If you have been told to write for "information gain," here is the short version. Google holds a patent for a method that scores how much new information a document might give a particular reader beyond what that reader already saw. That is confirmed: the patent exists, Google LLC is the assignee, and it was granted in 2022. What is not confirmed is whether Google actually runs that method inside ordinary search rankings today. The honest grade on "information gain SEO" as an active ranking factor is weak. The claim that Google patented the idea is confirmed. Those are two different sentences, and most of what you read online quietly merges them into one.
This matters because "information gain" gets pitched as the next big content-quality signal, the kind of thing that should reshape your whole content plan. Before you reorganize a team around it, it is worth knowing exactly what has been proven and what has been assumed, and whether information gain deserves to be called a ranking factor at all in the way people use that word.
What the claim says and where it came from
The phrase traces back to a real document: a Google patent titled "Contextual estimation of link information gain". The U.S. application published as US20200349181A1 on November 5, 2020, and the corresponding patent, US11354342B2, was granted on June 7, 2022. The listed inventors are Victor Carbune and Pedro Gonnet Anders, and the record shows an international filing date of October 18, 2018. Google LLC is the assignee of record.
The patent's abstract defines an information-gain score as an indication of the additional information a document contains, beyond what is already present in documents a particular user has previously viewed. A machine-learning model produces that score from document data, and the resulting scores can shape how a set of documents gets presented to that user.
SEO writers picked up the phrase and have mostly simplified it into something easier to act on: write things your competitors have not already covered, and you will score better on information gain. Amanda King wrote about this for Search Engine Land on July 25, 2023, describing information gain as a measure of how unique a piece of content is relative to the pool of documents that could answer a given query. King frames this as an interpretation of what Google might be doing, not as proof that it is doing it. That framing got lost on the way to a lot of content briefs.
What the patent actually describes
Read the patent itself rather than the summary of the summary, and a narrower picture shows up. The described process is contextual: it compares a document against what a specific user has already seen, not against every other page on the internet that targets the same keyword. The system can identify documents a user has already been shown, identify candidate documents the user has not seen, score those candidates for likely new information, and use the scores to decide what to present next.
The patent also describes results being reranked as new information becomes available, and it lists possible machine-learning inputs such as document contents, extracted salient information, or semantic representations of a page. None of that amounts to a formula you can calculate yourself. There is no published score, no stated threshold for what counts as "high" information gain, and no public evidence of a production rollout attached to the patent record.
One detail trips people up: the patent title contains the word "link." That does not make this a conventional link-authority signal like PageRank. The scoring described in the patent is about documents and the information inside them, evaluated against what a user has already encountered, not about the links pointing at a page.
What supports the claim that Google could use this
The strongest piece of evidence is the patent itself. Google LLC disclosed and was granted a method for calculating information-gain scores and using those scores, at least in part, to guide how a set of documents is ranked or presented. That is direct evidence that Google built, and patented, a technique capable of doing exactly what "information gain SEO" claims it does.
Roger Montti covered the patent for Search Engine Journal on August 15, 2024, and connected it to follow-up documents and to the broader idea of additional value beyond material a searcher has already seen. Montti raises the possibility that the technique could be relevant to AI-assisted search experiences. He presents that as a plausible connection worth watching, not as a confirmed fact. A patent existing, and a journalist noting it could matter for AI search, is useful context. It is still not the same claim as "Google uses this to rank pages right now."
What the evidence does not show
Here is where the weak grade comes from. Google's own public reference, titled Guide to Google Search Ranking Systems, last updated December 10, 2025, describes how the company's automated ranking systems work: many factors and signals, applied at the page level, with site-wide signals and classifiers contributing as well. The guide does not name information gain anywhere in it.
That silence is not proof that the method is unused. Google frames the guide as covering some of its more notable ranking systems, not as a complete inventory of every signal that touches a ranking decision. So the absence cannot be read as a denial. What it also is not, though, is confirmation. There is a real difference between "Google has not said this is active" and "Google has said this is not active," and the honest answer here is the first one, not the second.
Put the two primary sources side by side and they answer different questions. The patent answers whether Google disclosed a method where information-gain scores can affect how documents are presented: yes, clearly. The official ranking guide does not answer whether that particular method runs inside current Search, AI Overviews, or AI Mode. It simply does not say, one way or the other. King's 2023 piece and Montti's 2024 piece are both useful records of how the SEO industry read the patent, but neither one supplies direct evidence that the technique is running in live search results today.
