Can LLMs Reason About Brand Ownership? An Empirical Study of Domain Attribution Intelligence

This study evaluates the effectiveness of large language models (LLMs) in determining brand ownership of domains, a critical task for preventing phishing attacks. Four models—Gemini 2.5 Flash, Gemini 3.5 Flash, Claude Sonnet 4.5, and Claude Sonnet 4.6—were assessed on their ability to perform domain enumeration, brand attribution, and ownership classification, revealing that while models can achieve up to 82% precision in domain enumeration, their ownership verification performance is limited without external tools, with a maximum macro F1 score of 0.37 in in-context learning mode. The integration of WHOIS data significantly enhances ownership verification accuracy, demonstrating the potential for LLMs to improve brand protection strategies when combined with external resources.