Google
Google · AI / ML · Filed May 6, 2025 · USPTO published Nov 12, 2026 · This week's release
Google Files Patent for Ad Software That Picks the Best Layout in Milliseconds
An advertiser can supply enough images, headlines and templates to make millions of different versions of one ad. Google's patent application describes a cheap, fast guess that narrows the pile to a single finalist, so only that one gets a full review.
- Publication number
- US 2026/0359612 A1(US20260359612A1)
- Applicant
- Google LLC
- Filed · Published
- May 6, 2025 · Nov 12, 2026
- Inventors
- Haifeng Gong, Dongdong Wang, Xiaohang Li
- Status when we published
- Waiting for an examiner
- Document
- 20 claims
- US classification
- 706/15
What Google's ad-layout shortcut actually does
When a web page asks for an ad, the clock is already ticking. If the ad arrives late, the page can load with blank spots, or you may have moved on before it shows up. That is a problem when one advertiser's headlines, pictures and templates can be mixed into thousands or even millions of versions, and judging each one with a heavy AI model takes too long.
Google's patent application describes a shortcut. A lightweight guess scores every mix of image, text and layout for the person about to see the ad. Only the top pick gets a slower, fuller review. The system then builds that ad and sends it to your phone or computer.
The score is a prediction of how likely you are to click, take an action, or simply look at the ad.
… computing, for each asset and layout combination, a predicted quality metric using the parameters of the one or more Gaussian mixture models, the asset embedding for each asset of the asset and layout combination and the layout embedding for the layout of the asset and layout combination; …
Translation: The system scores every possible combination of pictures, text, and design templates to find the most appealing setup.
How the quick scoring picks one ad finalist
The system starts with a neural network (a type of AI that learns patterns from past examples). It reads details about the request, such as the page, the device type, the time of day and the region, plus details about the advertiser's product and landing page. From those it produces the settings for one or more Gaussian mixture models, which are sets of bell-curve-shaped clusters. The closer a combination sits to the center of a cluster, the higher its predicted quality. Fresh curves can be made for every request.
Every image, headline and layout has already been turned into an embedding, a list of numbers that describes it, usually prepared ahead of time. For each combination, the system joins the embeddings, scores them against each curve, weights the results and adds them up. It then picks the top score, or the top handful.
From there the steps are:
- A heavier AI model checks only the winning combination, predicting how likely you are to click, take an action or actually view the ad.
- A generator fills the layout with the chosen pieces, either by dropping them into template slots or by using an image-generating model.
- The finished ad goes to the device, and may first compete against other candidates, with its score multiplied by what the advertiser will pay.
Asset and layout combinations are identified based on the set of assets and the set of layouts. For each asset and layout combination, a predicted quality metric is computed using the parameters of the Gaussian mixture model(s).
Translation: It tests different mixes of ad parts and layouts to see which one works best.
Why faster ad building matters for Google ads
For you, the effect would mostly be invisible: ads built to match the page, the search or even the chat you are in, assembled in the moment instead of picked from a fixed set. The filing says new ads need to be ready in under a second, because late ads can leave blank spots on a page or arrive after you have moved on.
For advertisers, it means supplying many headlines, pictures and templates and letting the software find the best pairing. The filing notes that conversational AI agents can send these ad requests too, using what you ask as context. Keep in mind this is a patent application, so nothing here says it is running anywhere today.
Google's 26th filing we've tracked since May in our AI recommendation coverage follows earlier work like one on ranking data for AI training and one on generating ad scenes from photos.
Source: US 2026/0359612 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.
Add Patentlyze on Google
One click.
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 · 4 sheets
4 drawing sheets from US 2026/0359612 A1 · click any drawing to enlarge
More from Google
All 1,079 Google patents →
More in AI recommendation
All 86 AI recommendation patents →
Sony
Sony Patents a System That Sorts You Into Fan, Superfan or Hater
Adobe
Adobe Patents an AI System That Picks Your Email Templates and Test Variants for You
Disney
Disney Patents an AI That Matches Bonus Content to Shows and Their Viewers
Samsung
No comments yet