Processing, using generative model, content of user input and temporal encoding for the user input

Topics: AI Mode, LLMO / GEO, Personalization, Probably in use, Search Intent, User Signals

This patent by Google LLC describes a system and method for improving how generative models (such as large language models) respond to user input by considering not just what the user typed or said, but how they typed or said it. Specifically, the system captures temporal characteristics of user input — such as typing speed, time delays between keystrokes, or speaking rate — and encodes these as a “temporal encoding” that is combined with the traditional content embedding before being processed by the generative model. This allows the model to infer user intent or confidence level and tailor its response accordingly. For example, a user who types a query quickly and confidently may receive a detailed, fact-rich answer, while a user who types the same query slowly and hesitantly may receive a simpler, more explanatory response. The goal is to make AI-generated responses more personalized and contextually appropriate, even when the literal text of two queries is identical.