What Is Entity?

How Entities Work

Turning text into entities happens in two steps. First, a system spots the candidate things in a sentence — this is named entity recognition, which flags spans like “Marie Curie” or “Warsaw” as probable entities and guesses their type. Second, it resolves each candidate to a specific identity, linking “Marie Curie” to one node in the Knowledge Graph rather than to the string of letters. Resolution is the important half: it is what lets the system know that two differently worded mentions point at the same thing, and that one identically spelled mention points at two different things.

Once an entity is identified, a system can attach information to it. Google’s Cloud Natural Language API makes this concrete. Its analyzeEntities method returns, for each entity it finds, a type, a list of mentions in the text, and a salience score in the 0 to 1.0 range that measures how central the entity is to the document. When the entity is known, the response also includes metadata: a mid — the Knowledge Graph machine identifier, like /m/0cqt90 — and a wikipedia_url. That mid is the entity’s true name as far as the machine is concerned; the human-readable label is just one of many strings that point to it.

This is why entity-based search behaves so differently from string matching. A page that never uses a query’s exact keyword can still be recognized as being about the entity that query refers to, because the surrounding text, the linked facts, and the co-occurring entities all corroborate the identity. The words are evidence; the entity is the conclusion.

Types of Entities

Systems that detect entities assign each one a type, which constrains what facts make sense for it. Google’s Natural Language API uses categories including:

• PERSON — an individual, real or fictional.
• LOCATION — a place, from a city to a landmark.
• ORGANIZATION — a company, agency, team, or institution.
• EVENT — a happening with a time and place.
• WORK_OF_ART — a book, film, song, or similar creative work.
• CONSUMER_GOOD — a product or model.

Each mention is also tagged PROPER or COMMON, distinguishing a named entity (“Paris”) from a generic noun (“a city”). The type is not decoration — it tells the system that a PERSON can have a birth date while an ORGANIZATION can have a founding year, which shapes every fact the graph will accept about the node.

Example of an Entity

A documented, reproducible example comes straight from Google’s analyzeEntities reference. Feed the API a sentence such as “Lawrence of Arabia is a highly rated film biography about British Lieutenant Colonel T. E. Lawrence,” and it does not return keywords. It returns entities. “Lawrence of Arabia” comes back with the type WORK_OF_ART, a high salience because it is the subject of the sentence, and a wikipedia_url plus a Knowledge Graph mid identifying that specific film. “T. E. Lawrence” comes back as a separate PERSON entity with its own mid. “British” resolves as a LOCATION-flavored entity, lower in salience because it is peripheral to the sentence.

Two properties of that output are the whole point. First, identity: the API distinguishes the film named after the man from the man himself, even though both mentions share the word “Lawrence,” because each carries a different mid. A pure keyword system would see one repeated string; the entity system sees two distinct things. Second, salience: the same word can score near 1.0 in a document that is about it and near 0 in a document that mentions it in passing, which is how a machine decides what a page is really about versus what it merely name-drops.

This is the shift that reshaped search and now drives AI answers. When an assistant summarizes a topic, it is assembling facts about resolved entities, not stitching together keyword matches. Being understood as a clear, distinct entity — with a consistent name, a resolvable identity, and the right neighboring entities around it — is what puts a brand into that reasoning at all.

Entity vs Keyword

Keywords and entities are layers, not rivals: a query still arrives as words, but modern search resolves those words to the entity behind them before deciding what to show. Optimizing the string still helps you get found; making the entity unmistakable is what gets you understood.

Frequently Asked Questions

The Bottom Line

An entity is a thing, not a word: a person, place, organization, or concept that machines can pin to one identity no matter how it is spelled or phrased. Search and AI systems have shifted from matching the letters in a query to recognizing the entity behind it, which is why being understood as a clear, distinct entity now matters more than owning a particular keyword string.