Entity salience in SEO: how to build clear topical focus and authority

If Google can’t resolve what a page is mainly about, adding backlinks may not be the fix. Before you assume a ranking problem needs more links, check the cheaper thing first: is the page’s main entity clear enough for Google to work with? Below is the mechanism behind salience scoring, plus a worked numeric example. Then the steps I use to raise it, on the page itself and across the outside sources that describe a brand.

TL;DR

• Entity salience measures how central an entity is to a piece of content, not how often it is mentioned. Google scores it from 0 to 1, and grammatical role and position typically outweigh repetition.
• Consistent mentions across outside sources make a brand easier for retrieval systems to reconcile as one clear entity. Where profiles already exist on Wikipedia, Wikidata, review sites and in the press, their core facts shouldn’t contradict your own site.
• One URL should carry one dominant entity, named in the title, H1, and opening sentence. Splitting broad, mixed-topic pages into focused ones is usually the highest-leverage fix.
• Entity salience is not the same thing as keyword density. Repeating a term more often can lower salience if it dilutes context, while fewer, well-placed, well-connected mentions raise it.
• Structured data and internal linking sharpen what salience is inferred from. JSON-LD and descriptive anchor text give Google a machine-readable version of what your own prose already implies; neither replaces the underlying content.
• Small, focused sites benefit disproportionately from this work. A hundred tightly focused pages can outperform ten thousand pages that never resolve what the site is actually about.

What is entity salience?

An entity is a distinct, identifiable thing: a brand, person, product, place, or an abstract concept such as ergonomics or home lighting. Google’s Cloud Natural Language API, a public developer tool, assigns each entity a type, a list of mentions, and a salience score between 0 and 1. Search is another matter: Google has not stated that it uses this exact score as a ranking factor. Treat the number as a diagnostic model of how Google’s systems are likely to read a page, not a confirmed live metric.

Entity salience measures how central a specific entity is to the overall meaning of a piece of content, not how many times its name appears. A page can name an entity once, in the right place, and score higher than a page repeating the same name five times in passing. Entities do more work for Google than salience scoring, and how Google uses entities to understand content covers the rest.

A Cloud Natural Language example

Google’s own Cloud Natural Language documentation analyzes this sentence: “President Trump will speak from the White House, located at 1600 Pennsylvania Ave NW, Washington, DC, on October 7.” In the published output, “Trump” scores about 0.79 salience, while “White House” gets about 0.09 despite also appearing exactly once.

So the gap is not about frequency. “Trump” is the sentence’s subject and main topic; “White House” is a location detail mentioned in passing. Google does not publish which exact features drove each number, but the pattern matches what its documentation describes: position, grammatical role, and how central an entity is to the passage’s meaning carry more weight than mention count.

Entity salience vs. keyword density

Keyword density measures how often a term appears relative to a text’s total word count. It is a lexical count. Meaning, grammatical role, and how the term connects to the rest of the text play no part in it.

Entity salience asks a different question: how central a concept, person, organization, or thing is to what the document is actually about. Optimizing for salience means adding context and relationships, not repeating the name more often.

Signal	Keyword density	Entity salience
What it counts	How often a term appears	How central an entity is to the meaning
Typical range	A percentage of total words	A 0 to 1 score
Driven by	Repetition	Position, grammatical role, context, relationships
Risk of overdoing it	Keyword stuffing, diluted readability	Topic drift; irrelevant context can dilute the main entity’s clarity

A 5-step process to increase entity salience

Information architecture: one URL, one main topic, one leading entity

Each URL should focus on a single dominant entity. Make the URL, the title tag, and the H1 say the same thing, then let the introduction confirm it in the first line.

Example H1: “Wingback armchair: how to choose the right one for your living room.” Example intro: “A wingback armchair is a classic furniture entity: tall backrest, curved wings, and padded armrests. Below we compare upholstery types, dimensions, and positioning for optimal comfort.”

From my own audit and strategy work, splitting broad, mixed-topic articles into focused, single-entity URLs can improve long-tail visibility by 30 to 50 percent.

Content: describe entities through attributes and relationships

Expand the topic through properties, meaning the entity’s own features, and relations to other entities. Connect “armchair” to “upholstery fabric,” “living room lighting,” or “Scandinavian interior style.”

Mix specific names with general terms. Google’s language models lean on that variety to place an entity in the right class alongside related concepts.

Internal linking: signal what each page is about

• Use descriptive anchor text linking to category or pillar pages.
• Avoid orphan pages, and connect everything through menus, breadcrumbs, and contextual links.
• Place important contextual links early, inside the introductory paragraphs; links in the first 250 to 300 characters carry strong semantic weight for entity recognition.

Structured data and HTML that make entities explicit

Schema markup and clean code help Google’s systems disambiguate entities instead of guessing at them.

• Add JSON-LD for products, reviews, FAQs, or articles, but only when every marked-up value, including ratings, matches what is genuinely visible on the page.
• Use proper title, H1, alt, and aria-label attributes.
• Keep key copy available in HTML rather than hidden behind JavaScript that needs to render first.

