phenomena. Near-duplicate titles within a country are linked as one underlying item, and the tooltip, the ranked list, the country panel and the CSV export show both numbers.
Official Chinese sourcing, verification pending. Not a fifth category. A deterministic pass marks every article that cites Xinhua, CGTN, Global Times, China Daily, CCTV or People's Daily, quotes a foreign ministry spokesperson, a Chinese embassy or a named official spokesperson, or says that Chinese state media reported something. Those are the candidates for unchecked state sourcing. The same line holds the candidates for state origin that the model judges: an article with one weak signature, and one whose headline, summary or body nearly repeats a state origin item another outlet ran with its credit. When more candidates wait than the daily model ceiling allows, the ones sent are a random draw with the same fraction in every country, recorded per day so counts can be reweighted.
Gaps in collection. Feeds carry only their latest entries, so a skipped hourly run loses more items in a busy media market than in a quiet one; every poll records whether its window reached back to the previous one, and the estimated missed items are shown. Some outlets are collected only from the owner's machine, and many outlets with short feeds are polled from it as well as by the hosted runner; days on which it ran too few hours are marked on the timeline and on the countries whose outlets only it collects, so a sleeping laptop is never read as a decline. Ruleset changes and the days still carrying labels from an older ruleset are marked on the timeline too, so a discontinuity is not read as a trend. The relevance gate carries its own version, marked on the timeline as well: it decides what enters the corpus rather than how an article is labelled, and because gate decisions are recorded per item and never revisited, a change to it applies only to items discovered after that date. Items the earlier gate rejected are not re-examined, and rejected items are pruned after three days, except a random 2 percent kept permanently so the gate's recall can be checked by hand for any period.
The critical distinction is between unchecked state sourcing and independent journalism, and it is a judgement about verification, not about topic or tone. An article highly critical of China that repeats an unverified Xinhua claim is still unchecked state sourcing. An article sympathetic to China that independently confirms every claim is independent journalism. Only a factual claim central to the article counts; a position, policy, warning, threat, opinion or denial attributed to Beijing does not.
Report only: none of these changes a label, a count or a denominator. Each is provisional until its own hand check passes, and withheld if the latest completed check fails.
What limits the figures today, written from the data at the last export.
Operational detail: call caps, ruleset and gate versions, and reclassification status, written from the data at the last export.
What this can tell you is how often, in a fixed set of outlets that I can actually read, Chinese state text appears unaltered, and how often local journalists pass on official Chinese factual claims without checking them. What it cannot tell you is what happens behind paywalls I do not cross, in outlets I do not track, in sections I have not searched, in languages I excluded, or in the ordinary gaps of an hourly poll of RSS feeds. A country that looks empty here may simply be a country I have not yet found a working feed for. Read the hatching before you read the color.