How Marble works
Marble shows you what everyone’s talking about: every story from newsrooms across the left, centre and right, grouped so you see each story once, with who had it first, how it spread and who isn’t covering it.
What we do
We read the newsrooms. Every few minutes we read the public feeds of newsrooms and official sources: wire services, broadsheets, tabloids, broadcasters, specialist sites, and the leagues, studios and companies themselves. Each headline links to the original article.
We group reports into stories. Reports about the same event become one story, so you see it once with everyone who covered it. We count newsrooms, not links: several outlets repeating one claim is still one claim.
We measure how it spread. Every story shows who had it first and by how long, how fast other newsrooms picked it up, and how that compares with the section’s other stories this week. Who’s first ranks the newsrooms that break stories.
We show where a story is in its life. From the times of its reports: emerging, picking up speed, peaking or cooling, and how many reports it drew each day this week. A story with too little to measure gets no label rather than a guess.
We watch what people search for. Trending searches shows what Australia and the US are searching Google for right now, and how long each search has been trending. It’s kept apart from coverage: a search can be huge with no newsroom on it, and a story can be everywhere that nobody is looking up.
We show who’s covering it. News stories show which side of the political spectrum is covering them, using AllSides’s ratings of each newsroom. Blindspots lists stories only one side is running.
We follow the names. People, teams, companies, shows and games are picked out of the headlines and checked against Wikipedia, so a name means one thing. Storylines link the developments of a story that runs for days.
We pick five stories a day. Today’s five aren’t simply the stories with the most links. Each counts its newsrooms from the last day (older coverage counts for a quarter, and 40 newsrooms isn’t four times 10), then weighs how widely it reaches: whether left, centre and right newsrooms are all on it, how many countries are reporting it, how much of the coverage comes from Australia and the US, and whether an official source has spoken. Opinion, reviews, live blogs and stories carried mostly by tabloids count for less, and the five are spread across sections. Each story’s headline is the one that best matches the rest of its coverage, not one newsroom’s take.
We explain the biggest stories. Some stories carry an explainer: what happened, what the reports establish, what isn’t known yet and what to watch. Every line links to the reports it comes from, and an editor checks it before it appears.
We score our own calls. When we call a story emerging or a name rising, the call is logged at that moment and checked 48 hours later. Hits and misses stay on the Track record.
What we don’t do
We don’t report the news. We don’t have reporters, and we don’t rewrite other newsrooms’ stories. Read the original for the full report: it’s always one click away.
We don’t confirm claims. A story’s label tells you how many independent newsrooms have reported it, and whether it came from an official source. That’s a measure of coverage, not proof that something is true.
We don’t filter by viewpoint. Every story from every feed we read is shown, whatever it’s about and whoever it suits. We add newsrooms in balance across the spectrum.
We don’t rate newsrooms. Political lean comes from AllSides, not from us. A newsroom they haven’t rated is shown as not rated, never guessed.
The labels on a story
Official. The report comes from the source itself: a league, studio, publisher, company or government body.
2 newsrooms, 3 newsrooms… That many separate newsrooms have reported it. Sister mastheads that run one shared story (Nine’s Herald, Age, Brisbane Times and WAtoday; Seven West’s West Australian, PerthNow and 7NEWS) count as one newsroom, and so does the same headline word for word from different outlets. A story says which outlets it counted together.
One source so far. Only one newsroom has it. It may be an exclusive, or it may not hold up.
Where a story is in its life
Worked out from the times its reports were published, counted every few minutes. A story needs four newsrooms and three hours of reports before it gets one of these.
Emerging. Under six hours old and still picking up reports.
Accelerating. At least four reports in the last three hours, and twice as many as in the three before.
Rising. More reports in the last three hours than in the three before.
Peaking. Close to its busiest stretch so far, but no longer growing.
Cooling. Well below its busiest stretch, six hours or more in.
Back. Quiet for a day or more, now picking up new reports.
Where a story is heading
A projection says where a story is likely to be in the next 1 to 24 hours, measured in newsrooms. It comes from every story of the last eight days that was in the same position: how many newsrooms joined it in the last three hours, and how old it was. The likely range is where the middle 60% of those stories ended up; the most likely figure is the middle one. Nothing is guessed, and a position without enough history gets no projection.
We check it the same way every ten minutes: for moments in the last week, we project six hours ahead using only what was known then, and count how often the real figure landed inside the range. That figure is printed beside the projections. The same check found that ordering stories by projection was no better than ordering them by how fast newsrooms are joining, so lists are ordered by pace and the projection gives the range.
