Scalable high-accuracy transactional agents

Topics: Agentic Commerce Optimization (ACO), AI Mode, Entity based search, LLMO / GEO, Personalization, Product Context Optimization, Retrieval Augmented Generation (RAG), Shopping

This Google patent describes a conversational AI agent designed to handle transactional interactions, such as ordering food at a drive-through, with high accuracy and in a natural, human-like way. The system uses a multi-stage modular architecture built on Large Language Models (LLMs) to understand what users are saying, match their requests to real catalog items, and maintain the flow of conversation across multiple exchanges. It is directly applicable to Google AI systems, including Google Assistant, Google AI Mode in Search, and other Google-powered conversational agents. For GEO (Generative Engine Optimization), this patent is highly relevant because it reveals how AI agents parse, structure, and match user intent to catalog entries, offering direct insight into how content and product information should be structured to be accurately retrieved and acted upon by AI-driven systems.