Information Gain: What a Page Adds That the Index Does Not Already Hold
Information gain is the amount of new information a document adds beyond what a reader or the index already holds, and Holistic Radar scores it as information gain density, worth 15 of 100 points in the Holistic Authority Score. This article covers what information gain is, the Google patent that defines an information gain score, and what Google’s own guidance asks for in place of the term. It then sets out information gain density with its qualifying and non-qualifying elements, a worked calculation, an audit of a core section, and the relation of information gain to the other components of the score. It explains how to add information gain to a page, what information gain is not, information gain on holisticradar.com, and information gain in AI answers, and closes with questions on ranking, unique content, original research, measurement, and cited statistics.
What information gain is
Information gain is the amount of new information that a document adds beyond the documents a reader has already seen. A document that repeats earlier material adds little, and a document that states something new adds more.
Information gain is a property of a document in relation to other documents, and not a property of a document alone. The same page has high information gain on a topic where no other page states its facts, and low information gain on a topic where ten pages already state them.
The Google patent on information gain scores
Google holds a patent titled “Contextual estimation of link information gain”, US 11,354,342, issued on 7 June 2022 from an application filed on 18 October 2018. The inventors are Victor Carbune and Pedro Gonnet Anders. The patent describes an information gain score for a new, unviewed document that estimates how much information the document adds beyond the documents a user has already viewed.
| ELEMENT OF THE PATENT | WHAT THE PATENT DESCRIBES |
|---|---|
| Input | A set of documents the user has already viewed and a set of new documents on the same topic |
| Method | A machine learning model that reads semantic vectors, embeddings, or bag-of-words data and returns a score |
| Ranking use | New documents are ranked or re-ranked by score, and the ranking can change as the user views more documents |
| Assistant use | An automated assistant selects the new document with the highest gain and presents its information |
| Results list use | References to redundant documents can be demoted or removed when the user returns to a results list |
A patent grant describes a method and does not confirm that Google Search uses the method in ranking. The patent is evidence of how Google frames the problem: a document that repeats what the reader already holds is a candidate for demotion.
The patent measures information gain against the documents one user has viewed, so the same document can score high for one user and low for another. Holistic Radar measures information gain against the index, which is a fixed baseline: an element qualifies when no competing document in the index states it. The two baselines answer different questions, and the Holistic Authority Score uses the second because a practitioner can audit it.
Information gain in Google’s own guidance
Google’s guidance on helpful content asks the same question in different words. The guidance asks whether content offers original information, reporting, research, or analysis, whether it covers its topic thoroughly and goes beyond obvious points, and whether it adds real value compared with other search results instead of copying or rewording other sources.
Google also warns against two imitations of newness. The guidance advises against changing publication dates to make pages seem fresh when the content has not substantially changed, and against adding or removing large amounts of content only to appear fresh. Information gain is therefore a property of the facts a page states and not a property of its date or its length.
Information gain density
Information gain density is the proportion of core section pages that carry at least one element the index does not already contain. Holistic Radar measures it as one of the five components of the Holistic Authority Score, and the component is worth 15 points.
The formula is fixed: information gain density points equal core pages with a qualifying element, divided by total core pages, multiplied by 15. The unit is the core page and not the sentence, because a page either holds an element no competing document states or it does not.
| QUALIFYING ELEMENT | WHAT IT IS | EXAMPLE ON HOLISTICRADAR.COM |
|---|---|---|
| Original number from practice | A figure that comes from Holistic Radar’s own work | PropxPro: +47% demos and a 22% shorter sales cycle in 60 days |
| Original framework component | A named part of a method that no other document defines | The four types of content cannibalization |
| Original comparison | A comparison built from the site’s own criteria | Holistic SEO definitions compared by organizing principle |
| Original observation | A finding from the site’s own data that no competing document states | Six Library notes that were published at two URLs each during the migration |
Three kinds of content score zero. A restated definition scores zero, because the index already holds the definition. A cited third-party statistic scores zero, because the statistic belongs to its first publisher. A paraphrased competitor argument scores zero, because the argument is already in the index under another author’s name.
A worked calculation of information gain density
The worked calculation of the Holistic Authority Score runs against a core section of 34 pages, of which 6 carry a qualifying element. The ratio is 0.176, and 0.176 multiplied by 15 returns 2.6 points.
| CORE PAGES WITH A QUALIFYING ELEMENT | RATIO | POINTS OUT OF 15 |
|---|---|---|
| 6 of 34 | 0.176 | 2.6 |
| 12 of 34 | 0.353 | 5.3 |
| 24 of 34 | 0.706 | 10.6 |
| 34 of 34 | 1.000 | 15.0 |
In the worked calculation, information gain density scored 2.6 of 15 and was the largest gap in proportional terms. The remediation order therefore placed original elements in the core section ahead of further node coverage. Each additional core page that gains a qualifying element is worth 0.44 points on a 34-page core section.
The table shows the shape of the formula. The formula pays the same 0.44 points for every page: the six pages from 6 to 12 add 2.6 points, and the ten pages from 24 to 34 add 4.4 points. The formula therefore rewards completion of the core section and not a few strong pages, and a core section in which every page carries a qualifying element reaches the full 15 points.
Auditing information gain across a core section
An audit of information gain checks every core page against the index and records one result per page. The definition of a qualifying element requires the check, because an element qualifies only when no competing document states it.
| STEP | ACTION | RESULT RECORDED |
|---|---|---|
| 1 | List the core section pages | The denominator of the formula |
| 2 | List the factual claims on each page that come from the site’s own work | Candidate elements per page |
| 3 | Search the index for each candidate and read the top competing documents | Whether any competing document states the element |
| 4 | Classify each surviving candidate as a number, a framework component, a comparison, or an observation | One element type per qualifying page |
| 5 | Count the pages with at least one qualifying element | The numerator of the formula |
A page with no surviving candidate scores zero, and the page joins the remediation list. Recalculation follows the schedule of the Holistic Authority Score: quarterly, or after any publication run that adds 20 or more pages.