Minimal JSON-LD example:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Wingback Armchair LARS",
  "description": "High-back lounge chair with distinctive wings and soft armrests.",
  "brand": {"@type": "Brand", "name": "MebloHaus"},
  "aggregateRating": {"@type": "AggregateRating", "ratingValue": "4.7", "reviewCount": "128"}
}
</script>

Only include an aggregateRating when the rating and review count are real and shown on the page itself. Markup that does not match visible content risks a manual action, not just wasted effort.

Verification: measure salience and behavioral impact

Analyze text with Google Cloud Natural Language or a comparable NLU tool to see detected entities and their salience scores directly.

Cross-reference the result with Google Search Console: check which queries and pages are actually driving impressions and clicks. If a page ranks for unrelated queries, or Google seems to blend it with a different page on the site, the entity focus is usually unclear. Weak internal linking, overlapping content worth consolidating, and a search-intent mismatch worth rechecking are the other usual suspects.

Building entity consensus across sources

On-page entity clarity can support relevance, though Google hasn’t confirmed the Cloud API’s salience score itself as a Search ranking input. Cross-source consensus works on a different question, and an earlier one. Does your brand get treated as a single, real, well-defined entity at all, across sources you do not control? I treat this as a working model of how retrieval systems are likely to reconcile brand mentions, not a documented selection rule Google has published.

The mechanism is comparison. When Wikipedia, Wikidata, review platforms, press coverage, and your own site all describe the same name with the same category and the same core facts, that agreement counts as evidence. Retrieval systems have less reason to treat the entity as a stable, citable answer when those sources conflict, or barely mention it at all.

• Keep your brand’s name, category, and core description consistent across your site, social profiles, directories, and press mentions. “Software company” on one profile and “consulting agency” on another works against you.
• If your brand has a Wikipedia article, disclose your connection and use its talk-page edit-request process for substantive changes rather than editing it directly, that’s a conflict of interest under Wikipedia’s own rules. Wikidata has separate policies: add only verifiable claims and cite proper sources.
• Use sameAs in your Organization or Person schema to link your site’s entity to the external profiles that already exist for it, such as Wikidata, LinkedIn, or Crunchbase.
• Monitor press mentions and reviews for accuracy. A wrong category or an old product name repeated across several sources works against consensus, not for it.

That is a different discipline from on-page work: building and defending a brand’s identity as a search entity in its own right. It runs past what one article covers. Treat the off-page half as its own ongoing effort.

Editorial checklist for entity-first content

1. Define the topic early. Name and classify the entity in the opening sentence, such as “A wingback armchair is a type of…”, and disambiguate names that could mean something else, such as “Apple Inc.” rather than just “Apple.”
2. List key attributes. Dimensions, materials, use cases, whatever actually describes the entity.
3. Describe relations. Comparisons, alternatives, examples.
4. Include evidence. Diagrams, tables, captions, or descriptive alt text.
5. Follow the user’s journey. Add “how to choose,” common mistakes, and FAQ sections that match what someone actually needs next.

Application examples

E-commerce

On a category page, the job is building strong salience for an entity like “wingback armchair.”

If a product grid, headings, product names, and structured data don’t make the category’s scope clear on their own, add copy that does: a definition paragraph, detailed attributes, an FAQ. Back it with structured data that matches what’s visible, plus contextual links to guides such as “how to choose an armchair height.”

YMYL: finance and health

Goal: reinforce topical trust on sensitive subjects where accuracy carries real consequences.

Tactics: entity clusters such as glossary pages, transparent sources and methodology, and balanced pros-and-cons sections. Google’s Reviews system guidance belongs on pages actually structured as product or service reviews, though, and a generic finance or health explainer does not qualify just by adding a pros-and-cons list.

Content hub or blog

Goal: dominate a topic cluster such as “living room lighting.”

Tactics: one pillar article, supported by focused sub-articles on floor lamps, bulb color temperature, and layout, interlinked around the same set of entities.

Where entity salience work does not pay off

Entity salience optimization is not a fix for every ranking problem, and treating it as one wastes effort. There are a few situations where it generally will not move the needle.

• The content itself has nothing new to say. Clear entity focus on thin or duplicate content still leaves thin or duplicate content. It just makes the sameness easier for Google to detect.
• The site has a crawling or indexing problem underneath it. If Google is not reliably crawling or indexing the page, no amount of on-page entity clarity gets evaluated in the first place.
• The domain has little external trust on a competitive, sensitive topic. On YMYL subjects especially, on-page salience helps Google understand a page; it does not substitute for the off-page trust those topics also require.
• Mentions are split across pages competing for the same entity. That splits internal signals and forces Google to choose between competing URLs. Consolidating and redirecting usually helps more than adding another page.

A frequent technical cause is easy to miss: JavaScript that fails, stalls, or renders inconsistently by the time Google evaluates the page, so key copy never makes it into what gets read. Check the actual rendered HTML first. Move critical copy to server-side rendering when it is missing there, or late, or only present some of the time.

