How a news story travels: anatomy of fourteen days of press
By Kike Gandia · Research · 15 min read
Over fourteen days in September 2026 we logged every piece published by 162 Spanish-language outlets, with the timestamp each one puts on its own channel. That comes to 124,802 headlines.
With that material you can put numbers on a question that is almost always answered with gut feeling: when two outlets publish the same story, how much time passes between them?
The answer is that there isn't one answer; there are two. There is one set of outlet pairs whose stories land in the same minute 84.8% of the time, and another set where that overlap drops to 1.9% and the typical delay is 171 minutes. Two different statistical distributions living side by side on the same newsstand.
This article measures that gap and everything that follows from it: how much of the daily news agenda is shared, which types of outlet get there first, which topics travel fastest and how long a big story takes to work its way through the system.
One clarification before we start, because it shapes everything that follows: outlets are not named. Each one is classified by its profile — coverage area, volume and publishing behavior — and every figure is an aggregate of those categories. What we are studying is the system, not any particular newsroom.
What we measured
The corpus is the set of pieces published between September 4 and 17, 2026 by 162 Spanish-language outlets, captured from their public channels. For each piece we keep the outlet, the headline, the URL, the publication time declared by the outlet and a topic classification assigned from its own text.
Report corpus: 124,802 headlines from 162 Spanish-language outlets, fourteen full days from September 4 to 17, 2026, with minute-level timestamp precision and a daily volume of 7,000 to 12,000 headlines from 110 to 140 active outlets. Breakdown by outlet category, with number of outlets, headlines, share of the corpus and median pieces per outlet: high-volume pan-Hispanic, 16 outlets, 47,000 headlines, 37.7%, median 1,107; regional in a synchronized bloc, 20 outlets, 30,530 headlines, 24.5%, median 1,322; national general-interest, 11 outlets, 18,262 headlines, 14.6%, median 1,348; specialist or trade, 76 outlets, 12,233 headlines, 9.8%, median 32; independent regional, 9 outlets, 10,117 headlines, 8.1%, median 870; international Spanish-language, 9 outlets, 4,723 headlines, 3.8%, median 443; digital-native or alternative, 21 outlets, 1,937 headlines, 1.6%, median 55.
The table already tells a story on its own. The 76 specialist outlets add up to fewer headlines than the 16 pan-Hispanic ones, and each publishes a median of 32 pieces in fourteen days, against 1,107 for a pan-Hispanic outlet. The number of outlets and the volume they publish are very different things: 47% of the outlets in the corpus contribute 11% of the pieces.
That asymmetry matters for everything that follows. Any measure that counts raw events — who publishes first, who echoes the most — will be dominated by a handful of high-volume outlets unless it is corrected. Further down, it is.
How much of the agenda is shared
Two pieces from different outlets belong to the same story if their headlines exceed a lexical similarity threshold and are published less than 48 hours apart. Across the 124,802 headlines, that yields 8,961 shared stories.
Shared agenda: of the 124,802 headlines, 26,916 (21.6%) belong to a story that another outlet also published, and 97,886 (78.4%) have no equivalent in any other outlet in the corpus. Size distribution of the 8,961 shared stories by number of outlets publishing them: 2 outlets, 5,676 stories; 3 outlets, 1,666; 4 outlets, 745; 5 outlets, 403; 6 outlets, 162; 7 outlets, 92; 8 outlets, 54; 9 outlets, 33; 10 or more outlets, 130. The median is 2 outlets and the 90th percentile is 4; only 217 stories reach eight outlets or more in fourteen days. Share of its own output that each profile shares with another outlet: regional in a synchronized bloc 33.1%, independent regional 23.2%, national general-interest 22.2%, international Spanish-language 21.3%, high-volume pan-Hispanic 17.4%, digital-native or alternative 10.9%, specialist or trade 8.3%.
The result runs against the intuition that the press all publishes the same thing: four out of five pieces have no equivalent in any other outlet in the corpus. And the stories that are shared are mostly small — the median is two outlets.
What is interesting is the ranking in the last table. A regional outlet that is part of a bloc shares four times as much of its output as a specialist outlet. That isn't explained by them covering the same things: it is explained by how they publish, which is what the next section measures.
