What is a knowledge graph?

What is a knowledge graph?

A knowledge graph (“graphe de connaissances” in French) represents knowledge as a graph: entities connected by relationships.

The authors acknowledge that several definitions, some of them contradictory, coexist. Lisa Ehrlinger and Wolfram Wöß (Johannes Kepler University Linz, 2016) ask for more. In their view, a knowledge graph integrates information into an ontology and applies a reasoner to derive new knowledge. They noted that the term had been circulating without a definition since the launch of Google’s database in 2012.

The basic unit is the triple. The W3C RDF 1.1 Recommendation (February 25, 2014) describes it as three components: a subject, a predicate and an object. The predicate is always an IRI, an identifier shaped like a web address.

Example: (Elev8 Lab, has author, Baptiste Clair). A second triple, (Baptiste Clair, works as, SEO and GEO consultant), extends the graph from the same node. In GEO, a brand describes itself this way, through explicit entities and relationships. These two triples illustrate the model; they say nothing about what Google knows about Elev8 Lab.

Knowledge graph, database or ontology: what’s the difference?

These terms overlap without being synonyms. The table is based on Hogan et al. (2021) and on the notes of Stanford’s CS520 course (2020).

On the standards side, RDF 1.1 is still the latest W3C Recommendation, and RDF 1.2 has been at the Candidate Recommendation stage since April 7, 2026.

What is Google’s Knowledge Graph?

Google’s Knowledge Graph is the model Google uses to connect real-world entities and their relationships. Google announced it on May 16, 2012 under the phrase “things, not strings”.

The post by Amit Singhal, then Senior Vice President of Engineering, takes the example of the query [taj mahal]. It can refer to the monument, a musician, a casino or a restaurant. The graph is there to resolve this ambiguity.

At launch, Google cited public sources such as Freebase, Wikipedia and the CIA World Factbook. In 2020, it describes a system fed by web content, open databases and licensed databases. Wikipedia is a frequent source among hundreds, including licensed data (music, sports, television) and medical partnerships.

In December 2012, Google explained that the expansion was not a simple translation: “football” does not mean the same sport in the United States and in Europe. No official figure more recent than the 2020 one was found from Google (check of October 19, 2026).

Knowledge Graph, knowledge panel, Business Profile: what’s the difference

The Knowledge Graph is the database; the knowledge panel is one of the ways it is displayed. According to the Google Help page About knowledge panels, a knowledge panel appears when you search for an entity that is in this database: a person, place, organization or thing.

The panel is generated automatically from web sources, sometimes from partners (movies, music) and from changes suggested by verified entities. It updates on its own.

The Google Business Profile follows a different path. It is for businesses that serve customers at a location or within a service area, and it is managed in Google Business Profile (formerly Google My Business). The featured snippet is yet another feature of the results page.

How does a brand or a person get into the Knowledge Graph?

Google documents no sign-up procedure. Its two “Knowledge panel” help pages, read on October 19, 2026, only describe claiming an existing panel and sending feedback.

Google does, however, describe two uses of the structured data it finds on the web (Google Search Central, updated on December 10, 2025). It helps Google understand pages and gather information about the world: people, books or companies included in the markup.

Organization markup on the home page can help Google tell an organization apart. According to its documentation (updated on September 8, 2026), some properties can influence the logo shown and the knowledge panel; none is required. Our guide to structured data covers this markup in detail.

The schema.org sameAs property points to a reference page that identifies the item unambiguously: its Wikipedia page, its Wikidata item or its official website. The rest of the identity work (canonical name, About page, official profiles) is described in our method for the brand entity.

Claiming and correcting a knowledge panel

If a knowledge panel already exists for you or your organization, Google lets you claim it. The help page Get verified on Google describes the procedure.

1. Google Account: sign in to the account that will manage the panel.
2. Search: search Google for your name or the entity’s name and find the knowledge panel.
3. Claim: at the bottom of the page, in the “About” section, click “Claim this knowledge panel”.
4. Verification: sign in to one of the official sites or profiles offered (YouTube, Google Search Console, Twitter or Facebook). Otherwise, provide additional information.
5. Representatives: once your identity is verified, add authorized representatives if needed, with or without admin rights.

Not every knowledge panel can be claimed. The “Feedback” link then lets you request an update, a correction or a removal, with links to web pages that prove the information. On mobile, it sits behind the three dots at the top of the panel.

In its 2020 post, Google acknowledges that its automated checking systems are not perfect: hence this reporting option open to everyone.

How can you check whether your entity exists in the Knowledge Graph?

First test: search for your name. If a knowledge panel appears, the entity is in Google’s database.

Second test: Google’s Knowledge Graph Search API, which returns entities in JSON-LD format, with schema.org types. A request queries the address https://kgsearch.googleapis.com/v1/entities:search with three parameters: query (the name searched for), limit (the number of results) and key (a Google Cloud API key).

Google points out three limits. The API is read-only. It is not suited to a critical production service. Finally, Google is migrating it to Cloud Enterprise Knowledge Graph, which it recommends to new users.

The identifiers returned start with /g/ or /m/. Wikidata documents the former in its property P2671 (“Google Knowledge Graph ID”, alias kgmid); the latter fall under the Freebase ID (P646). A practitioner’s method, not a Google rule: the Wikidata QID is used in public markup (sameAs), the Google identifier is used for diagnosis.

What role does the knowledge graph play in AI search engines?

At Google, this role is stated. Google introduced its AI Mode on March 5, 2025 (Google blog). On top of web content, it gives access to fresh, real-time sources, including the Knowledge Graph, real-world information and shopping data. Our article on Google AI Mode explains its “query fan-out” technique.

Google’s guide to generative AI features (updated on July 10, 2026) states that structured data isn’t required for generative AI search. There is no special schema.org markup to add; Google advises keeping it in your SEO strategy.

Outside Google, the knowledge graph also serves LLMs. Most RAG (retrieval-augmented generation) systems look for passages by vector similarity before answering. With GraphRAG (Microsoft Research, February 13, 2024), an LLM first builds a knowledge graph from a private corpus, then this graph enriches the prompt at query time.

The related research paper (Darren Edge et al., 2024) targets global questions, such as “what are the main themes in this corpus?”, on which classic RAG fails. Microsoft, which designed the method, reports substantial improvements on complex documents. For mainstream engines, see how AI search engines cite their sources.

How do you build a knowledge graph?

You start with the ontology: which classes of entities, which relationships between them. Then comes the choice of a technical model. Hogan et al. cite several; two families serve as reference points.

So Neo4j is not a knowledge graph: it is a graph database, used to build one. Elev8 Lab has no commercial relationship with the vendors mentioned.

At the scale of a website, the same logic applies to markup. Each entity on the site (organization, author, offer) gets a stable @id in JSON-LD, and its relationships become triples. Some practitioners call this an “internal knowledge graph”. A sameAs does not replace a contextual link between two pages: this graph complements internal linking. Its effect on citation by AI search engines has not been established.