Information gain — what Google's patent describes and why it matters
What is information gain?
Information gain is a concept described in Google patent US11354342B2, "Contextual estimation of link information gain," filed in 2018 and granted in 2022. The patent describes estimating how much additional information a document offers beyond what the user has already seen in other documents; the patent alone does not prove that Google uses such a signal in ranking. In SEO, information gain is understood as how much new, unique information your content brings relative to pages already ranking for the same query.
In practice: if 10 pages in SERP say the same thing, the 11th page with identical content doesn't deserve ranking. A page that adds a new angle, data, experiments, fresh perspectives brings information gain — and gives a reason to show it alongside, or instead of, pages that repeat the same thing.
Why does it matter?
- Helpful Content — according to Google's guide to its ranking systems, the system announced in 2022 was designed to ensure people see original, helpful content written for people; since March 2024 it has been part of Google's core ranking systems. Google has not confirmed that these systems use the mechanism from the patent, but the direction is the same
- AI Overviews and GEO — according to Google's documentation on AI features in Search, the same best practices apply there as in regular SEO, and content that repeats other pages gives no reason to pick it in particular
- Defensive SEO — a page that only repeats the competition has no argument with which to outrank it
What exactly does the patent describe?
The patent does not talk about comparing a page with the top ten results; it describes the situation of a specific user:
- Documents already seen — the system identifies a set of documents on a topic whose information the user has already encountered (in the patent's claims, among other things, information previously presented to them by an automated assistant)
- New documents on the same topic — candidates whose content the user has not yet seen
- Information gain score — for each new document, the system estimates how much information it adds beyond what was already seen; the patent describes, among other things, applying a machine learning model to semantic feature vectors. Based on that, the new documents can be ranked and information from the selected one presented to the user
The patent alone does not show that Google uses this mechanism to rank search results. The practical tips below follow from the idea of the patent, not from a documented algorithm.
How to increase information gain?
1. Audit SERP before writing
- Check top 10 pages for your query
- Extract their H2/H3 (topic structure) and dominant entities
- Identify gaps — topics covered by 1-2 pages but not the rest
2. Bring unique data
- Internal metrics — own case studies, A/B tests, Google Search Console data
- Fresh benchmarks — comparison of multiple tools/strategies with your setup
- Experiments — description of test and conclusions nobody published yet
- Insider knowledge — market context, product decisions, expert perspectives
3. Add entities missing from corpus
- Enrich text with named entities (people, companies, tools, standards, dates)
- Each entity is a potential topical signal — link to entity in knowledge graph
4. Atomic answers
- Look at content as a set of atomic answers (entity + image + content atoms)
- Each fragment must carry one unique piece of information extractable by AI
- Helpful formats: FAQ, definitions, statistics, step-by-step lists
Common mistakes
- Paraphrasing top-rankers — such content adds no new information (zero gain), regardless of whether any tool catches it: according to the study "Can AI-Generated Text be Reliably Detected?", paraphrasing can significantly reduce the effectiveness of AI detectors, and Google does not publicly describe using third-party detectors. An AI detector score is not proof of a text's origin or quality either.
- Theory only, no practice — corpus already contains theory; add "how we did it"
- No numbers — raw data (percentages, KPIs, times) is a strong gain signal
- Copy-pasted FAQ from competitors — Google groups duplicates and picks one URL as canonical, not necessarily yours
Information gain and GEO
In the context of GEO (Generative Engine Optimization) the idea is the same: a generative answer in ChatGPT, Perplexity or Gemini combines many sources, and a page that adds nothing beyond them gives the engine no reason to cite it in particular. This is reasoning, not a documented mechanism — the patent concerns Google, and a citation in AI answers cannot be guaranteed.
Related terms
- Topical authority — thematic authority
- Topical mapping — content cluster mapping
- E-E-A-T — expertise signals
- GEO — AI optimization
- Helpful content — helpful content
- SERP — search engine results page