Information gain and the other components of the score
Information gain density is one of five components, and each of the other four interacts with it. The interaction explains why the Holistic Authority Score sequences information gain after coverage and before further expansion.
| COMPONENT | POINTS | RELATION TO INFORMATION GAIN |
|---|---|---|
| Tuple coverage | 30 | Coverage decides which tuples a page states, and information gain decides whether the page states anything new about them |
| Extraction integrity | 20 | A qualifying element is stated as a triple, so the same sentence earns both components |
| Network integrity | 20 | Duplicate coverage across the site’s own pages adds no gain and raises cost of retrieval |
| Historical depth | 15 | Historical depth counts months of continuous publication and queries ranked, so a qualifying element published early starts its history early |
Gains at the Contested band, 60 to 79, come from information gain and historical depth and not from more pages. A site at that band already covers its map, so the next point comes from stating something the index does not hold.
How to add information gain to a page
A page gains information gain when it states one fact that no competing document states, in a sentence that a machine can extract. The audit lists the qualifying elements on each core page, and a page with none receives one.
| ORIGINAL ELEMENT | SOURCE OF THE ELEMENT | HOW THE SENTENCE STATES IT |
|---|---|---|
| Number from practice | A client record | Subject, result, unit, period, and scope in one sentence |
| Framework component | A named method | One fixed name with its attributes, repeated unchanged |
| Comparison | The site’s own criteria | The items compared, the attribute compared, and the result in one table |
| Observation | The site’s own data | The date range, the sample size, and the finding in one sentence |
The sentence that states the element is a triple: a subject, a precise predicate, and an object that carries a number, a date, or a named quantity. Extraction integrity, another component of the Holistic Authority Score, rewards the same sentence form, so one sentence earns points in both components.
What information gain is not
Information gain is not length. A longer page restates more of what the index holds, and a restatement scores zero. Information gain is not a new angle on old facts either, because a paraphrased competitor argument scores zero.
Information gain is not freshness. Google’s guidance treats a changed date without a changed substance as an imitation of freshness, and the Holistic Authority Score treats a restated definition with a new date as zero.
Information gain is not a difference between a site’s own pages. Two pages of the same site that state the same tuples add nothing to each other, which is the condition of near-duplication in content cannibalization.
Information gain is not a reader-experience measure. The Holistic Authority Score does not measure whether a reader enjoys the page, and a page can hold a qualifying element while being unpleasant to read.
Information gain on holisticradar.com
Holistic Radar measures information gain density on its own site from three sources: named methods, published map rows, and client results. Each source produces qualifying elements that a competing document does not state.
| SOURCE | QUALIFYING ELEMENT |
|---|---|
| Named methods | The Six-Layer Pyramid, the Holistic Authority Score, the Radar Scan, and the four types of content cannibalization |
| Published map rows | The tuples of the topical map, published as headings and table rows |
| Client results | PropxPro: +47% demos in 60 days. Vital Transportation: +312% organic growth and $1.4M in pipeline in 90 days |
Every client result carries its period, and every named method carries a fixed name. Holistic Radar keeps a locked vocabulary, so a method name appears unchanged wherever it is used, which keeps each qualifying element attached to one entity.
Information gain in AI answers
An automated assistant that selects the document with the highest information gain avoids repeating what the user already heard, according to the patent. A page that restates the index offers the assistant nothing to select.
A qualifying element is also the sentence an answer engine can attribute. A client result stated with its subject, unit, period, and scope can be quoted as a single claim with a source, while a restated definition has many equal sources.
→ Six-Layer Pyramid: where information gain sits among the six layers, and why page semantics comes before retrieval.
Information gain questions on ranking, unique content, original research, measurement, and cited statistics
Is information gain a Google ranking factor?
Google has not confirmed information gain as a ranking factor. Google holds a patent for an information gain score, and Google’s helpful content guidance asks whether a page offers original information and adds value compared with other results.
Is information gain the same as unique content?
Information gain is not the same as unique content. A page can have unique wording and no new fact, and unique wording around an old fact scores zero.
Does information gain require original research?
Information gain does not require formal research. A number from practice, a framework component, a comparison built from the site’s own criteria, and an observation from the site’s own data each qualify.
How is information gain measured?
Holistic Radar measures information gain density as core pages with a qualifying element divided by total core pages, multiplied by 15. The patent describes a different measure, a machine learning score per new document, which Holistic Radar does not claim to reproduce.
Does a cited statistic count as information gain?
A cited third-party statistic scores zero in the Holistic Authority Score, because the statistic belongs to its first publisher. A cited statistic supports a claim, and an original element adds one.
Information Gain in the semantic SEO topical map
Information Gain is the added value attribute of semantic SEO, the central entity of the Holistic Radar topical map, and sits in the outer section under the Know predicate.
The outer section holds the page because the page explains a ranking input without a direct sale, and its bridges pass session signals toward Radar Scan.
→ Historical Data in SEO: the record that the gain is compared against
→ How to Write a Content Brief: where the gain target is set
Sources
The information gain score is described in US Patent 11,354,342, the originality questions come from Google Search Central, and the density formula, element types, and worked calculation come from the Holistic Authority Score.
- Google, US Patent 11,354,342, Contextual estimation of link information gain, 2022
- Google Search Central, Creating helpful, reliable, people-first content
- Holistic Radar, Holistic Authority Score
- Holistic Radar, Content cannibalization