Google's Knowledge Graph is Google's database of entities (people, places, organizations and things) and the facts and relationships that connect them. Google launched it on 16 May 2012, and it supplies the knowledge panels that appear beside or among search results.
How the Knowledge Graph works
At launch Google said the graph held more than 500 million objects and more than 3.5 billion facts about and relationships between them, drawn from sources such as Freebase, Wikipedia and the CIA World Factbook. By May 2020 Google put it at over 500 billion facts about five billion entities.
Those facts come from material across the web and from open-source and licensed databases. Google's 2020 explainer adds licensed data for music, sports and TV panels, vetted content from medical providers for health panels, and markup that site owners add, such as event details. Automated systems gather the facts and other systems try to keep errors out; anyone can report a problem through the Feedback link on a panel.
A knowledge panel shows a selection of those facts, typically a title and summary, a longer description, images, key facts and links to official sites and social profiles. On mobile, Google often splits the information into one or more panels mixed in with the results. The subject of a panel, such as a person or company, can claim it through Google's verification process, which lets them give Google feedback directly and suggest changes such as a preferred photo.
Why the Knowledge Graph matters for AI answers
When Google introduced AI Mode on 5 March 2025, it said the feature can tap fresh sources such as the Knowledge Graph as well as web content, and uses query fan-out to run related searches across those sources at once. That makes any facts the graph holds about your company one of the inputs Google AI Mode can draw on, alongside your own pages and what others publish about you.
In classic search, the graph's facts show up in knowledge panels, and Google's help page says a featured snippet can also appear next to Knowledge Graph information. If a fact about your brand is wrong in a panel or in an AI answer, start with how to correct wrong information about your brand in AI answers.
An example: one entity record
Google's Knowledge Graph Search API documentation shows what a single entry looks like. A query for "taylor swift" returns a record with the ID kg:/m/0dl567, the schema.org types Thing and Person, the short description "Singer-songwriter", a longer description taken from Wikipedia, an official URL and a resultScore. The response is JSON-LD using schema.org types, the same vocabulary sites use in their own markup.
Common misconceptions
- "The Knowledge Graph is Wikipedia." Wikipedia is a commonly cited source, but Google says it draws from hundreds of sources across the web.
- "The public API shows the whole graph." The Search API returns individual matching entities, not the connections between them; for a connected graph, Google points developers to Wikidata data dumps. Google also warns that the API is not suitable as a production-critical service and is migrating it to Cloud Enterprise Knowledge Graph.
- "The graph is fixed once built." Google's 2012 launch post said it is tuned based on what people search for and what Google finds on the web, and the 2020 post describes removing inaccurate facts that violate Google's policies once they come to its attention.