What is entity salience?

Entity salience is how central a particular entity is to a piece of text. In an article titled "How Spotify pays artists," Spotify is the most salient entity; Taylor Swift, mentioned once as an example, has low salience. Both are entities in the text, but only one is what the text is about.

An entity is a distinct, identifiable thing: a person, organization, place, product, event or concept. Named entity recognition finds the entities in a text. Entity salience then ranks them by importance to the document as a whole.

In Google's Cloud Natural Language API, the tool most SEOs use to see salience, every entity gets a salience score between 0 and 1.0. Google's documentation says the score "provides information about the importance or centrality of that entity to the entire document text. Scores closer to 0 are less salient, while scores closer to 1.0 are highly salient" [4].

How is entity salience calculated?

Entity salience is calculated by a machine learning model that weighs signals such as how often an entity is mentioned, where it first appears, and its grammatical role. Google has not published the exact model behind the Cloud Natural Language API's score, but the research it grew from is public.

In April 2014, Google researchers Jesse Dunietz and Daniel Gillick published "A New Entity Salience Task with Millions of Training Examples" at EACL [1]. Their insight was a way to label salience automatically at scale: an entity counts as salient if it also appears in the human-written summary of the article. Using the New York Times annotated corpus, they labeled 110,639 documents containing 2,229,728 entities, about 14% of which were salient [2].

What the model learned is the most useful part for writers [2]:

  • Mention count was the strongest signal, counting every way the entity is referred to: by name ("Spotify"), by description ("the streaming company") and by pronoun ("it").
  • Position of the first mention mattered: entities introduced in the first sentence are more likely to be salient.
  • The head word of the mention also contributed.

A simple rule, "the entity is salient if it is mentioned in the first sentence," scored an F1 of 46.2%. The full model reached 61.6% [2]. Put plainly, position helps, but sustained reference throughout the text matters more.

What does a salience score mean?

A salience score is a relative measure of how much a document is about an entity, compared with the other entities in the same document. It is not a quality score, a ranking score, or a measure of how well the page matches a query.

Three practical points that most explanations skip:

  • Scores are comparative within one text. A score of 0.3 means little alone. What matters is whether the target entity is the top-scoring entity, and by how much.
  • Changing other entities changes the score. Adding a long section about a competitor lowers the target entity's share, even if the target entity's own mentions are unchanged.
  • The score depends on the text submitted. Pasting a full HTML page with navigation and footer into the API produces different results from pasting only the main content.

The Cloud Natural Language API also attaches metadata to many entities. Google's documentation says that for most entity types this is "a Wikipedia URL (wikipedia_url) and Knowledge Graph MID (mid), if they are available" [3]. An entity that comes back with a Knowledge Graph MID has been matched to a known thing in Google's Knowledge Graph, which Google introduced in 2012 [8].

Does Google use entity salience for rankings?

Google has not said that Google Search uses entity salience, or the Cloud Natural Language API's salience score, as a ranking factor. The API is a commercial product for developers. Its score is a window into how Google-built language models read text, not a readout of Search's ranking systems.

Most careful SEO coverage says the same. A 2019 Smart Insights article states that entity salience "should not necessarily be considered a ranking factor" [5]. A 2019 G2 guide acknowledges "there's no sure-fire way to optimize content for entity salience, nor are we certain entity salience means higher rankings" [6].

What is known versus claimed

Known: Google researchers formalized entity salience as a task in 2014 [1]; Google's public API scores it from 0 to 1.0 [4]; and Search has been entity-based since the Knowledge Graph [8]. Claimed but unsupported: that a specific salience score (such as "0.5 or higher") is needed to rank, that salience is a direct ranking factor, or that raising a page's API score will raise its position. Search Insider found no published test that isolates salience from the many other changes made when a page is rewritten.

When did entity salience become an SEO topic?

Entity salience was formalized as a research task by Google researchers in 2014 and became an SEO topic around 2019, once practitioners began testing pages in Google's Cloud Natural Language API [1] [7]. It rests on Google's shift to entity-based search in 2012 [8].

2012: Search becomes entity-based

Google launched the Knowledge Graph on May 16, 2012, describing the change as "things, not strings" [8]. Once search was organized around entities, the question of which entity a page is mainly about became a practical one.

2014: salience defined at scale

Jesse Dunietz and Daniel Gillick presented "A New Entity Salience Task with Millions of Training Examples" at EACL in Gothenburg in April 2014, defining salient entities as those that appear in an article's summary [1] [2].

2019: SEOs adopt the salience score

In 2019, several SEO publications explained how to read the Cloud Natural Language API's salience score, including Majestic in March 2019 [7], Smart Insights in October 2019 [5] and G2 in November 2019 [6]. Most were careful to present the score as a diagnostic, not a ranking factor.

What is the difference between entity salience and keyword density?

Entity salience measures how central an entity is to the meaning of a text, while keyword density counts how often an exact string appears relative to total words. Keyword density has no documented threshold in any search engine's guidance, and repeating a keyword does not reliably raise salience.

The 2014 research explains why: salience counts every reference to an entity, including descriptions and pronouns, and weighs where and how the entity appears [2]. A page that names a brand twenty times in a list of links can score lower than a page that names it four times but makes it the subject of the opening sentence and every paragraph.

What is the difference between entity salience and relevance?

Entity salience describes a document on its own, while relevance describes how well a document matches a specific query. Salience asks "what is this page about?" Relevance asks "does this page answer what was searched?" A page can be highly salient for "Spotify" and still be irrelevant to the query "Spotify student discount" if it never covers discounts. Salience is a property of the text; relevance only exists in relation to a search.

How do you check entity salience?

To check entity salience, paste a page's main text into the free demo of Google's Cloud Natural Language API and read the Entities results, which list each entity with its salience score. The demo has been available on Google Cloud's Natural Language product page, where anyone can paste text and see entity, sentiment, syntax and category results without writing code [6] [7].

A useful routine:

  1. Paste only the main content, not navigation, sidebars or footer.
  2. Check that the target entity is the top-scoring entity, and note the gap to the second.
  3. Check whether the target entity has a Knowledge Graph MID or Wikipedia link attached, which shows it was matched to a known entity [3].
  4. Run the same test on the top three ranking pages for the target query and compare which entities they make salient.

Search Insider's practical check, borrowed from how the 2014 researchers defined salience [2]: write a one-sentence summary of the page. If the target entity would not naturally appear in that summary, the page is not really about it, whatever the score says.

How do you improve entity salience?

To improve entity salience, make the target entity the grammatical and structural subject of the text. Based on the signals in the published research [2]:

  • Name the entity in the title, the first heading and the first sentence.
  • Keep referring to it throughout, by name and by clear descriptions, rather than drifting to other subjects.
  • Make it the subject of sentences ("Spotify pays artists per stream") rather than an object buried in a clause.
  • Move long digressions about other entities to their own pages.
  • Use the entity's full, unambiguous name at least once so it can be matched to the right Knowledge Graph entry.

When should you care about entity salience?

Care about entity salience when a page is meant to be about one specific thing, such as a brand's About page, a product page or a glossary entry, and diagnostic tools suggest search engines read it as being about something else. Salience is a useful editing check for focus and clarity. It is not a number to chase: a page that clearly serves the reader on one subject will usually make that subject salient as a side effect.