What is the Knowledge Graph?

Three things get mixed up in day-to-day work and are worth keeping apart. The entity is the real-world object: a company, a municipality, a drug. The Knowledge Graph is the store where Google keeps those entities with their attributes and their links. The knowledge panel is only one of the surfaces where that store becomes visible, and its absence does not prove the entity is missing from the graph.

Nor is it the same as the search index. The index holds documents: URL, text, crawling signals. The graph holds facts about things and the connections between them, with its own identifier for each node. A company can have a hundred indexed URLs and no recognised node, and the reverse happens too: a person with no website of their own appears in the graph because other sources describe them.

Structured data is an entry channel into the graph, not the graph. Marking up Organization on the home page helps Google understand the organisation's administrative details and tell it apart from others with a similar name, according to the Search Central documentation updated on 15 April 2026. That same page warns that Google does not guarantee that features which consume structured data will appear in the results.

The graph is built from sources that compile factual information. Google describes its base as billions of facts about people, places and things, and acknowledges that it also licenses third-party data for blocks such as sports scores, stock prices or weather forecasts. The rest comes from crawling the open web and from public repositories.

Every entity receives a stable identifier. The Knowledge Graph Search API returns it with the kg: prefix, for example kg:/m/0dl567, alongside types aligned with schema.org (Person, Place, Organization, Movie, Book, Event, among others). That identifier is what resolves “Springfield” in a text about Illinois and “Springfield” in a text about Massachusetts as two distinct nodes.

The mechanism has three steps. Google extracts fact candidates from text and markup, resolves which node each candidate belongs to (disambiguation), and consolidates the value when several independent sources agree. The sameAs property is the usual signal for declaring that your website, your listing in a directory and your social profile describe the same entity. Less visible properties such as iso6523 or naics work in the background towards the same end: separating you from organisations that share your name.

There is one limitation of the query tool worth knowing. Google states that the legacy API is being migrated to Cloud Enterprise Knowledge Graph, recommends the new product to anyone starting now, and warns that the old endpoint is not suitable as a production-critical service. It serves as an audit probe, not as a permanent operational source.

The concrete decision that hangs on this is where you invest when your brand does not show up recognised. If the problem sits in the documents, content and links fix it. If the problem sits in the entity, no volume of articles resolves it: what is missing is Google's ability to identify who you are, under which canonical name, and with which sources corroborating it.

That difference has become expensive since generative answer systems began citing brands. A model summarising a commercial query works on retrieved entities, not on your meta description. If your company is not resolved as a node, it appears described generically, confused with a namesake, or simply outside the candidate set.

Basic data management and crisis handling change as well. An old address, a founder who has left, or an obsolete legal name propagate across many surfaces at once when they live in the graph. Fixing the website without touching the sources that feed the node leaves the stale value circulating. Google notes that panels are generated automatically and that verified representatives can claim the listing and suggest corrections.

• Designate exactly one entity page for the brand (the home page or a stable “About us”) and mark it up with Organization in JSON-LD, with name, url, logo and legalName consistent with the commercial register.
• Declare sameAs towards the profiles that already describe you: Wikidata if one exists, LinkedIn, the Google Business Profile listing and the directories of your sector. Use the canonical URL of each profile, without campaign parameters.
• Fix a single canonical name and use it on the website, in invoicing and in the profiles. If “Nordia”, “Nordia Group” and “Nordia Consulting SL” coexist, pick one and relegate the rest to alternateName.
• Corroborate the hard facts (address, founding year, key people) in at least two external sources a crawler can read. A fact that only appears on your own website is an unconfirmed fact.
• Check whether a node already exists before planning: a query to the Knowledge Graph Search API with the exact brand name returns the identifier and the assigned type, or nothing.
• Claim the knowledge panel as a verified representative and use the suggest-a-change flow for factual errors, instead of rewriting the website over and over.

• Taking the knowledge panel as proof of existence. The panel is generated automatically once there is enough information on the open web, so its absence points to a lack of corroboration rather than to the node not existing.
• Inserting the Organization markup into every template of the site. The entity is declared on one page; repeating the block across thousands of URLs multiplies versions that contradict each other.
• Expecting a guaranteed effect from the markup. Google states expressly that it does not guarantee that features which consume structured data will show up in the results.
• Pointing sameAs at any page that mentions the brand. That property declares identity, not mention: an article writing about you is not a profile of yours.
• Building the audit on the legacy API. Google is migrating it to Cloud Enterprise Knowledge Graph and warns that it is not suitable as a production-critical service, so an empty response can be a limit of the endpoint and not the absence of the entity.