Search system having task-based machined-learned models
Topics: AI Mode, LLMO / GEO, Navboost, Personalization, Probably in use, Query Fan Out, Ranking, Reranking, Retrieval Augmented Generation (RAG), Search Intent, Search Query Processing, User Signals
This Google patent describes a task-based search system that uses machine-learned models to break down a broad user query into specific subtasks. Instead of just returning generic search results, the system identifies what a user is actually trying to accomplish (a “task”), splits that into smaller steps (“subtasks”), and then ranks and displays content based on how other users have interacted with similar results. The system also allows advertisers to target their sponsored content at the task or subtask level, rather than just matching keywords.
This patent represents a significant shift in how Google may structure search results and ad targeting — moving from keyword-level matching to goal-level (task-based) understanding, powered by a combination of LLMs and specialized multi-task machine-learned models.
