Study: 85.7% of AI brand citations come from third-party sites like Wikipedia

A new study by Dmitrij Zatuchin, published on arXiv, reveals how large language models (LLMs) source brand reputation across languages and markets. The research analyzes 167,551 URL-grounded citations (189,974 total attribution rows) across 128 brands in 12 home markets and 13 languages. By looking at the citations themselves—not just the model's answer text—the study uncovers four universal patterns.

First, LLMs overwhelmingly rely on third-party sources: 85.7% of citations come from sites the brand does not own, versus 14.3% from owned media. Second, the source distribution is highly concentrated and long-tailed—80% of citations come from about 18% of domains, fitting a Zipf law (alpha = 0.86, R² = 0.983). Third, Wikipedia is the most-cited domain in 11 of 12 languages, with the sole exception being Lithuanian, where the business daily vz.lt edges it by 4.38%. Finally, the source mix is market-specific: for 46 Polish national brands, YouTube is the top-cited domain, and four HR/career portals supply 637 citations versus just 297 for Polish Wikipedia—more than double. The study suggests that brand reputation in AI is shaped by a narrow, third-party-dominated set of sources that vary significantly by language and local market.