What is information gain in SEO?
Information gain is, in SEO, the idea that a page is worth more when it adds facts the reader has not already seen on other pages about the same topic. The name comes from a Google patent filed in 2018, which scores documents by how much new information they offer a particular user. Google has not confirmed it as a ranking factor.
Key takeaways
- Information gain measures how much new information a document adds compared with what a user has already seen on the same topic.
- The term comes from a Google patent filed in October 2018 and granted in June 2022, which describes it mostly for an assistant that converses with the user.
- A patent protects a method: it does not show that Google uses it, where, or with what weight.
- What is documented is something else: Google asks whether your page adds substantial value compared with the other pages in the results and asks for content that is not interchangeable.
- For a store, the practical question is what your product page says that the manufacturer's and your competitors' do not.
In SEO, "information gain" is used for a simple idea: a page is worth more when it says something the reader has not already read on the others. It is not the information gain of machine learning decision trees, which shares the name and little else. It reaches a store from two sides: through agency pitches and tools that promise to "optimize information gain", and through AI search, where an assistant reads several pages about the same product and composes a single answer. Before paying for it, separate two things that often get mixed up: what a Google patent describes and what Google has documented about how it ranks.
What the patent describes
The name comes from a Google LLC patent titled Contextual estimation of link information gain, with Victor Carbune and Pedro Gonnet Anders as inventors. It was filed in the United States on October 18, 2018 and granted on June 7, 2022 as US11354342B2. The family has at least four US patents with the same title; the latest was filed in May 2024.[1]Google Patents (Google LLC; inventors Victor Carbune and Pedro Gonnet Anders). US11354342B2: Contextual estimation of link information gain
What it describes is an information gain score for each document: how much additional information it contains beyond that of the documents the user has already seen. The score is computed by running data about the documents through a machine learning model: their full content, the highlighted information extracted from them or a semantic representation, such as embeddings. With those scores, new documents can be ranked, and re-ranked as the user sees more.[1]Google Patents (Google LLC; inventors Victor Carbune and Pedro Gonnet Anders). US11354342B2: Contextual estimation of link information gain
The patent's example is a computer problem: someone asks for help, reads one document about software crashes and another about hardware repair, and the system then prefers a document that brings something those two did not have. The first claim, the one that defines the protected method, talks about an automated assistant that responds to a natural-language request, presents information from a first group of documents and picks the next one by its score.[1]Google Patents (Google LLC; inventors Victor Carbune and Pedro Gonnet Anders). US11354342B2: Contextual estimation of link information gain The text also mentions search engines that show a summary next to the links, as another place where information from several documents is presented.[1]Google Patents (Google LLC; inventors Victor Carbune and Pedro Gonnet Anders). US11354342B2: Contextual estimation of link information gain
Two traits matter for reading it correctly. The score is not a fixed property of a page: it depends on what each user has seen before. And a patent protects a method; it does not say whether Google uses it, in which product or with what weight.
What is documented and what is not
| Claim | Documented? | Where |
|---|---|---|
| Google has a patent for scoring a document's new information against what has already been seen | Yes | Patent US11354342B2[1]Google Patents (Google LLC; inventors Victor Carbune and Pedro Gonnet Anders). US11354342B2: Contextual estimation of link information gain |
| Google uses that score to rank its results | No | Google's guide to its ranking systems, which lists the ones it uses and the retired ones, names none for information gain[2]Google Search Central. A guide to Google Search ranking systems |
| Google tries to show original content ahead of that of those who only cite it | Yes | Its original content systems, including original reporting[2]Google Search Central. A guide to Google Search ranking systems |
| Google asks whether your page adds something the others do not have | Yes, as guidance | The helpful content guide[3]Google Search Central. Creating helpful, reliable, people-first content |
| For its AI features, Google asks for content that is not interchangeable | Yes, as guidance | Its guide to generative search[4]Google Search Central. Optimizing your website for generative AI features on Google Search |
| An outside tool can give you your information gain score in Google | No | Google says no third-party tool has access to its internal ranking or AI systems[4]Google Search Central. Optimizing your website for generative AI features on Google Search |
What is documented, then, is not the patent but the guidance. The helpful content guide asks whether the content provides substantial value when compared to other pages in search results, whether it offers analysis that goes beyond the obvious and whether, when it draws on other sources, it avoids copying or rewriting them and adds substantial value and originality. Among the warning signs it lists summarizing what others say without adding much.[3]Google Search Central. Creating helpful, reliable, people-first content The guide to Google's generative AI features goes the same way: its AI systems look at many sources, so having a point of view that stands out helps; a first-hand review brings a perspective of its own, while a summary of what has been published only repeats what is already somewhere else.[4]Google Search Central. Optimizing your website for generative AI features on Google Search
In a store
A fictional example: a photography store sells a 25-liter camera backpack. The manufacturer publishes its page and three other stores copy its text. What each page adds on the questions a shopper asks:
| What the shopper asks | Manufacturer | The other three stores | This store |
|---|---|---|---|
| Capacity | 25 liters | 25 liters | 25 liters |
| Does it fit as carry-on? | Does not say | Does not say | Outside dimensions measured in the store, to compare against your airline's limits |
| What gear fits? | "For camera and lenses" | Same | Tested with a mirrorless camera, two lenses and a 14-inch laptop |
| Does it hold up in rain? | "Water resistant" | Same | Rain cover included, and which pocket it lives in |
The three columns agree on the first row and part ways on the rest. Those rows are the page's information gain, whether or not the patent is in use: what an assistant or a shopper does not find repeated elsewhere. And it comes from something only the seller of the product has: having measured it and tested it.
