Google Patents a System That Generates Custom Ad Images for Each Individual User
Google has filed a patent describing a system that combines your personal data with a merchant's brand assets, feeds them through AI text and image generators, and produces a custom ad image built specifically for you. Every user could see a visually different version of the same promotion.
What Google's per-user AI ad images actually do
Ever noticed how most online ads feel like they were made for someone else? That's because they were, or at least for a broad category of people rather than you specifically.
Google's patent describes a system that works differently. It takes what it knows about you as a user, combines that with a brand's logos, colors, and product details, and runs both through AI models to write descriptive text and then turn that text into a generated image. The result is an ad picture that is assembled fresh for your profile, not pulled from a library of pre-made creative.
In practice, two people searching for the same sneaker brand could see two completely different-looking ads, each one shaped by their own browsing habits, preferences, or demographics. The merchant doesn't have to design dozens of variations by hand; the AI handles the visual output automatically.
… processing the user personalization data and the asset data with a text generation model to generate one or more model-generated terms; processing the one or more model-generated terms with an image generation model to generate one or more model-generated images; …
Translation: It feeds personal details and product info into a text model to create terms, then turns those into custom pictures.
How the text and image models build each personalized ad
The system described in the patent works in a pipeline with two main AI stages.
Stage one: text generation. The system collects user personalization data (things like interests, demographics, or past behavior) and merchant asset data (the brand's products, colors, slogans, or other creative materials). A text-generation model processes both inputs together and produces what the patent calls model-generated terms, essentially a tailored text description that bridges what the user cares about with what the merchant sells.
Stage two: image generation. Those generated terms are then fed into a text-to-image model (an AI that converts written descriptions into pictures, similar in concept to tools like DALL-E or Imagen). The model outputs one or more images constructed specifically to match that user-and-merchant combination.
The final ad unit shown on the user's device is built from those AI-generated images. The system is designed so that:
- No single human designer has to produce every variant
- The merchant's brand identity can still be baked into the output through asset data
- The same merchant ad request can yield visually different results for different users
The patent covers the full loop from data ingestion through display on a user's screen.
Techniques for presenting a content item using text-to-image machine-learned models are presented. For example, a system can obtain user personalization data associated with a user and merchant assets data of a merchant.
Translation: The system collects information about your personal preferences along with product details provided by a merchant.
What this means for the ads you see every day
For you as a user, the most noticeable effect would be ads that feel less like interruptions and more like things you might actually want. Whether that feels convenient or invasive depends on your comfort with personalization, but the stated goal is relevance rather than volume.
Google has been filing around AI-generated advertising since at least 2024, and this patent extends that direction into image creation itself. For advertisers, the upside is obvious: less manual creative production and potentially higher engagement. The open question is how well AI-assembled images hold up against professionally designed ones, and whether users will ever notice or care that what they saw was made just for them.
Google's 35th filing in the AI image and video work we've tracked since May adds to a run that includes training without photo-text pairs and editing photos guided by text.
If this system works as described, the ad you see for hiking boots would be generated fresh, styled to match your apparent taste rather than pulled from a generic photo shoot. That's a real change in what appears on your screen, even if you'd never know it happened.
The gap worth noting is quality. An AI painting a custom image for a paid ad can produce something odd or off-brand, and the patent describes no step where a human checks the result before it reaches you. That means the failure mode isn't a boring ad, it's a weird one.
For most people, the honest range of outcomes runs from "slightly more relevant" to "mildly unsettling when you think about it." The system cuts production costs for advertisers, and it uses your profile to literally illustrate something you're about to be sold. Whether that feels like a feature depends entirely on how you already feel about targeted advertising.
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
10 drawing sheets from US 2026/0278018 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →