Counting voices, not trusting labels: how LSN News measures it

Series: How minds are moved · Fake news, fact-checks and how to check for yourself · 6 of 7

On 12 November, between 15:45 and 21:45 UTC, 362 websites published the same headline, word for word: “iHeartRewind 2026: Check Out iHeartRadio’s Top Artists & Songs Of The Year”. Every one of them is a radio station site of the same company. On LSN News, my news engine, the story shows 384 outlets and 5 independent voices. A search engine would have shown you 384 results. The honest number is 5. This post explains, plainly, how LSN News gets from the first number to the second, and why I chose to count voices instead of labelling sources as “trusted” or “untrusted”.

Contents · 10 sections

TL;DR #

• LSN News shows two numbers for every story: how many outlets carried it, and how many independent voices those outlets represent.
• A voice is one piece of writing. Sites that run the identical headline count as one voice. Every other site counts as one.
• A second layer finds outlet networks: sites that keep publishing the same text over a week. On 13 November there were 194 networks covering 1,597 sites.
• In the last three days, 12% of stories carried by two or more outlets had only one voice.
• I do not label sources as reliable or unreliable. Labels are someone’s verdict; a count is something you can check. A count says nothing about truth, and I say so on every story page.

Two numbers instead of one #

When you hear “everybody is reporting it”, you are usually hearing a count of websites. That is the number search results, feeds and aggregators show you. It is also the easiest number to inflate, often without anyone intending to.

Most local news today is produced centrally and published many times. A newspaper group writes one piece and puts it on fifty mastheads. A radio company pushes one article to every station site. An agency sells the same copy to thousands of customers. None of that is wrong. It just means that “50 outlets” can mean one newsroom.

So every story on LSN News carries two numbers: outlets, and independent voices. The gap between them is the interesting part.

How a voice is counted #

The rule is deliberately simple, so that anyone can reproduce it:

1. Take every article in the story.
2. Compare headlines, ignoring upper and lower case. If two or more different sites ran exactly the same headline, that headline is one voice, however many sites carried it.
3. Every site that wrote its own headline is one voice.
4. A story never has more voices than outlets. (A site that ran its own piece and also a syndicated copy would otherwise count twice.)

That is all. No model, no reputation score, no secret list.

Two real stories from this week show the range:

• The Kraków shooting on 7 November: 2,168 outlets, 1,900 voices, in 60 languages. Almost every newsroom wrote its own copy. That is what broad, independent coverage looks like.
• iHeartRewind: 384 outlets, 5 voices. One company promoting its own year-end list through its own stations.

The pattern is not rare. In the same three days, one helicopter emergency landing headline ran word for word on 562 outlets, and a retail bankruptcy headline on 491. Those are mostly radio-network and wire copies. Whoever writes the copy decides what hundreds of “local” sites say.

The second layer: outlet networks #

Identical headlines catch syndication inside one story. But I also wanted to know which sites habitually publish the same text, across all stories. So once a day LSN News looks at the last seven days and asks, for every pair of sites: how many headlines do you share?

• If two sites share at least 8 headlines, and the shared headlines are at least a fifth of the larger site’s output, they are sister sites: one newsroom or one owner feeding several mastheads. Sister sites that connect to each other form a network.
• A headline that ran on more than 60 sites is ignored for this purpose. That is agency copy that everyone runs, and it says nothing about who is related to whom.
• Smaller overlaps are kept as “shares headlines with” on each site’s page, but never merged into networks. When I tried merging them, agency copy chained 1,038 unrelated sites into one giant group. That taught me to keep the rule strict.

On 13 November the result was 194 networks covering 1,597 sites. The largest has 196 sites, and Wikidata lists 174 of them as stations of one radio company, Townsquare Media. Next come a 113-site family of city and country news sites, a 68-title Australian regional chain, and a group of US papers whose members Wikidata attributes to several owners, including MediaNews Group and Tribune Publishing.

The page says these sites “publish the same text as” each other. It never says “are owned by”. Two sites that buy the same feed look exactly like sister sites. Ownership is a fact you check elsewhere; co-publication is what I can measure.

What the numbers say this week #

Across all stories active in the last 72 hours on 13 November:

• 64,406 stories were carried by two or more outlets. Of those, 7,780 (12%) had a single voice: every outlet ran the same piece.
• Of 5,327 stories carried by ten or more outlets, 1,690 (about a third) had fewer voices than half their outlets.