Why this keeps getting repeated as settled fact
A patent is a strong-sounding word. It reads like proof, and a filing with Google LLC's name on it feels like it should settle the question. But holding a patent on a method and deploying that exact method inside a production ranking system are two different events, and plenty of granted patents never ship in the form they were written. Nothing in the public record reviewed here demonstrates that this specific scoring method made it from patent filing into the live Search stack, as written, with this exact mechanism intact.
There is also a second collapse happening alongside the first. "Information gain" as the SEO industry uses it, meaning content that adds something competitors missed, is a much broader idea than the patent's actual definition, which is about one document compared against what one particular user has already viewed. Treat those as the same thing and you end up telling clients that Google computes an "information gain score" for every page on a topic and ranks accordingly. That specific, universal, query-wide score has not been shown to exist anywhere in the sources reviewed here.
This is not the first time a patent has been treated as a confirmed information gain ranking factor when the public record only supports a narrower claim. SEO has a long history of this exact pattern: a company files a patent for a technique, the patent becomes public, and the industry skips straight from "they could do this" to "they definitely do this, and here is how to optimize for it." The same thing has happened with Google's Search Quality Rater Guidelines, debated for years as a hidden ranking factor despite Google saying human ratings do not directly control rankings. Anyone curious about that parallel case can read what the guidelines actually say about good content, which turns out to be a similar story of a real document stretched past what it actually says. A patent is evidence that an idea was worth protecting. It is not evidence that the idea shipped unchanged into the product people use every day.
How confident should you actually be
It helps to separate three different levels of confidence, because "information gain" gets discussed as if it only has two settings: real or fake. The patent's existence is a fact, not a belief: anyone can read US11354342B2 and confirm Google LLC holds it. Whether some version of the underlying idea, scoring new information against what a user has already seen, shows up anywhere in Google's current systems is plausible, given that Google files plenty of patents that later inform real features in some form. But plausible is not the same as proven, and nothing in the sources reviewed here moves "plausible" up to "proven." The same gap shows up in the long-running debate over click-through rate as a ranking factor, where leaked documents show click-derived signals matter somewhere in the system without confirming that raw click-through rate itself is the input. The narrowest, least supported claim here is the one repeated most often: that there is a specific, calculable "information gain score" that ranks pages against every competing page on a topic, and that writing to it will move your rankings. That version is the one with no public evidence behind it at all.
The verdict, and what to do differently
Grade: weak, for the claim that information gain functions as an active Google ranking factor. Confirmed, for the narrower claim that Google patented a method using information-gain scores. Keep those two sentences separate every time this topic comes up, because conflating them is exactly how the myth spreads.
Does Google measure information gain today, inside ordinary rankings? The public record does not settle it either way. That is an honest unknown, not a disappointing one. You do not need Google to confirm a named algorithm signal before "does this page add something a reader could not already get elsewhere" becomes a useful editorial question. Use it as a lens for deciding what to cover and how deeply, not as a target score to hit or a guaranteed ranking lever. A page that genuinely adds detail, context, or a clearer explanation than what is already out there tends to earn attention on its own merits, patent or no patent. That standard holds up even as search itself changes, including the open question of whether traditional SEO fundamentals still matter once AI answer engines are doing more of the retrieval.
If you are briefing writers, the practical move is to stop asking "does this hit our information gain score" and start asking what a specific reader, with a specific question, does not yet know that this page could tell them. That question holds up regardless of what Google's ranking stack turns out to contain, and it keeps your content decisions anchored to evidence rather than to a term that sounds more settled than it is. Teams that produce a lot of content, especially at the pace content marketing leads are now expected to hit, can lose this discipline fast once volume goes up and every brief starts repeating the same borrowed claims. Keeping facts and interpretation separate in your own content process, the same way this review tries to, is a habit worth protecting as output scales. It is also worth asking, on any page that claims to add something new, how unique a piece of content actually needs to be before that uniqueness is worth anything at all, since the honest answer is less about a threshold and more about whether the page genuinely serves the reader.
What would change this verdict
The grade would move from weak to confirmed if Google explicitly stated that this information-gain method runs inside a current ranking system, or if the official ranking-systems guide named it as an active signal, or if credible, reproducible evidence demonstrated the score operating in production. None of the following would be enough on their own: another patent grant, another SEO blogger's interpretation of the existing patent, or a correlation study showing that distinctive content tends to rank well. Correlation between unique content and good rankings is a real and separately useful finding, but it does not prove this particular patented mechanism is the reason.
