Determing uniform resource locator (URL) characterics using generative artificial intelligence models
Topics: AI Mode, AIOverviews, Freshness, Indexing, LLMO / GEO, Scoring
This U.S. patent application describes a URL understanding system that uses multiple language-based generative AI models to quickly evaluate newly discovered URLs. It first builds a URL language model to learn the syntax, structure, and “grammar” of URL strings by treating them like sentences. This knowledge is then transferred to a URL prediction model that determines selection probabilities (e.g., click likelihood), enabling near-real-time decisions on whether to add high-value URLs to a search index, which improves freshness and efficiency over traditional multi-day analysis methods.
