The Promptification of the Search Bar: Evidence of a Behavioral Spillover from Generative AI

Abstract Large language models (LLMs) have produced documented behavioral changes across information tasks, raising the question of whether they have also changed how users formulate requests to the legacy search bar. This study documents a durable structural shift in the syntactic register of web search queries following LLM diffusion, a shift termed “the promptification of the search bar” and characterized through panel co-movement and temporal precedence at population scale. Using 196,500 weekly US search-frequency observations from Google Trends across 25 semantic categories and 30 intent markers (2021 to 2025), this study classifies intent query markers into three classes of increasing conversational delegation and estimates their differential response to LLM diffusion. Conversational delegation rose by 58.7% after a structural break in June 2023, moving from no relation with LLM diffusion to tight co-movement, with temporal precedence running from diffusion to query syntax and not the reverse. An epoch decomposition shows the shift was established roughly eleven months before the platform deployed its own generative layer within the legacy tool, and the elasticity to LLM diffusion declined rather than increased after deployment, the opposite of what a platform-driven account predicts. The shift is strongest in domains serving cognitively complex needs, deepens over time as longer, more elaborate conversational forms standardize, and replicates worldwide. These findings are consistent with a spillover from LLM use into the legacy search bar: freed by exposure to conversational machines, users are returning to the natural register that keyword era had compressed, and the legacy interface is adapting to a behavior it did not create.