Two publishing regimes
We took each shared story, identified which outlet published first and measured how long each of the others took. When the cases are split by whether both outlets belong to the same synchronized publishing bloc, two curves emerge that look nothing alike.
Two distributions of the delay between the first and second publication of the same story, with the three cumulative cut-offs measured. Within the same bloc (4,374 cases): 78.3% of second publications are out within two minutes, 79.9% within fifteen and 84.2% within the hour; median delay 0 minutes and 90th percentile 190 minutes. Between different blocs (12,269 cases): 1.8% within two minutes, 10.8% within fifteen and 29.2% within the hour; median delay 171 minutes and 90th percentile 1,280 minutes, twenty-one hours. The first cut-off separates the two populations by a factor of forty-three. The between-bloc distribution fits a lognormal with a median of 154 minutes, the characteristic shape of a diffusion process in which each participant picks up the story at its own pace, and its 1st percentile on the fast side is 3.1 minutes.
The distribution for the synchronized regime is a spike: almost eight in ten cases have already published within two minutes, and from there to the one-hour mark it barely moves six points. A median of zero minutes does not describe a fast process: it describes a simultaneous event.
The other one does have a shape. It climbs slowly — 1.8% at two minutes, 10.8% at fifteen, 29.2% at one hour — and trails a long tail out to the twenty-one hours of the 90th percentile. It fits a lognormal distribution with a median of 154 minutes, the characteristic shape of a diffusion process in which each participant picks up the same story at its own pace: some hear about it earlier, some have a busy newsroom, some see it the next day.
That lognormal also has an operational use. Its 1st percentile on the fast side is 3.1 minutes, so an overlap below that value between outlets from different blocs is an event the reference distribution produces once in a hundred times. Put another way: if two outlets with no known connection publish the same thing less than three minutes apart, that isn't fast journalism. It is a data point that needs explaining.
A median of zero minutes isn't speed. It's a single decision being executed in several places at once.
The synchronized publishing blocs
The blocs are not defined in advance: they are detected by measurement. And the procedure uses no external data about the outlets — no ownership, no media group, no declared agreements — only their publication times.
The procedure, in two steps
- Two outlets are linked if they have overlapped on 20 or more stories and published less than two minutes apart on at least 60% of them.
- The connected components of that graph are identified.
Across 162 outlets, the procedure identifies three blocs of 8, 7 and 5 outlets. Twenty outlets in total, all of them regional; the other 142 stand alone.
Detection of synchronized publishing blocs: three connected components of 8, 7 and 5 outlets, all regional, and 142 standalone outlets with no synchronized partner. Of the 9,541 outlet pairs where both are in the same bloc, 84.8% publish less than two minutes apart; of the 31,701 pairs from different blocs or standalone outlets, only 1.9%. Of the 8,682 simultaneous co-publications recorded over fourteen days, 8,090 — 93.2% — happen inside one of those three blocs. The 30,530 pieces from bloc outlets make up 24.5% of the corpus.
Those twenty outlets account for almost all the simultaneity in the system: of the 8,682 simultaneous co-publications over fourteen days, 8,090 — 93.2% — happen inside one of those three blocs.
The result has a clear practical consequence. Simultaneity is not spread across the system: it is concentrated in 12% of the outlets. For the rest of the ecosystem, publishing in the same minute as another outlet is a rare event, with a frequency of 1.9%. And 24.5% of the corpus reaches readers through a channel in which the decision to publish is made once and executed across several outlets at the same time.
It is worth spelling out what this does not prove. Synchrony of this kind is consistent with several perfectly ordinary explanations: a media group with a central newsroom, a content syndication agreement, or a system that automatically publishes wire agency copy. What the data establishes is the structure, not the intent.
Who gets there first
Counting raw first publications rewards whoever publishes the most, so the measure is taken against a null model. If a story is covered by k outlets and the order of arrival were random, each would have a 1/k probability of being first. Summing that probability over every story an outlet takes part in gives each category a number of first publications expected by chance.
The index is observed ÷ expected. A value of 1.00 is indistinguishable from chance. When several outlets publish within the same two-minute window, the credit is split among all of them.
