How Large Language Models Source Brand Reputation Across Languages and Markets

Large language models source brand reputation overwhelmingly from third-party websites, with a single reference site dominating citations in nearly every language examined, according to a new analysis of 167,551 URL-grounded citations across 128 brands and 13 languages [1][2]. The study, posted to arXiv on 24 June 2026, merged three datasets from Rankfor.AI covering 12 home markets and found that 85.7% of citations point to sites the brand does not own, against 14.3% that point to owned properties [1][2]. The source base is highly concentrated: 80% of citations come from about 18% of domains, a distribution that fits a Zipf law with an alpha of 0.86 and an R² of 0.983 [1][2]. Reputation, defined broadly as an opinion about a social entity developed through social evaluation, has long been understood as a mechanism of social control that operates across individuals, organizations, and nations [3]. The new findings show that AI systems are now shaping that mechanism by leaning on a narrow set of external references. Wikipedia is the most-cited domain in 11 of the 12 languages studied [1][2]. The exception is Lithuanian, where the business daily vz.lt edges it out, accounting for 4.38% of citations [1][2]. The source mix also shifts at the market level: for 46 Polish national brands, YouTube is the most-cited domain, and four human-resources and careers portals together supply 637 citations, roughly twice the 297 citations attributed to Polish Wikipedia [1][2]. The findings arrive as large language models become embedded in consumer-facing products. Google’s Gemini chatbot, for instance, integrates into the company’s mobile assistant and cloud platform, processing text, images, audio, and video through a single architecture [5]. The study’s authors note that most prior analysis of AI brand visibility examined only the answer text, not the underlying citations [1][2]. By tracing the citations, the research reveals that a brand’s presence in AI-generated answers depends less on its own websites and more on its footprint in a concentrated set of third-party sources that vary by language and market [1][2].