Context
Kai-Cheng Yang analyzed 24,069 AI Search Arena conversations and 366,087 citations from 12 OpenAI, Google and Perplexity search models in March–May 2025, finding distinct model-family patterns and concentration among news-source citations.
What changed
The July 7, 2025 original preprint quantified news-source selection across multiple AI search model families using real user comparisons: 32,865 (9.0%) of 366,087 citations pointed to news domains, and the top 20 news domains captured 67.3%, 31.9% and 28.5% of news citations in the OpenAI, Google and Perplexity model families, respectively.
Why it matters
Provider-specific citation concentration and diversity can differ substantially. AI visibility measurement should track source distribution and publisher diversity separately for each model family instead of relying on pooled citation totals.
Visibility OS interpretation
Visibility OS interpretation: citation concentration is a measurable source-selection dimension, not proof that an outlet's visibility can be optimized by any particular tactic. The findings describe a 2025 Arena sample, not live 2026 product behavior.
What remains uncertain
Do these concentration patterns persist in 2026 commercial products, languages, and query mixes? How would owner-level aggregation change results? Do cited sources materially support generated answers? How sensitive are results to the Arena user sample and domain-level source classifications?
Evidence claims
Yang's analysis of March–May 2025 AI Search Arena data extracted 366,087 citation URLs from 24,069 conversations; 32,865 citations (9.0%) were classified as news-domain citations.
Among news-domain citations in the 2025 Arena sample, the 20 most-cited news domains accounted for 67.3% of OpenAI-family citations, 31.9% of Google-family citations, and 28.5% of Perplexity-family citations.
In 1,534 judged AI Search Arena head-to-head comparisons where both responses cited news sources, the study did not find statistically significant associations between user preference and the measured news-source political-leaning or quality variables after its specified controls.
Related historical records
Research finds AI news citations more concentrated than audience traffic across countries
This record related to the linked event.