What Is Entity Density?

How Entity Density Works

An entity is a thing an engine can recognize and disambiguate: a named person, a place, an organization, a product, or a concept that exists as a node in a knowledge graph. Entity density describes how many of the relevant entities for a topic a page actually names, and how clearly. It is a coverage measure, not a frequency measure — the question is not “how often does the keyword appear” but “does this page mention the concepts a knowledgeable writer could not avoid mentioning.”

Machines read entities in two steps. First, named-entity recognition (NER) scans the text and tags spans that look like entities. Second, entity resolution links each tagged span to a canonical node — matching the string “Apple” to the company rather than the fruit by reading the surrounding context. Once entities are resolved, the engine scores how central each one is to the document. Google’s Natural Language API exposes this directly as a salience score, defined on a 0 to 1.0 scale, where a value near 1.0 means the entity is central to the page and a value near 0 means it is incidental.

This is also how large language models make sense of a passage. A model doesn’t retrieve on keyword match; it represents your text as concepts and relationships. A page that names the specific entities a topic implicates gives both a search engine and an LLM more hooks to understand what the page is about, judge whether it is comprehensive, and decide whether to retrieve or cite it. Sparse, generic prose that repeats one phrase gives them almost nothing to anchor to. High entity density is, in effect, a legible signal of expertise — you tend to name real things only when you actually know the subject.

Retrieval-heavy systems reward this even more directly. In retrieval-augmented generation, a passage is pulled into an answer because its embedded meaning matches the query’s meaning. Entity-rich passages sit closer to the queries they should answer, so they surface more often. Density here is not a dial you turn to a target percentage; it is a byproduct of covering the topic properly.

Example of Entity Density

Google’s own Natural Language API makes entity density concrete. Its Analyzing Entities documentation runs a short passage through entity analysis and returns a salience score for each recognized entity. In the documented example, the API identifies entities and ranks them by centrality to the text:

• Trump (person) — salience 0.79
• White House (location) — salience 0.092
• Pennsylvania Ave NW (location) — salience 0.086
• Washington, DC (location) — salience 0.029

The salience score is defined in the Entity reference as a value in the 0 to 1.0 range indicating the entity’s importance or centrality to the entire document. That single mechanism is exactly what “entity density” is describing from the writer’s side: a page’s substance is the set of resolvable entities it contains and how central each is, not the count of any one phrase.

Consider two drafts targeting the query best noise-cancelling headphones. Draft A repeats “best noise-cancelling headphones” a dozen times and hits a textbook keyword density, but names nothing specific. Draft B mentions the Sony WH-1000XM5, Bose QuietComfort Ultra, Apple’s AirPods Max, the H2 processor, active noise cancellation, ANC, transducer size, and Bluetooth LE Audio. Run both through entity analysis and Draft B lights up with a dense, salient entity set while Draft A resolves to almost nothing. To an engine, Draft B demonstrably covers the topic; Draft A only claims to. (This two-draft contrast is an illustration of the mechanism, not a measured Google result.)

Entity Density vs Keyword Density

The two ideas are constantly confused because both sound like “how much of my topic is on the page.” They measure opposite things. Keyword density is a legacy metric — the percentage of total words that match a single target phrase — built for an era when search engines genuinely counted term frequency. Entity density is about coverage of distinct recognized concepts, the signal modern engines and language models actually use.

The critical fact underneath the confusion: keyword density is not a Google ranking factor, and per Google it never really was. Google published a video on the topic as far back as 2011 telling site owners to stop chasing a target percentage, and analyses of top-ranking pages routinely find keyword densities scattered from under 1% to over 5% with no correlation to position. Repeating a phrase past the point of natural usage does nothing for ranking and reads as manipulation. Covering the entities a topic requires is what correlates with topical comprehensiveness.

The practical takeaway is that “density” was the right instinct pointed at the wrong object. The goal was always to prove a page covers its subject. Term frequency was a crude proxy that engines abandoned; entity coverage is the thing that proxy was reaching for. Optimize for the entities — see the full entity density vs keyword density comparison — and you optimize for how understanding actually works, both in classic search and in generative engines that read your page as concepts rather than strings.

Frequently Asked Questions

The Bottom Line

Entity density reframes on-page relevance from "how many times did I say the keyword" to "how completely did I cover the concepts this topic is made of." Engines and language models resolve meaning through recognized entities and how central they are to the page, not through repetition counts. Write like an expert naming real things, and the coverage takes care of itself — the keyword percentage was always a proxy for something it never actually measured.