Grouping news sites by owner reveals more concentrated markets
Who Owns Digital News, and Who Gets Cited: Measuring Media Concentration Across 77 Countries and Seven AI Systems · arXiv:2610.40164
In plain words
The paper introduces a dataset linking digital news websites to their owners across countries. The authors pair those ownership records with estimates of website visits, then compare how concentrated visits are when sites are counted alone or grouped under the same owner.
They also compare traffic with the sources named in AI answers. In five systems’ citation data, citations were more concentrated than visits on the same outlets, though that does not show AI answers changed where people went.
The research
If a country has many news sites, do readers really have many independent sources, or do those sites belong to a few owners? Cross-country comparisons have been hard because regulators collect ownership and concentration data in different ways.
The same gap also makes it difficult to tell whether AI answers cite sources as broadly as audiences visit them.
The authors build a cross-country ownership dataset and use it to compare concentration in news traffic with concentration in AI citations.
From ownership records linked to traffic estimates, the authors measured whether traffic clustered among a few outlets, then among their owners. Across 77 countries, the median owner-level score was 1.30 times the outlet-level score.
Comparing citations with visits over the same outlets, they found citation concentration higher in five AI systems, both before and after grouping sites by owner. This supports a narrower-source conclusion for citations, not a claim about where people went.
Ownership labels cover 70.1% of traffic and only 28.8% of domains, with thinner coverage associated with higher measured concentration. The AI data are snapshots, so citation patterns do not establish that traffic will shift to follow them.
What the evidence shows
- A cross-country ownership dataset
The paper’s cross-national digital news ownership dataset links websites to their owners and pairs those records with traffic estimates. It covers 9,761 domains, with concentration estimates for 77 countries.
Because the dataset includes domains with identified owners after filtering, it offers a basis for comparing covered news markets, not a census of every outlet.
- Ownership changes the picture
Counting each website separately can make a market look less concentrated when several sites share an owner. The authors calculate traffic concentration both by outlet and after grouping outlets by owner.
Across 77 countries, the median owner-level measure was 1.30 times the outlet-level measure. In a separate comparison across 55 countries, journalists’ survey ratings of owner influence also tended to be higher where traffic concentration was higher; this association does not establish cause.
- Citations cluster more than visits
The authors then ask whether AI systems cite news sources as broadly as audiences visit them. For each country, they compare citation concentration with traffic concentration over the same outlets.
Across five systems, the country-level median citation-to-traffic ratios exceeded one both when outlets were counted separately and when grouped by owner. This supports a narrower distribution of citations than visits in those comparisons.
- Language reveals a separate pattern
A separate analysis used AI Search Arena prompts in countries’ own languages to examine citations to domestic state-controlled outlets. In that setting, those outlets received 1.65 times as many citations as a commercial group matched to the same traffic share.
The pattern was driven primarily by Russia and China; the English-prompted Ahrefs data did not show the same preference.
- Coverage bounds the conclusions
The evidence has limits that affect its reach. Ownership labels cover 70.1% of traffic but only 28.8% of domains, and thinner coverage is associated with higher measured concentration.
The citation datasets are snapshots, and the Arena data have lower coverage. The findings describe traffic and citations; they do not show that AI citations change actual traffic or measure editorial influence, revenue, or political power.
Deep
- Original paper authors
- Peter Carragher, Kathleen M. Carley
- Submitted to arXiv
- License
- CC BY 4.0