New Google Patents · Filed Mar 30, 2026 · Published Aug 6, 2026 · verified — real USPTO data

Google Patent Ensures Search Results Name the Specific Entity You Queried

When you ask Google a question, you often get a list of links instead of an answer. This patent describes a system that identifies the specific person, place, or thing your question is about and names it directly in the reply.

Google Patent: AI Answers That Name Their Sources — figure from US 2026/0228294 A1
Figure from the official USPTO publication.
See all 8 drawings from this filing ↓
Publication number US 2026/0228294 A1
Applicant GOOGLE LLC
Filing date Mar 30, 2026
Publication date Aug 6, 2026
Inventors Dvir Keysar, Tomer Shmiel
CPC classification 707/730
Grant likelihood Medium
Examiner CENTRAL, DOCKET (Art Unit OPAP)
Status Docketed New Case - Ready for Examination (Apr 29, 2026)
Parent application is a Continuation of 18600264 (filed 2024-03-08)
Document 20 claims

How Google wants Search to answer questions more directly

Imagine asking Google "Who invented the polio vaccine?" and instead of ten blue links, you get a clean sentence back: "Jonas Salk invented the polio vaccine." That's the kind of response this patent is designed to produce.

The system works by figuring out what type of thing your question is looking for, whether that's a person, a company, a date, or a location. It then scans the top search results, finds the specific name that fits, and builds a natural-language answer around it.

The key detail is that the answer includes text that wasn't in your original question. You asked "who" and Google fills in the name. That sounds simple, but getting a machine to reliably extract the right entity from messy web text and package it into a grammatical sentence is harder than it looks.

How the system pulls entity references from ranked results

The patent describes a pipeline with a few distinct steps:

  • Query classification: When a question arrives, the system identifies what category of thing the question is hunting for (a named entity type, like a person, organization, or location).
  • Search and ranking: Standard search results are generated and ranked by relevance.
  • Pre-cached entity extraction: Here's the interesting part. Rather than parsing the top results on the fly, the system retrieves previously generated data already attached to those pages. That data contains "entity references" found inside the page's unstructured text, meaning names and labels that were extracted earlier during indexing.
  • Answer generation: A natural-language answer is assembled that includes at least one term from the original question and the resolved entity name, producing a direct, readable response.

The phrase "collective reconciliation" in the abstract suggests the system may be resolving conflicts when different top results point to different candidate entities, picking the most consistent answer across sources.

We find one patent like this every day. Get the best of each week in your inbox, free →

What this means for Google's AI-powered search answers

This patent sits squarely in the infrastructure behind Google's AI Overviews and the older "featured snippet" answers. The specific contribution here is using pre-indexed entity data rather than running fresh text analysis at query time, which would be faster and cheaper at the scale Google operates.

For you as a searcher, the ambition is fewer clicks and more direct answers. For the broader web, it continues a pattern where Google's search result page provides the answer itself, reducing traffic to the source sites that supplied the underlying information.

Editorial take

This is solidly interesting infrastructure work, not a flashy consumer feature. The pre-cached entity reference approach is a meaningful engineering choice that makes AI-generated answers faster and more consistent. It won't make headlines, but it's the kind of plumbing that makes Google's AI search answers actually reliable at scale.

There are more where this came from

We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.

The drawings

8 drawing sheets from US 2026/0228294 A1 · click any drawing to enlarge

Patent filing page

Source. Full patent text and figures from the official USPTO publication PDF.

Editorial commentary on a publicly published patent application. Not legal advice.