In other words, for one big story in three, the outlet count overstates the number of independent newsrooms by more than double.

Why I count, instead of labelling sources #

There is a whole industry that rates sources: browser extensions that put a green or red badge next to a site, indexes that score “disinformation risk” for advertisers, platform labels such as “state-affiliated media”. I thought hard about adding something like that. I decided against it, for three reasons.

First, a label is a verdict, and whoever writes it has a point of view. Two documented cases from opposite directions:

• In December 2022 the Global Disinformation Index, a non-profit that sells risk ratings to the advertising industry, rated 69 US news sites. Its least risky group included NPR, AP and The New York Times; its riskiest included Reason, RealClearPolitics and the New York Post. Critics, led by the Washington Examiner, reported that GDI had received US government-linked grants and that its lists were used by ad platforms; GDI says its index does not assess partisanship. Whatever you conclude, a rating used to steer advertising money is a powerful thing to hand to anyone.
• In April 2023 Twitter labelled NPR “US state-affiliated media”, although its own policy had named NPR as an example of a publicly funded outlet that was not state-affiliated. NPR said the label was untrue and stopped posting on its 52 official accounts. The label was soon changed to “government-funded media”, and later dropped. Here the label came from a platform’s owner, and the target was an outlet the first case had rated least risky.

Second, the evidence that labels help is thin. A 2022 study in Science Advances got more than 3,000 people to install a popular source-rating extension and followed their browsing. On average the labels did not measurably shift people to more reliable sources, and did not reduce belief in widely shared false claims. There was a small improvement among the heaviest consumers of low-quality news.

Third, a count can be checked. Anyone can open a story on LSN News, read the headlines, and see whether 52 outlets really are one piece. A badge you have to take on trust.

To be transparent about my own system: the source registry has a field for a reliability score. Every one of my 231 curated sources carries the same value in it, the neutral midpoint, and the voice count never reads it. Sources do carry a type (official, state, public broadcaster, wire, fact-check and so on), and the story page shows how many outlets of each type covered it. That is a description, not a weight. Fact-checkers count exactly like any other outlet: one fact-check is one voice.

What counting cannot do #

I want to be as clear about the limits as about the method.

• Voices are not truth. Many independent newsrooms can repeat the same wrong paraphrase, as I showed in an earlier post about the Microsoft outage. Every story page says it: corroboration is not a verdict.
• One voice can be right. A local scoop starts as one outlet. Single-voice stories are usually local, not suspicious.
• Syndication is undercounted. Two outlets that rewrite the same agency copy with different headlines count as two voices.
• A careful coordinated campaign, rewriting every copy, would pass. Catching that needs reading, not counting.
• The method only sees headlines. Grouping by text inside the article would be stronger, and is on my list.

How to notice it #

• When something “is everywhere”, ask how many different pieces of writing you have actually seen.
• Open three versions from different-looking sites. Same headline and first paragraph means one voice.
• Look at the footer: a group name, “originally published by”, the same byline.
• Treat a trust badge as one organisation’s opinion, and ask who made it and who pays for it.
• Count voices first, then look for a primary source among them.

Frequently asked questions #

What is an independent voice on LSN News? #

One distinct piece of writing. Sites that ran the identical headline count as one voice; every other site counts as one; a story never has more voices than outlets.

What is an outlet network? #

A group of sites that kept publishing the same headlines over seven days, enough to make up a large part of their output. It shows co-publication, not ownership.

Does LSN News rate sources as reliable or unreliable? #

No. Every curated source has the same neutral reliability value, and the voice count ignores it. Sources have a type (official, wire, fact-check…), shown as a description.

Are fact-checkers treated differently? #

No. A fact-check is one voice, like any other article.

Does a high voice count mean a story is true? #

No. It means many newsrooms wrote it independently. They can still all be wrong.

The lesson #

“384 sources” and “5 voices” describe the same story. The first is what the web shows you; the second is closer to how many people actually wrote something. I chose to count because a count is a fact anyone can check, while a trust label is a judgement someone asks you to accept. Counting will not tell you what is true. It tells you how many separate minds stand behind a claim, and that is a good first question to ask before deciding what you believe.