Get the newsletter
Google · AI / ML · Filed May 14, 2025 · USPTO published Nov 19, 2026 · This week's release

Google Files Patent for Ads That Blend Into AI Search Answers

Google has filed a patent application for an AI search assistant that writes ads into its own answers, so a sponsored link can read like part of the reply instead of sitting in a box beside it.

An AI search answer containing embedded shopping links for juicers and fresh vegetables woven into the advice text. Drawing from patent filing US 2026/0368496 A1.
An AI search answer containing embedded shopping links for juicers and fresh vegetables woven into the advice text.All 14 drawings ↓
Publication number
US 2026/0368496 A1(US20260368496A1)
Applicant
Google LLC
Filed · Published
May 14, 2025 · Nov 19, 2026
Inventors
Matthias Heiler, Shashidhar Anil Thakur, Dorothea Wiesmann Rothuizen, Vivek Rao
Status when we published
Waiting for an examiner
Document
20 claims
US classification
706/11

What Google's ad-in-the-answer patent does

Someone types a question into an AI search box: is it healthy to make vegetable juice at home? The assistant writes back a few paragraphs. Inside them, the name of an ingredient is a link where you can buy it.

That is the idea in this Google patent application. Instead of putting a banner or a boxed ad beside the answer, Google describes rewriting the advertiser's content so it reads like part of the answer. The filing says people learn to ignore anything that looks like a typical ad, so the goal is to make sponsored content look organic and less intrusive.

For you, that could mean an AI answer carrying a sponsored link, image or short video that matches the tone of the text around it. The filing's own example is a juice recipe with a shopping link for every ingredient.

From the filing · CLAIM 1
… generate a preliminary response and a full answer to a first query; a first machine-learned model, wherein the first machine-learned model is configured to generate a proxy phrase based on a long-form content; and a second machine-learned model …

Translation: The system uses two separate AI models to pick keywords and shape external content before blending it in.

How a short keyword phrase finds the ad

Google's filing describes a pipeline that runs behind an AI search assistant (a chatbot-style search tool that writes full answers instead of listing links). The assistant first produces a preliminary response. That can be a cached answer to a similar earlier question, a best guess, a summary of key points, or the first paragraph of the answer.

From there, the steps look like this:

  • A first model boils the preliminary response down to a proxy phrase, a handful of keywords (the filing says fewer than six) that capture what you want. Its example turns "leather pants men fashion nova" into "men's leather pants fashion nova."
  • That phrase goes to a content server (an ad backend), which returns a content item such as a text ad, image or video. The filing says these servers handle short phrases well but can pick irrelevant items when fed long text.
  • A second model rewrites the content item to fit the answer, taking its tone into account, and decides where to place it. The result can be text with a link, an image or a video.

The ad lookup runs in parallel with writing the full answer, so you are not kept waiting. The first model can also judge whether the answer involves a product or service at all. If it does not, no link is added. The assistant then combines the rewritten media with the full answer into the final reply. The filing also mentions filters for brand safety and inappropriate content.

From the filing · THE ABSTRACT
… process, using the second machine-learned model, the first content item and the full answer to generate a final answer to the first query.

Translation: The final step combines the ad content with the AI answer so everything flows together smoothly.

Why ads inside AI answers matter for search

If you use AI search, this filing shows one way those answers could carry ads. Search has long run on ads, and a chatbot-style answer has no obvious slot for one. The filing's fix is to put the ad inside the answer: a link on an ingredient, a product image, or a short video, written to match the tone of the text.

For Google's ad business, the useful part is the plumbing. The filing says existing ad servers are built for short phrases, so the first model translates a long answer into a few keywords. That lets the current ad machinery plug into AI answers. The filing also describes a check that leaves links out when the answer has nothing to do with a product or service.

This is the 99th filing from Google in AI assistant and agent coverage we've tracked since May, joining one on speeding up chatbots and one on shrinking example lists.

Source: US 2026/0368496 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.

Get our take in your Top Stories

Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Google patent news breaks.

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 · 14 sheets

14 drawing sheets from US 2026/0368496 A1 · click any drawing to enlarge

How the AI selects ads to embed into search answers, drawing sheet 1 from US 2026/0368496 A1
Sheet 1 How the AI selects ads to embed into search answers
The step-by-step exchange between the user and ad servers, drawing sheet 2 from US 2026/0368496 A1
Sheet 2 The step-by-step exchange between the user and ad servers
Steps for weaving sponsored product links into an answer, drawing sheet 3 from US 2026/0368496 A1
Sheet 3 Steps for weaving sponsored product links into an answer
Shopping tags blended directly into an AI recipe answer, drawing sheet 4 from US 2026/0368496 A1
Sheet 4 Shopping tags blended directly into an AI recipe answer
Training the AI model with feedback signals, drawing sheet 5 from US 2026/0368496 A1
Sheet 5 Training the AI model with feedback signals
Turning user inputs into text, image, or video media, drawing sheet 6 from US 2026/0368496 A1
Sheet 6 Turning user inputs into text, image, or video media
Predicting output text from an incoming sequence of data, drawing sheet 7 from US 2026/0368496 A1
Sheet 7 Predicting output text from an incoming sequence of data
Combining different types of input data into a single line, drawing sheet 8 from US 2026/0368496 A1
Sheet 8 Combining different types of input data into a single line
Show the other 6 sheets
Software tools used to build and test the AI, drawing sheet 9 from US 2026/0368496 A1
Sheet 9 Software tools used to build and test the AI
Training steps that teach the AI using human feedback, drawing sheet 10 from US 2026/0368496 A1
Sheet 10 Training steps that teach the AI using human feedback
Cloud servers answering search queries with live web data, drawing sheet 11 from US 2026/0368496 A1
Sheet 11 Cloud servers answering search queries with live web data
Connected computers and servers sharing the search workload, drawing sheet 12 from US 2026/0368496 A1
Sheet 12 Connected computers and servers sharing the search workload
Each phone app running its own dedicated AI model, drawing sheet 13 from US 2026/0368496 A1
Sheet 13 Each phone app running its own dedicated AI model
A shared system managing AI models for all apps, drawing sheet 14 from US 2026/0368496 A1
Sheet 14 A shared system managing AI models for all apps
Patent filing page

More in AI assistant & agent

All 436 AI assistant & agent patents →
Reader comments

No comments yet

Start the discussion

Real name or a handle, either is fine. Comments are read by a person before they appear, so allow a little time. Keep it about the filing.