arXiv News

First-hand information for everyone

Site

AboutPrivacy PolicyTerms of Use

© 2026 arXiv News

AI Platforms Converge on the Same News, Increasing Concentration in AI Cited Sources | arXiv News

arXiv News

EnglishJapanese

Switch language

EnglishJapanese
Loading account…

First-hand information for everyone

Back to front page
cs.CYEnglishPublished

AI Platforms Converge on the Same News, Increasing Concentration in AI Cited Sources

November 1, 2026arXiv: 2610.40164v1

Researchers built the world’s first public inventory of ownership for 9,761 digital news outlets in 77 countries, using data from Wikipedia and Wikidata. Using this inventory, they calculated traffic‑weighted Herfindahl–Hirschman Index (HHI) concentration scores for each country. The results show that media markets vary widely, with some nations dominated by a handful of owners and others more evenly split.

Map of 77 countries with each country’s HHI colour‑coded, highlighting which nations have high concentration—values near 10 000 mean a single owner dominates, while values near zero indicate many low‑share owners.
Figure 1. Traffic‑weighted Herfindahl–Hirschman Index values per country.Carragher, Peter, Carley, Kathleen M. · arXiv:2610.40164v1 · Figure 1. · CC BY 4.0

Applying the same ownership map to the ways AI systems cite news, the study compared AI‑citation concentration to ordinary audience traffic. Across seven AI models, including popular tools like Gemini, Copilot, and Google AI Overviews, the median amplification of concentration ranged from 1.23‑x to 2.08‑x. That is, for a given country, AI citations were concentrated on a narrower set of outlets than the outlets actually visited by readers.

Table showing median AI ÷ traffic HHI multipliers for each AI model, with statistical confidence intervals and the proportion of countries where the AI system’s concentration exceeds traffic concentration.
By outletBy owner
DataModelAI ÷ traffic HHInAI>trafficAI ÷ traffic HHInAI>traffic
AhrefsGemini2.08×*** [1.65, 2.56]65 / 761.79×*** [1.49, 2.00]47 / 53
Copilot1.60×*** [1.38, 1.86]59 / 761.33×*** [1.13, 1.57]42 / 54
Google1.57×*** [1.36, 1.73]60 / 761.34×** [1.14, 1.46]39 / 54
ChatGPT1.35×*** [1.22, 1.63]61 / 761.33×*** [1.16, 1.58]41 / 54
Grok1.23×** [1.09, 1.38]51 / 761.11×* [1.02, 1.31]35 / 54
ArenaPerplexity1.76×*** [1.47, 1.95]23 / 311.19×* [0.99, 1.53]20 / 31
Gemini1.70× [1.02, 2.70]13 / 171.28× [0.99, 1.67]12 / 16
GPT 4.01.60×* [1.23, 6.05]10 / 131.47×* [1.09, 1.73]10 / 13
Table 2. AI citation‑to‑traffic HHI amplification ratios across AI systems.Carragher, Peter, Carley, Kathleen M. · arXiv:2610.40164v1 · Table 2. · CC BY 4.0

Because the AI amplification effect appears strongest in countries that already have lower traffic concentration, the results suggest that AI intermediaries can reinforce existing media dominance. However the study notes that the dataset covers only 28.8% of known news domains, and that reliance on Wikipedia‑based information may bias concentration estimates.

Visual 1 illustrates the traffic‑weighted HHI values per country, while Visual 2 (Table 2) details the amplification ratios for each AI system at both outlet and owner levels.

Source paper

Who Owns Digital News, and Who Gets Cited: Measuring Media Concentration Across 77 Countries and Seven AI Systems

cs.CY
View on arXiv