Categories
A category is a fixed list of English Wikipedia articles (Generative AI, for example, is ChatGPT, OpenAI, Anthropic, Claude, Gemini, Large language model, Generative artificial intelligence and Vibe coding). Its attention is the total weekly views of those articles by people, from Wikimedia, over two years. Growth compares the last 13 weeks with the 13 before and with the same 13 weeks a year earlier. The projection carries the last 26 weeks’ growth rate forward 13 weeks; its range is the middle 60% of how far off that method was at every past week we can check, across all categories, and we print how it compared with assuming nothing changes. It measures attention, not money or users, and it can’t see what’s coming that isn’t already in the numbers.
Trend explorer
The long view of the names in the week’s news: weekly views of each one’s English Wikipedia article over the last year (people only, from Wikimedia), beside how many of our stories named it. The labels go by three-month stretches, so a one-week jump is never called a trend. Rising: the last 13 weeks are at least one and a half times the 26 weeks before, and the last month holds it. Fading: the last 13 weeks are at most two thirds of the 26 before. Spiking this week: the last week is four times the year’s usual, with no longer rise behind it. One-off spike: one week earlier in the year at five times the usual, since settled. Steady: none of these.
How the coverage changed
On bigger stories we cut the reports into stretches of time and show two things for each. New angles: words several newsrooms started using then, having rarely used them before. Frames: what the headlines were doing, decided by fixed cue words, so the same headline always gets the same frame. A stretch gets a frame when two or more newsrooms and a quarter of its headlines share it. Every one links to the headlines behind it.
Denials and walk-backs. Headlines with words like denies, rejects, declines, walks back, only kidding.
Courts and police. Headlines with words like court, sues, charged, arrested, trial.
Questions and scrutiny. Headlines with words like probe, investigation, inquiry, calls for, resign.
Reactions. Headlines with words like responds, slams, hits back, defends, backlash.
Money and markets. Headlines with words like shares, investors, billion, box office, earnings.
The people involved. Headlines with words like family, tributes, fighting for life, mourns.
Explaining it. Headlines with words like what we know, explained, why, timeline, analysis.
Two headlines
On a story with three or more newsrooms, we show the two headlines that put it most differently: the pair that shares the fewest words, both clearly about the same event (each uses at least two of the words most of the story’s headlines share). Explainers, questions and live blogs are left out. When pairs tie, two newsrooms on different sides of the spectrum are preferred, but only once their words already differ. The highlighter marks framing words one headline uses and the other doesn’t, from a fixed list of words that judge or colour what happened: verdicts (escapes, fails, blocks), conflict (slams, rebels), emotion and alarm (humiliating, chaos), hedging (alleged, claims, insists) and spin (bombshell, sham). Names, numbers and job titles are never highlighted, and often nothing is.
Headlines, photos and credit
Headlines are the newsroom’s. We show each outlet’s own headline and opening line, credited, with a link to its article.
Photos are the newsroom’s, or free to use. A story shows its outlet’s own picture, loaded from the outlet and credited to it. When there isn’t one, we use a freely licensed photo from Wikimedia Commons and name the photographer and licence.
Background is from Wikipedia. The short description on a name’s page is the opening of its Wikipedia article, credited under CC BY-SA 4.0.
Every word is the newsroom’s. Headlines and summaries are the outlets’ own, credited and linked. Nothing on the site is written by AI: see our AI policy.
When we get it wrong
Grouping is done by rules, and rules make mistakes: two stories merged that shouldn’t be, one story split in two, a name matched to the wrong person. The rules check their own work (a story is filed under the section most of its reports come from, and an official post is only credited with a story it’s actually about), and our editors fix what they miss. A fix stays fixed: a story split by hand isn’t merged back. Tell us when you spot one. If a newsroom’s own report is wrong, that’s for the newsroom to correct: we show its headline as it published it, and link to the original.
Where the news comes from
Public feeds from newsrooms and official sources, focused on Australia and the United States, with newsrooms from the UK, Europe, Asia and beyond. In News they span the left, centre and right: the rated newsrooms are listed on Blindspots. “Official” on a story means it came first-hand from the league, studio, publisher or company itself.
Marble is independent, curated in Melbourne, Australia and San Francisco. Questions, corrections and partnerships: contact us. See also Privacy and Terms.