A Google patent has been circulating since the start of the year, usually under a headline saying Google has bought the right to replace your website. One of the more widely read write-ups puts it exactly that way, and the same patent number has appeared in enough posts and threads since that most people who own a site have now heard some version of it.
The document is real. It is worth reading rather than reading about, because the version that has spread is not quite the version that was granted, and the differences change what you would sensibly do about it.
What the document is
It is US12536233B1, titled AI-generated content page tailored to a specific user, assigned to Google. The earliest priority date is July 25, 2024, the formal filing is January 3, 2025, and it was granted on January 27, 2026. There is a European counterpart. All of that is on the page linked above, and so is the full text.
The first claim describes a system that takes a query from a user, builds a results page carrying a result for a business, calculates what it calls a landing page score for that business’s landing page, and then builds an updated results page carrying a link to an AI-generated page for that business instead. Later claims fill in what the generated page can hold: a call to action pointing at a product page, a product feed, an assistant to chat with, a headline written for the person looking, suggested filters, and suggested groupings of products. Another claim has the model reading the person’s previous searches to decide what to put there.
So the shape of the story is accurate. A page can be scored, and a generated alternative can be offered in its place. Three details in the coverage are not accurate, and each one matters.
The score runs the other way
Almost every version of this says the generated page appears when your score falls below a threshold. The claim says the opposite. The generated page appears when the score exceeds one.
That sounds like a quibble until you look at how the document defines exceeding. One claim gives a worked example: the score exceeds the threshold when the landing page does not have a filter for products. It is not a quality grade where high is good. It is closer to a measure of what is missing, where a higher number means there is more room for something else to do a better job.
The practical consequence is that the most common piece of advice attached to this story, watch your landing page experience score in your ads account, is aimed at a different number. Whatever the score in this document is, it is not that one, and no published threshold exists for it.
It is not specific to organic results
Several write-ups describe the scored page as the top organic result. The word organic does not appear anywhere in the patent. What does appear is a claim stating that the link to the generated page can sit inside a sponsored content item, which is to say an ad.
Read with that claim in view, the document looks less like a plan to intercept ordinary search results and more like something aimed first at paid placements. That is a considerably narrower thing. It also puts a genuinely awkward question on the table for anyone buying ads, which is who pays for a click through to a page they did not write, and nothing in the document answers it.
What is not in the document at all
This is the one worth knowing, because the advice attached to this story almost everywhere rests on it.
The usual counsel is that the generated page is built from your own content, so the cleaner and more complete your product information and trust signals are, the more control you keep over how you are represented. It is a reasonable guess. It is not in the patent. There is no mention of crawling, of an index, or of reading the content of the page being replaced. Your landing page is the thing that gets scored, and the described inputs to the generation itself are the current query, an account profile holding preferences and previous searches, and a data resource the document never defines.
Maybe an implementation would draw on your pages. It would be the obvious way to build it. But it is an assumption, and it has been repeated often enough to start sounding like a finding. Anybody telling you that structured content is your defense here is telling you something the document does not say.
What a patent is
A granted patent is permission to stop somebody else from building a thing. It is not a commitment to build it, a schedule, or a product. Google files these in very large numbers, and most of what is in them never ships in the form described. Nothing here has been announced, and Google has not commented on it.
What does deserve weight is the company it keeps. This patent cites an earlier Google filing on optimizing landing pages with machine learning, and it lands in the same stretch as shopping ads inside AI answers, checkout happening inside an assistant rather than on a retailer’s site, and a protocol for agents to buy things across platforms. Those are shipped, not filed. Any one patent is weak evidence. A direction of travel with products already in it is not.
What this actually asks of your site
Less than the headlines suggest, and nothing that is new.
The only concrete trigger the document names is structural. Not a tone of voice, not a design trend, not a score you cannot see: a missing filter on a page selling products. That is the same class of problem as a price that only appears after a script runs, or a product page that does not say what the product costs. It is a page failing to do an ordinary job. Whether or not this system ever exists, a page like that is already losing people who arrive on it.
Which lands where this work usually lands. A site that says plainly what it is, loads its content before anything else has to happen, lets somebody narrow a long list, and agrees with itself across every page, is a site that is hard to improve on by generating something else. That was true before this patent and it will be true after it. What is SEO, and why it matters if you own a website covers the groundwork in full.
The part genuinely worth watching is not your own site at all. It is how much of a search result is still a link to somewhere else. That has been shrinking for years, through answers printed on the results page, through summaries above the links, and now through transactions completed without leaving. This patent belongs to that sequence rather than to a separate emergency, and the defense against all of it is the same one: some share of your customers reaching you by a route that does not pass through a search result.
One thing to do
Open the patent yourself. It is public, and the claims are in plain enough language that twenty minutes gets you further than any summary of it, this one included. Then go and look at whichever of your own pages sells the most, and ask the only question the document actually raises about it: if somebody landed here wanting one specific thing out of everything you offer, could they narrow it down, or would they have to scroll?
Written September 2026, from the text of US12536233B1 as published. Dates, claim numbers, and quoted wording were read from the patent itself rather than from coverage of it. Nothing here has been announced as a product, and no threshold, score, or launch date for this system has been published.