First-to-publish index by category, calculated as observed first publications divided by those expected by chance, where 1.00 is indistinguishable from chance. High-volume pan-Hispanic: 7,318 appearances, 3,572 firsts, 2,939 expected, index 1.22, median delay of 205 minutes when not first. Specialist or trade: 977 appearances, 410 firsts, 411 expected, index 1.00, delay 196 minutes. National general-interest: 3,920 appearances, 1,203 firsts, 1,290 expected, index 0.93, delay 121 minutes. Regional in a synchronized bloc: 9,912 appearances, 2,893 firsts, 3,127 expected, index 0.93, delay 160 minutes. International Spanish-language: 961 appearances, 264 firsts, 349 expected, index 0.76, delay 532 minutes. Independent regional: 2,304 appearances, 565 firsts, 754 expected, index 0.75, delay 142 minutes. Digital-native or alternative: 212 appearances, 54 firsts, 91 expected, index 0.59, delay 893 minutes.
Three takeaways, all of which can be read straight off the chart.
Volume is the main timing advantage. The pan-Hispanic category is the only one clearly above chance: it gets there first 22% more often than it should. These are also the outlets with the most pieces per outlet in the corpus. Publishing a lot, about everything, raises the odds of being out in front on any given story — a less romantic finding than the exclusive, but it is the one the data produces.
The papers of record run with the pack. The national general-interest category has an index of 0.93: across the daily agenda as a whole, its order of arrival is essentially what chance would produce. What they do have is the shortest delay in the corpus when they don't arrive first — 121 minutes — in other words, they react faster than anyone else. They don't set the agenda; they follow it quickly.
Some profiles are on the receiving end. The international and digital-native categories combine low indexes with median delays of 9 and 15 hours. Their timing pattern is one of picking up the agenda, not setting it — to be expected in the first case because of time-zone differences and in the second because of newsroom size.
Which topics travel, and how fast
Each piece carries a topic classification assigned from its own text. Cross-referenced with the shared stories, it produces two independent measures per topic: how much it gets picked up and how long that takes. And the two don't move together, which is the interesting part.
Pickup and speed by topic, with headlines, share picked up, outlets per story and median delay. International: 8,601 headlines, 32.6% picked up, 2.89 outlets per story, 129 minutes. Politics: 21,028 headlines, 25.1%, 2.90 outlets, 63 minutes. Health: 2,953 headlines, 22.1%, 2.75 outlets, 93 minutes. Sports: 10,197 headlines, 21.8%, 2.99 outlets, 86 minutes. Society: 67,386 headlines, 20.2%, 2.84 outlets, 100 minutes. Technology: 6,538 headlines, 17.4%, 2.81 outlets, 105 minutes. Economy: 8,099 headlines, 14.4%, 2.70 outlets, 98 minutes. Pickup ranges from 14.4% to 32.6%, so a piece's topic more than doubles its likelihood of also appearing in another outlet.
International is the most picked-up topic and one of the slowest: a third of its pieces appear in several outlets, but they take a median of 129 minutes to spread. Lots of overlap, little urgency — the profile of a story everyone ends up covering and nobody considers urgent.
Politics is the fastest topic in the corpus, at 63 minutes: less than half the time of international, despite a similar pickup rate. It is the only topic where the median interval drops below an hour.
Economy is the least shared, at 14.4%, and also the topic with the fewest outlets per story (2.70). When it does get picked up, it travels in narrower circles.
The rhythm of the day and the path of a story
Counting, for each hour of the day, how many follow-up publications there are for each story that broke in that same hour, the day splits in two. From the small hours until midday the ratio holds at around 1.2: almost as many stories break as get picked up. From 14:00 UTC it rises and stays high for the rest of the day, peaking at 1.95 at 15:00. The day's agenda is set in the morning; the afternoon distributes it.
The second question is about shape: when a story travels far, how does it get there? For the 217 that reached eight outlets or more, the path is not linear.
Rhythm of the day and the path of a big story. From the small hours until midday the ratio of follow-ups per newly broken story holds at around 1.2; from 14:00 UTC it rises and stays high for the rest of the day, peaking at 1.95 at 15:00. For the 217 stories that reached eight outlets or more, the median time to reach 25% of the outlets that will eventually publish it is 0.9 hours; 50%, 2.5 hours; 75%, 5.9 hours; and 100%, 15.4 hours. In other words, the first half of the path is covered in the first two and a half hours and the second half needs almost thirteen more.
