Google's newly granted patent US12536233B1 reveals a system that scores landing pages and replaces low-quality ones with AI-generated pages personalized per user search history.
A patent granted to Google, published on January 27, 2026, has drawn sharp attention from the search and advertising community - describing a system that could dynamically replace an advertiser's landing page with an entirely AI-generated version, assembled in real time from a user's search history, contextual signals, and performance data pulled from the original page.
The patent, numbered US12536233B1 and titled "AI-generated content page tailored to a specific user," was filed on January 3, 2025, by inventors Caren Zeng, Rushil Grover, Timothy Benjamin Whalin, Lauren Marjorie Bedford, Pallavi Satyan, and Ethan Milo Mann. The assignee is Google LLC. A provisional application, US 63/675,726, was first submitted on July 25, 2024, establishing the original priority date. The patent was granted on January 27, 2026, with publication of US20260030300A1 following on January 29, 2026. A parallel European application, EP25191927.0A, was filed on July 25, 2025.
The core technical claim is deceptively straightforward. According to the patent document, the system receives a user query from a device tied to a user account, generates a standard search result page for that query, and then calculates what the patent calls a landing page score for the top organic result associated with a given organization. If that score exceeds a defined threshold value, the system generates an updated search results page containing a navigation link to an AI-generated version of that landing page. The user then sees the updated page, not the original.
What determines the landing page score? The patent describes several criteria operating in parallel. The score can be determined based on a conversion rate associated with the landing page, its bounce rate, its click-through rate, or a qualitative factor - specifically described in the claims as "page design quality or content quality." In a further variation, the score can flag poor performance when the first landing page simply does not have a filter for products, a remarkably low bar that could apply to a wide range of e-commerce sites.
The AI-generated replacement is not a blank canvas. According to the patent's description of the system, the machine-learned model processes both the user's current query and contextual information associated with the user account - including a previous query - to produce the generated page. Each component of the AI-generated page, the patent states, "is annotated with dynamically generated content based on the user query." The page can include a call-to-action button linking to a product page, a product feed providing an overview of relevant items, an AI chatbot, sitelinks to product detail pages, and a personalized headline with suggested filters and suggested clusters.
The updated search result page containing the navigation link to the AI page can, according to claim 12, be included in a sponsored content item - an advertising unit. That detail carries significant implications for how costs and attribution would be allocated in any commercial implementation.
The patent describes a machine-learned content page generation pipeline that takes input from two primary data sources: a general data resource and an account profile. The account profile contains user preferences, the current user query, previous searches, and past signals and controls previously submitted to the pipeline.
The pipeline includes four separate generation modules: a machine-learned text generator, a machine-learned image generator, a machine-learned audio generator, and a machine-learned video generator. An optimizer module applies optimization algorithms to the outputs of any combination of these generators. A ranking module then orders the outputs. An asset feedback layer sits above the ranked outputs, accepting user feedback on generated assets and triggering regeneration of updated or different components. Only after that feedback loop does the system output the final AI-generated content page.
The patent's model architecture section describes support for neural networks including feed-forward networks, recurrent neural networks, long short-term memory models, convolutional neural networks, diffusion models, and generative-adversarial networks. The preferred architecture, described in detail, is a transformer-based sequence processing model leveraging multi-headed self-attention. The training workflow covers pre-training over large-scale noisy datasets, fine-tuning on higher-quality labeled data, and refinement using reinforcement learning from human feedback - a now-standard pipeline across Google's AI products.
Training methodology described in the patent includes backwards propagation, gradient descent, truncated backpropagation through time, weight decays, and dropout regularization. The patent also describes a model development platform incorporating alignment toolkits, fine-tuning pipelines, prompt libraries with few-shot and chain-of-thought prompts, and computational optimization tools including quantization, weight pruning, and model distillation for deploying lightweight versions in resource-constrained environments.
The patent's description section includes a concrete example. A user's journey involves three previous searches - "best laptop for architecture," "best laptop for 3D modeling," and a specific laptop brand name. On a subsequent search, the system generates a link to an AI-generated page customized to that user based on their prior interactions. The page is not the advertiser's actual website. It is a version assembled by Google's model, populated with content it determines is most relevant to that user's demonstrated research path.
The figures in the patent depict several graphical user interfaces for these AI-generated pages, though the patent text describes rather than shows the specifics in detail. The interfaces include personalized headlines, suggested filters, suggested clusters, call-to-action buttons, and product feeds, all dynamically populated per user.
The patent drew immediate commentary from digital marketing professionals after it circulated on LinkedIn. Joshua Squires, Associate Director of SEO at Amsive, noted the patent and raised concerns. According to the LinkedIn post, he described the system as one in which "Google would use AI to generate a page that looks like your website but rebuilds the entire structure of a page dynamically, in real time, and places it at the top of the SERP."
Squires drew a comparison to Google's AI Max for Search campaigns, noting that Google's AI efforts in that product "cause about as many problems as they solve, including but not limited to: black-boxed attribution, budget cannibalization, brand misalignment/brand safety issues, hallucinated offers."