What people often confuse
That information gain is a ranking factor confirmed by Google. What exists is a patent that describes a score relative to what each user has already seen, designed mostly for an assistant that converses.[1]Google Patents (Google LLC; inventors Victor Carbune and Pedro Gonnet Anders). US11354342B2: Contextual estimation of link information gain Google's documentation on its ranking systems does not mention it.[2]Google Search Central. A guide to Google Search ranking systems What Google does ask for in writing is to add value compared with the other pages, and that is not achieved by adding sections or words: Google says it has no preferred word count.[3]Google Search Central. Creating helpful, reliable, people-first content
How to check it
- Type the main query for a product page and open the top results. Note, one per row, the facts each page gives.
- Mark the rows only your page has. If there are none, your page repeats what is already published, which is exactly what Google's guide advises against.[3]Google Search Central. Creating helpful, reliable, people-first content
- For each row of your own, note where it comes from: a measurement, a test, a customer's question, a fact from your sales. A fact with no origin is the first one anyone will question.
- Check that those rows are in the page's visible text, and not only in an image or an attached PDF.
- Be wary of anyone offering "your information gain score" in Google. Google says no third-party tool has access to its internal systems.[4]Google Search Central. Optimizing your website for generative AI features on Google Search A tool can compare your text with others, but that comparison is its own, not Google's.
Related concept: helpful content
Helpful content is the framework Google documents: content created for people and not for the search engine, with its self-assessment questions.[3]Google Search Central. Creating helpful, reliable, people-first content Information gain is an industry idea, taken from a patent, that covers only one part of that framework, originality compared with what is already published. A page can have a lot of information gain and still not be reliable, for example if the new facts are wrong; that is why E-E-A-T, with trust as its most important aspect, completes the picture.[3]Google Search Central. Creating helpful, reliable, people-first content
To go further, the AI search reading path orders the rest of the terms, and the answer engine entry explains how an assistant composes an answer from several sources.
Synonyms
- English
- information gain score
- Español
- ganancia de información, puntuación de ganancia de información
Sources
- [1] Google Patents (Google LLC; inventors Victor Carbune and Pedro Gonnet Anders). US11354342B2: Contextual estimation of link information gain. Accessed . ↑ ↑ ↑ ↑ ↑ ↑
- [2] Google Search Central. A guide to Google Search ranking systems. Accessed . ↑ ↑ ↑
- [3] Google Search Central. Creating helpful, reliable, people-first content. Accessed . ↑ ↑ ↑ ↑ ↑ ↑
- [4] Google Search Central. Optimizing your website for generative AI features on Google Search. Accessed . ↑ ↑ ↑ ↑
Change history
- First English version.
This definition is published under CC BY 4.0. Reuse this definition freely, crediting Clione with a link to this page.
Related terms
Helpful content
What helpful content is for Google, what the helpful content system was from 2022 to 2024, what it became and how to apply it to an online store.
E-E-A-T
What E-E-A-T (experience, expertise, authoritativeness and trustworthiness) is, what Google says about it and which trust signals an online store needs.
Answer engine
What an answer engine is, the three ways it answers (knowledge base, snippet or generative answer) and where your store can show up.
LLMO
What LLMO (large language model optimization) is, how it differs from GEO and what a store can control about what ChatGPT says about it.
Scaled content abuse
What Google's scaled content abuse policy bans, the examples it gives, where AI fits, how bulk pages in a store can trip it, and how to audit your site.
Product description
What a product description is, what Google Merchant Center and AI assistants ask of it, and how to review it without mixing it up with the meta description.
Make your catalog easy for bots to read
Clione fills every product page with what bots read best: the JSON-LD graph with its structured data, the meta description and the FAQs. You review and press Push; then we check your live store so it all stays published.