A big story covers half its path in the first two and a half hours, and needs another thirteen to complete the second half. The last quarter of outlets takes almost ten hours longer than the third: nearly all the noise happens on the first afternoon, and after that it trickles on into the next day.
How a report like this is produced
This article is a small sample of what 0data, our data processing and cross-referencing platform, does. The full report isn't written by hand: it is generated, and the human work lies in deciding which questions deserve an answer and in checking that the answers hold up.
The five layers underneath
- Continuous ingestion. Collection from each source's public channels, with retries and gap tracking: if an outlet stops responding for two hours, that gets logged and is not mistaken for “published nothing”.
- Normalization. Everything moves into a single schema: outlet, UTC timestamp to the minute, headline, canonical URL. Without this step nothing can be compared, because every source declares the time in its own way.
- Similarity clustering. The cross-referencing step that turns 124,802 loose headlines into 8,961 stories. It is the layer that most determines the result and the one that gets the most manual review.
- Cross-referencing and correlation. This is where the blocs come out — connected components on a temporal-overlap graph — along with the null-model indexes and the distribution fits. None of this is in the raw data: it emerges when the data is cross-referenced against itself.
- Reference model. Fitting the lognormal isn't decoration: it is what lets us say that 3.1 minutes is the 1st percentile and turn “this looks odd” into “this happens once in a hundred times”.
That same engine powers CyberPanorama, 0data's cybersecurity module, which cross-references more than six hundred sources in real time to keep a live threat model. And it powers the KOMPLIANCE agents, which read that context before deciding what to test in an audit. The raw material changes; the machinery stays the same: ingestion, normalization, clustering, cross-referencing and a reference model to measure against.
What this is good for beyond an article
A press report makes a good showcase because the data is public and verifiable, but the real value of the machinery lies elsewhere.
- Coordination detection. The 3.1-minute threshold is reusable: it works for any publishing ecosystem where organic spread has to be told apart from synchronized spread. Social networks, forums, messaging channels.
- Brand and mention monitoring. Knowing who is talking about you is the easy part. What is useful is knowing whether that mention is going to travel: the topic, the profile of the outlet that publishes it first and the time of day change its expected path by more than a factor of two.
- Response window. If a big story covers half its path in two and a half hours, the window to respond to a reputational crisis isn't “the same day”: it is that first afternoon.
- Context for threat intelligence. The same pipeline that cross-references headlines also cross-references leaks, exposed credentials, lookalike domains and threat actor activity. It is what sits behind our threat intelligence services.
Limits of what we measured
This is the section missing from almost every data report, and the one that decides whether the rest deserves trust.
- The time is the one the outlet declares, not the actual publication time. An outlet that updates the timestamp when it edits a piece shows up later than it really was.
- Lexical similarity groups similar headlines, not facts. Two pieces covering the same event with very different headlines are not grouped, so the 21.6% shared agenda is a floor, not a ceiling.
- Fourteen days are fourteen days. There is no seasonality, no August, no election campaign, no major unforeseen event that reshuffles the agenda. Repeating the measurement in another window is the only way to know what is structural.
- Synchrony does not imply a single newsroom. It implies a single publishing decision, which may come from a media group, a syndication agreement or a wire agency with automatic publishing.
- The corpus is public channels. Whatever an outlet doesn't publish on its channel isn't there, and 162 outlets are not all Spanish-language media.
None of these limits affects the main finding, and it is worth explaining why: the gap between 84.8% and 1.9% is a factor of forty-five. The timestamp and clustering biases act on both populations at once, so they can shift the absolute figures, but on their own they cannot produce a separation of that size.
A report on your sector
What you have just read covers fourteen days of general-interest press because it was the easiest public corpus to verify. The same machinery can be pointed at whatever you need: your sector, your competitors, the channels where people talk about you, or the ecosystem where you want to spot a campaign before it becomes obvious.
If you'd like to see what your own data would show, half an hour is enough to find out whether the question on your mind can be measured — which is the part that saves the most time.
Related service
threat intelligence and OSINT services
Topics
#0data #OSINT #Threat-Intelligence #Data-Analysis #Media #Disinformation