How to measure and report entity performance

1. Use Google Cloud Natural Language or a comparable tool to compare salience scores across page drafts, as a diagnostic, not as a client-facing KPI.
2. Monitor keyword coverage for entity-related and attribute-related queries in Search Console.
3. Watch CTR, conversions, and rankings together, but read them separately: CTR also reflects position and snippet wording, while conversions depend on the offer and page experience.

Why entity salience matters for Google’s ranking systems

Several of Google’s documented ranking systems overlap with this work, though Google has never published a direct link between Search ranking and the Cloud API’s specific salience score.

• Reviews system: rewards pages with expert analysis and original research about specific products or services. Clear entity references help Google interpret what is actually being reviewed.
• Link analysis and PageRank: evaluates how pages link to each other to understand what a page is about. Descriptive anchor text is a separate, long-standing practice that gives extra context to both readers and crawlers.
• Core ranking systems: absorbed the separate Helpful Content System in March 2024, per Google’s own ranking-systems documentation, so content-quality signals are now part of ongoing core ranking rather than a standalone, periodic update.

Alongside these named systems, Google does document a topic authority system. It is narrower than how SEO practitioners usually use the term “topical authority”: it evaluates publisher expertise for news and newsy queries, and weighs signals such as a track record of original reporting. Broad “topical authority,” in the SEO-industry sense of being a consistent, trustworthy source on a subject, describes a general outcome these systems reward, not that specific named system. Steady entity salience across a site’s content still strengthens it.

Google patents relevant to entity salience

These are the patents assigned to Google that sit closest to how entity salience and entity-based ranking work. None of them confirms any specific method is active in Search today. Links go straight to the filings.

Patent	Number, year	What it covers
Techniques for automatically identifying salient entities in documents	US9619457B1, 2017	Trains a classifier on document-abstract pairs: entities that show up in a document’s own abstract are treated as salient, an abstract-based method for creating salient-entity training labels.
Ranking search results based on entity metrics	US10235423B2, 2019	Scores knowledge-graph-derived search results using weighted metrics that vary by entity type, then combines them into a single ranking score.
Document ranking based on entity frequency	US9679018B1, 2017	Ranks documents partly on how rare or common their entities are across a corpus, surfacing content about less-common topics that still match user interest.
Associating an entity with a search query	US9336211B1, 2016	Maps queries to relevant entities to improve both results and query suggestions tied to a specific entity.

The patent above lists Daniel Gillick as an inventor, and he co-authored the Dunietz and Gillick paper below. Both cover the same abstract-based approach to identifying salient entities.

Academic research relevant to entity salience

Paper	Authors, venue, year	Relevance
A New Entity Salience Task with Millions of Training Examples	Dunietz and Gillick, Google, EACL 2014	Defines entity salience as a machine learning task, using the entities in a document’s own abstract as ground truth for which entities are salient.
Entity Query Feature Expansion Using Knowledge Base Links	Dalton, Dietz and Allan, University of Massachusetts Amherst, SIGIR 2014	Uses knowledge-base links (Freebase, the Google Knowledge Graph) to expand entity queries and improve retrieval; relevant to entity-aware search, not direct proof that document salience is a ranking factor.
Entity Linking via Joint Encoding of Types, Descriptions, and Context	Gupta, Singh and Roth, EMNLP 2017	Improves entity disambiguation by jointly modeling an entity’s type, description, and surrounding context, rather than name matching alone.

How this research connects to the 2024 Google leak

In May 2024, more than 2,500 pages of internal documentation for Google’s Content Warehouse API became public. A Google spokesperson later confirmed the documents were authentic, while cautioning against drawing conclusions from them. Fair caution. Leaked attribute names are not proof of how the live algorithm weighs anything today, so treat them as a working model of Google’s likely internal vocabulary rather than a confirmed, current ranking formula.

Leaked attribute name	What the name and surrounding docs suggest
EntityAnnotations	Appears to attach Knowledge Graph entities extracted from a page. The docs do not confirm a direct link to the patents above; treat that as this article’s own inference, not a documented connection.
topicEmbeddingsVersionedData	Appears to be versioned vector embeddings of a page’s entities and topics. The leak does not document an exact data flow into salience or site-focus scoring; treat any link between them as inferred, not confirmed.
siteFocusScore / siteRadius	Appear to measure how concentrated a site is on one topic, and how far a given page strays from that core topic.
contentEffort	Appears to estimate the human effort a page represents. Independent leak-analysis sites describe it as language-model-scored, but the exact method is not documented.
OriginalContentScore	Appears to estimate content originality. Independent leak-analysis sites report conflicting ranges for it (some say 0 to 127, others 0 to 512); treat any specific number as unconfirmed.

FAQ

If you want this diagnosed on your own site rather than applied generically, an SEO audit is where I’d start. Once you already know which pages are competing for the same queries, or ranking for the wrong ones, book a strategy call instead.

Pozdrawiam serdecznie!

Direction and strategy: Szymon Slowik. Research and drafting support: LLM. Reviewed, corrected and finalized: Szymon Slowik.

Created by Szymon, edited with an LLM… see any difference? 🙂