He also cited data from the ANA Q2 2025 Programmatic Transparency Benchmark, according to which lost global media value in programmatic advertising saw a 34% increase from 2024, reaching $26.8 billion. "A technology like this will skyrocket that number," he wrote.
Hank Azarian, a full-stack marketing and AI professional, responded that there are "generative UI experiments running in labs right now." Eric Hoover, Director of SEO and GEO, called it "a terrible idea for everyone involved aside from Google," and added that AI "seems less like a democratizing tool and more like a tool for consolidating wealth and power at the top."
Squires also flagged concerns about legal liability, customer lifetime value erosion from inconsistent brand experiences, a heightened reliance on Google's ecosystem, a decrease in organic traffic, and an increase in the overall cost of conducting business online.
The patent does not arrive in isolation. It fits within a broader sequence of Google infrastructure changes documented by PPC Land over the past 18 months that together suggest a systematic effort to intermediate the relationship between users and advertisers within Google's own surfaces.
On January 11, 2026, Google launched the Universal Commerce Protocol, establishing open-source standards for AI agents to execute purchases across retail platforms. That same day, Target and Walmart enabled checkout directly within Google's Gemini and AI Mode, embedding full transactions inside AI-generated responses. On February 11, 2026, Google unveiled a shopping ad format for AI Mode, which by that point had reached over 75 million daily active users.
In October 2024, Google overhauled its Shopping platform with Gemini-powered AI, incorporating personalized feeds, dynamic filters, and AI-generated product briefs. Ads began appearing in AI Overviews for US mobile users in October 2024, and the format expanded to 11 countries in December 2025.
The navigation link to the AI-generated page can, under the patent's claims, appear within a sponsored content item - meaning advertisers could be billed for traffic sent to a page they did not create, brand, or control. Whether Google intends to implement this system commercially, and under what disclosure or opt-in requirements, is not addressed in the patent document itself.
The patent also has a technical relationship to a previously cited document in its references. The patent cites US20250086246A1, a Google application titled "Landing Page Optimization Using Machine-Learning Techniques," filed September 12, 2023, as a related work. It also cites US9767157B2, a 2017 Google patent titled "Predicting site quality."
The scoring mechanism described in the patent operates as an automated quality gate. A landing page with a poor conversion rate, high bounce rate, low click-through rate, or inadequate design quality may trigger AI page generation. The patent also states that the system triggers when a landing page "does not have a filter for products" - a functional deficiency rather than a purely qualitative judgment.
This threshold-based logic differs from how quality signals currently function in Google Ads, where landing page qualityaffects Quality Score and minimum CPCs but does not replace the page itself. The patent describes a system where Google makes the replacement decision autonomously, without advertiser input or consent - at least as currently architected.
Critically, the AI-generated page can, according to claim 2 of the patent, "be presented to another organization that is different from the first organization," and "be utilized for future search." That means a page generated in the context of one advertiser's query response could potentially surface for a different brand's query at a later time.
The system's reliance on previous queries and contextual user account information raises questions under data protection frameworks, particularly the EU's General Data Protection Regulation. Google's parallel European application, EP25191927.0A, filed July 25, 2025, suggests the company is pursuing protection for the technology across major markets.
The processing of search history to generate a personalized page constitutes profiling under the GDPR's definitions. Whether the generation of a third-party AI page using that profile data, and its display as a search result, constitutes lawful processing under Article 6 of the GDPR has not been tested. Google has been navigating similar tensions in other product lines: non-personalized ad controls in European AdSense were deprecated in May 2025, and Google's Privacy Sandbox program was substantially retired in October 2025 after adoption failures left most of its replacement technologies without sufficient industry uptake.
The ANA figure cited by Squires - $26.8 billion in lost global media value in 2025, up 34% from 2024 - reflects how existing automation already erodes advertiser control. A system capable of replacing the destination page itself would represent a more fundamental shift in that dynamic.
Who: Google LLC, with inventors Caren Zeng, Rushil Grover, Timothy Benjamin Whalin, Lauren Marjorie Bedford, Pallavi Satyan, and Ethan Milo Mann. Industry reaction came from SEO and digital advertising professionals including Joshua Squires (Amsive), Hank Azarian, Eric Hoover, and Lily Ray.
What: A patent granted today, US12536233B1, describes a system that scores an organization's landing page using conversion rate, bounce rate, click-through rate, and design quality. If the score exceeds a threshold - or if the page simply lacks a product filter - the system generates an AI replacement page, personalized using the user's previous search queries and contextual signals, and displays a navigation link to it in the search results. The link can appear within a sponsored content item.
When: The provisional application was filed July 25, 2024. The formal US application was filed January 3, 2025. The patent was granted and published January 27, 2026. The European parallel application was filed July 25, 2025.
Where: The patent was filed with the United States Patent and Trademark Office and has a parallel European application. The technology would operate within Google Search's results infrastructure.
Why: The patent describes the system as improving user experience for people who struggle to navigate poorly designed or low-converting landing pages, while improving performance metrics for content providers by generating pages that better match individual user intent. Industry professionals have raised concerns about brand control, attribution transparency, organic traffic displacement, and increased advertiser dependency on Google's ecosystem.
