Knowledge Graph

Google describes its search results as sometimes showing information from the Knowledge Graph, a database of billions of facts about people, places and things, which lets it answer factual questions such as how tall the Eiffel Tower is^[1]. It introduced the graph in 2012 as a model of real-world entities and their relationships, with the line “things, not strings”, and said then that it held more than 500 million objects and more than 3.5 billion facts about and relationships between them^[2].

The facts come from several places. Google says they come from a variety of sources that compile factual information, from data it licenses, such as sports scores, stock prices and weather forecasts, and from content owners directly, including those who suggest changes to knowledge panels they have claimed^[1]. Knowledge panels are information boxes for entities in the graph, generated automatically, and kept up to date as the information changes on the web^[3].

Keep three things apart. The graph is the store. An entity is the thing a record in the store stands for: a company, a person, a place. A knowledge panel is one way the store is shown to a searcher. Google’s Knowledge Graph Search API lists the types of record it holds, among them Organization, LocalBusiness, Person, Place and WebSite^[4].

For a brand, the graph is a place where the brand’s facts appear in results that the brand does not control. A site is one input among several. Google says organization markup on a home page can help it understand an organization’s administrative details and tell it apart from others, and that some properties can influence which logo shows in Search results and in a knowledge panel^[5]. A claim a site makes about itself can also disagree with what is found at its address. In QueryRev’s study, 125 robots.txt files name an llms.txt file, and 20 of those sites had none that the study could find^[6]. A system that gathers facts has to weigh such claims against the rest of the web.

Search for the brand by name, and by name with its category, and record whether a panel appears, which facts it shows and where each could have come from. A business that serves customers at a location has a Business Profile, which looks like a knowledge panel but is specific to such businesses^[3] and is created and claimed through Business Profile^[1]. Google says its automated systems decide when, where and how a query shows a panel, and that this is not influenced by hand^[1]. A panel that disappears is therefore a finding, not a fault to repair.

Look the brand up in the graph. The Knowledge Graph Search API finds entities and returns them with standard schema.org types, in JSON-LD, one entity at a time rather than as a graph of connected entities^[4]. A lookup shows whether the brand is in the graph and under which type. Google warns that the API is not suitable for a production-critical service, so use it for checks and do not build a product on it^[4].

Compare facts. For each fact a panel or a lookup shows, find the page where the brand states it, and the Organization markup that repeats it. A fact that differs, or that appears only in the panel, is the finding.

Sample data: an invented site, to show the method.

A sample bed retailer compares its Organization markup with the panel Google shows for its name. The name and the address match. The panel shows the old logo, and the markup carries the new one. The panel also gives a founding year that appears nowhere on the retailer’s own pages.

The panel is built from many sources, so the markup alone does not decide it. The retailer corrects the logo in its markup, states the founding year in visible text on its about page and in the markup, and, as the official representative of the entity, claims the panel and suggests the change through Google’s verification. It checks the panel again after Google has had time to read the pages.