Generating and Utilizing compressed grounding data data for search engines that utilize generative artificial intelligence models

Topics: Chunk Relevance, Freshness, Information Gain, LLM Readability, LLMO / GEO, Microsoft, Retrieval Augmented Generation (RAG), Semantic Search

This patent application describes a system called the Grounding Compression System, which is designed to intelligently reduce the volume of information (called grounding data) fed into generative AI-powered search engines before they generate a response. Instead of simply cutting data down in size, the system distills, prioritizes, and compresses the most relevant information based on the actual search query. The patent applies directly to Generative Search Engines (GSEs) that use large language models (LLMs) such as systems similar to Google AI Mode, ChatGPT-based search, and other AI-driven search tools that retrieve web content and generate answers. For GEO (Generative Engine Optimization), this patent is highly relevant because it reveals exactly how AI search systems decide which content gets selected, summarized, and passed to the AI model, meaning it directly influences which sources and content types appear in AI-generated answers.