A New Patent Rewrites Your Search Query Before Looking Anything Up
Before Google even starts searching, this patented system would rewrite what you typed, using where you are and what you've searched for in the past to build a better question than the one you asked.
How Google's query-rewriting AI uses your history and location
Imagine you type "best place nearby" into Google Search. It's vague, but Google already knows you're in Chicago, and that you've been searching for vegan restaurants all week. This patent describes a system that takes all of that context and uses an AI to rewrite your query into something more specific before the actual search even begins.
The idea is that what you type and what you mean are often two different things. Instead of matching your exact words to a list of web pages, the AI generates a refined version of your question, one shaped by your location and your recent search behavior.
The system can even chain these rewrites together: the AI's first attempt at rephrasing feeds back into itself to produce an even better version. Only then does Google use the result to find and display answers.
How the generative model chains outputs into a final query
The patent describes a generative model (an AI similar in spirit to the kind that powers chatbots) that takes three inputs at once:
- Your original query, whatever you typed or spoke
- Your location, where you are right now
- Attributes built from your past searches, a profile of your interests and behavior inferred from earlier queries
The model processes all three and produces a query variant: a rephrased version of your question that better captures what you probably meant. That output is then fed back into the same model as additional input, producing a second, further refined output. The final answer Google returns to your screen is based on this chained, context-enriched version of your original question, not the raw text you submitted.
The technique is called autoregressive generation (the model uses its own previous output as part of the next step), which is the same basic mechanic behind large language models like Gemini or GPT. The novel angle here is applying it specifically to the query-reformulation problem inside a search engine, with user context baked in from the start.
What this means for how Google Search reads your mind
Search has always struggled with the gap between what people type and what they actually want. Keyword matching got better, then semantic search got better, but both still depend heavily on the user writing a good query. This patent describes a system that simply sidesteps that dependency by treating your query as a rough draft that an AI will clean up.
For you as a user, this could mean fewer "refine your search" dead ends and more results that feel like Google read between the lines. For advertisers and publishers, it means the query that determines which results (and ads) you see may look nothing like what you actually typed, raising real questions about transparency and control over how intent is interpreted.
This is a genuinely consequential patent because it formalizes something Google has been moving toward for years: treating your typed words as a suggestion, not a command. The chained-generation approach is technically interesting, but the bigger story is the explicit use of past query behavior as a permanent input to how your current search is interpreted. That's a design choice with privacy implications worth watching.
Which company should we read for you?
We track 17 companies here. Pro is the same weekly breakdown for any company you choose, delivered privately. Type a name and we'll scope it and send you a quote.
Get one Big Tech patent every Sunday
Plain English, intelligent commentary, no hype. Free.
Editorial commentary on a publicly published patent application. Not legal advice.