The first month, in numbers

2026-10-07 by A River of News | Tags: findings, data

We have collected headlines for 30 full days, from September 8 to October 7, 2026 (UTC). This is the first of a monthly series of plain findings from that data. Every number below can be checked: the script that produced them, scripts/monthly-findings.py, is in the code, and the headlines themselves are on the open data page.

The month at a glance

• 101,150 headlines from 92 publications, an average of 3,372 a day.
• Weekdays are busier: 3,734 headlines a day on average, against 2,377 on weekends, 36% fewer.
• The busiest day was Wednesday, October 7 (4,442 headlines). The quietest full day was Friday, September 18 (1,714). We have not checked whether that was a quiet news day or a gap in our own collection, so please do not read anything into it.
• 2,706 stories were covered by three or more publications. The middle story had 4 publications. The largest had 54.

Stories take a day to spread, not an hour

On the stories where one publication was clearly first (2,561 of them), the second publication followed a median of 159 minutes later. Only 32% were matched within an hour. The last publication to join a story arrived a median of 24.8 hours after the first.

How fast depends on the beat. For beats with at least 100 stories:

Entertainment and sports stories are matched quickly. World, business and tech stories take about four hours to get a second publication. A beat's category is the most common category of the headlines in the story.

Who is first depends on the beat

Out of the stories each publication covered, these were first most often:

Specialist outlets lead because on their beat few others cover the same thing. This is not a ranking of quality. A publication that is first often on a narrow beat is doing something different from one that covers everything and is first less often. The full table is on the first to report page.

Almost nobody copies a headline

We looked for headlines that repeat an earlier headline in the same story word for word. There were 103 out of 41,039, or 0.25%. Of the 47 publications with at least 200 headlines in stories, 27 had none. The highest was The Japan Times, with 29 of 670 (4.3%).

So on this list of publications, wire copy under identical headlines is rare. Publications write their own headline even when they report the same thing. That says nothing about whether the articles beneath are original.

What we cannot say yet: bias

We can sort publications by who shares their links, using the scores from a published study (see how we use it and the perspective spectrum). Of 886 news stories, 753 had three or more publications with a score. Of those, 700 included at least one publication whose audience leans Democratic, and 313 included at least one whose audience leans Republican. In 118 stories, three or more Democratic-leaning publications covered it and no Republican-leaning one did. The reverse never happened.

We do not read this as a finding about the press. Of the 28 scored publications that carry news, 12 have a Democratic-leaning audience, 12 a mixed audience and only 4 a Republican-leaning one. With that mix, a story is far more likely to be missed by the small group. The numbers describe our list of publications, not the news. Until the list is more balanced, we will not publish a conclusion from them.

How to read these numbers

• A "story" is a group of headlines that NewsMarkets, or our own matching, judged to be about the same thing. The largest groups, with 40 or more publications, are mostly topics in the AI industry and some are broader than one event.
• "First" means the earliest time a publication's feed gave for its article. A feed can be late, so a publication can look later than it was.
• We only count stories with at least three publications. Stories covered by one or two outlets are in the river but not in these figures.
• The list is the 92 publications we collect from. It is not the whole press.

Audience-lean scores are from Robertson, Jiang, Joseph, Friedland, Lazer and Wilson (2018), Auditing Partisan Audience Bias within Google Search, Proceedings of the ACM on Human-Computer Interaction 2(CSCW), article 148, used with the authors' permission.