Do AI answers narrow the news people see? A 77-country preprint found citations more concentrated than traffic
Across two citation panels, seven AI systems drew on a narrower mix of outlets than audience traffic did. Missing ownership labels, proprietary prompts and few citations per answer mean the result is a hypothesis about future market power—not measured traffic loss.
The short version
- What happened
- The researchers linked 9,761 news domains to 3,540 owners and estimated traffic-weighted concentration for 77 countries; grouping outlets by owner raised the median concentration measure by 1.30 times.
- Who it affects
- Readers may experience a smaller set of news owners even when an AI answer lists several brands. Journalists and publishers could face weaker referral diversity if clicks follow citations, but this preprint did not measure that behavioural step.
- What remains unproven
- Unreviewed preprint
Keep in mind
Unreviewed preprint
Crowdsourced ownership records
Labels cover 28.8% of domains and 70.1% of traffic
Commercial citation panel uses proprietary prompts
English and US weighting
Smaller second panel underpowered
AI systems cite few sources by design
Owner-level AI analysis limited to 54 countries
No click-through or revenue outcome
No causal effect on pluralism
State-media claim especially uncertain
Source analysis · 2 linked sources. Primary analysed source date: 29 October 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 4 min The evidence, methods, implications and limitations
AI citations came from a narrower ownership mix than news traffic
A new cross-national preprint suggests that AI-assisted search may concentrate attention on fewer news organisations than readers do today. The authors assembled ownership labels for 9,761 digital-news domains linked to 3,540 owners, then compared traffic shares with source citations from seven AI systems. Across five systems in the larger Ahrefs panel, the median outlet-level concentration ratio ranged from 1.23 for Grok to 2.08 for Gemini. After websites were grouped by owner, the range was 1.11 to 1.79.
The pattern appeared in most measured countries and survived adjustment for multiple comparisons in the larger panel. A second panel, AI Search Arena, covered fewer domains and countries but pointed in the same direction for Perplexity, Gemini and GPT-4.0. The authors also found that grouping outlets by their common owner made national markets a median 1.30 times more concentrated than counting sites separately—important because several apparently distinct brands can belong to one company.[1][2]
The result combines ownership records, traffic estimates and citation panels
The dataset begins with Wikipedia and Wikidata, then adds state-media labels and monthly traffic estimates. Countries needed at least five identified owners for the 77-country analysis; a stricter 42-country set required at least 20 owner-labelled domains and 25% traffic coverage. Validation against 638 overlapping US domains found 73.5% current-owner agreement with the Medill directory, rising to 93.6% when the immediately previous owner counted. On 200 hand-labelled domains, the news filter reached 0.744 precision and 0.970 recall.
For each country and AI system, the researchers calculated the Herfindahl–Hirschman Index over the same outlets for traffic and citations. The larger panel used proprietary, English- and US-weighted prompts; the Arena panel used organic questions, about 40% not in English. Owner-level AI analysis fell to 54 countries because only 1,653 of 4,584 AI-cited domains had ownership labels. The released data and code make those choices inspectable.[1][2]
The distinction that matters
What was compared
The paper compares two distributions over the same news outlets; it does not observe readers moving between them.
Ownership
9,761 domains → 3,540 owners
National concentration estimates for 77 countries
Audience
Monthly traffic shares
Current attention benchmark
AI
Citation counts from 7 systems
Narrower distributions in both panels
Missing step
Citation → click → revenue
Not measured longitudinally
More concentrated citations do not yet prove fewer readers or less pluralism
Coverage is the largest caveat. Ownership labels reached 28.8% of domains and 70.1% of traffic, and thinner label coverage was associated with higher measured concentration. AI systems also cite only a handful of sources by construction, so some increase against the full traffic distribution may be unavoidable. The analysis ignores citation order and does not measure clicks, revenue, political power or editorial influence. Its Google AI Overviews calculation is a scenario in which citations reshape all organic traffic, not an observed market outcome.
The paper's native-language state-media claim needs extra restraint: the Arena evidence is underpowered, and the authors themselves say statistical claims from it are difficult to defend. For publishers and regulators, the study is best treated as a monitoring method and an early warning that ownership—not just outlet count—matters. Longitudinal citation and click-through data, stronger official ownership registers and comparisons that hold the number of available sources constant would show whether AI answers actually narrow what people read. The work is funded partly by the US Office of Naval Research and Knight Foundation.[1][2]
Watch the background, then check the evidence
Illustrative photo · not a video frame
The Royal Society · 44 min 28 sec · Background explainer
Is this the AI future we want? | The Royal Society
YouTube loads only when you choose; playback starts when you press play.
A Royal Society discussion with Madhumita Murgia and biomedical engineer Alison Noble on AI’s potential in healthcare, science and culture, recorded at the 30 June 2026 event.
Evidence check: A discussion of opportunities and risks, including future possibilities. It is not evidence that every application described is available or clinically proven.
Published 2026-07-01. Source checked 14 October 2026. Check captions on the publisher’s video page.
Check the healthcare evidencePhoto: Daniel Sone / National Cancer Institute / Unsplash · Licence
Continue the story
Read next
A second perspective on AI in society & media, with its own sources and qualifications.
Why do people see ChatGPT so differently?
Convenience samples in Germany and Serbia differed sharply in AI literacy and in whether they saw ChatGPT as useful or risky. The study maps associations among knowledge, beliefs and attitudes, but its self-report design cannot show that literacy caused trust or that the samples represent either country.
Read the next reportExplore the full source and evidence record
What this means for people
- Readers may experience a smaller set of news owners even when an AI answer lists several brands. Journalists and publishers could face weaker referral diversity if clicks follow citations, but this preprint did not measure that behavioural step.
Global context
The dataset spans 77 countries, but ownership coverage and Wikipedia maintenance vary by language and market. The stronger within-country comparison is more defensible than treating every country ranking as equally complete.
What the evidence does not yet show
- Unreviewed preprint
- Crowdsourced ownership records
- Labels cover 28.8% of domains and 70.1% of traffic
- Commercial citation panel uses proprietary prompts
- English and US weighting
- Smaller second panel underpowered
- AI systems cite few sources by design
- Owner-level AI analysis limited to 54 countries
- No click-through or revenue outcome
- No causal effect on pluralism
- State-media claim especially uncertain
What to watch next
- Peer review
- Official ownership-register validation
- Longitudinal AI citation and click data
- Fixed-source-count comparisons
- Citation rank and visibility
- Country-specific high-coverage replications
- Publisher revenue effects
- More robust native-language state-media tests
Living evidence record
Impact record IAI-186UIKM
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Not yet
Record status
Monitoring
Last checked
31 October 2026
Source trail
2 direct sources across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Related-source reporting disclosure
This record analyses 2 linked source records around the same underlying development. The extra records add method, date or context, but they do not by themselves constitute independent replication of every performance claim or predicted outcome.
Evidence trail
Sources used for this report
Links checked 31 October 2026
Research question, topics and regions
Cross-national comparison of digital-news traffic concentration and citations by generative AI search systems
Topics: AI search · News media · Ownership concentration · Publisher traffic · Platform power · Media pluralism
Regions: Global · United States · United Kingdom
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.