Knowledge graphs organize information using entities, relationships, and attributes. They do more than store text: they record who or what it refers to, that entity’s characteristics, and its relationships with other things. Regular search results primarily list links to web pages; knowledge graphs can help search systems identify what a query refers to and display relevant facts directly. However, a knowledge graph is an underlying data structure, while a knowledge panel on a search results page is just one way to present information. They are not the same thing.
Breaking information down into entities, relationships, and attributes
According to Semrush, the basic components of a knowledge graph include nodes, edges, and attributes. A schema defines the language and data structure used in the graph. The article’s Apple example illustrates these components:
- Nodes represent entities: Apple, Steve Jobs, and iPhone are distinct entities, rather than just keywords within the same article.
- Edges represent relationships: Steve Jobs can be connected to Apple through “founded,” while iPhone can be connected to Apple through “manufactured by.”
- Attributes describe details: Information such as the company’s official name, founding date, and headquarters can be recorded under the Apple entity.
Entities also have unique identifiers to help distinguish things with the same name. This means Apple the company and apple the fruit need not be confused simply because they share a name. The focus is on identifying what a name refers to, rather than merely matching strings of text.
Identifying what a query refers to, then finding relevant information
Welly outlines the process as follows: first, extract entities and relationships from data and convert them into structured information. Then use natural language processing and machine learning to link entities with attributes. When a user submits a query, the system uses the data network and relationship model to find matching information. This is only an overview of the process; not every knowledge graph is implemented in the same way.
A query does not necessarily need to include an entity’s full name. Semrush gives the example of asking, “When was the company that makes the iPhone founded?” Google can use its knowledge graph and other methods to understand that the user wants Apple’s founding date. The system must both identify the company and determine which attribute the user is asking about. That understanding cannot be attributed entirely to the graph.
Web page listings link to further reading; panels bring facts together
Regular web search results primarily provide links that users can follow to read individual pages. Search systems can also use graph information to bring facts, such as a person’s date of birth or company details, together on the search results page. This reduces the need to move between pages to find a single piece of information. The difference is in how information is organized and presented; these are not mutually incompatible approaches to search.
According to Semrush’s article, knowledge panels may combine graph information with content from other sources—for example, displaying images or videos alongside a person’s date of birth. Whether a panel appears depends on whether the algorithm considers it useful. So not every direct answer on a search results page should be called a knowledge graph, nor should every item in a panel be assumed to come from the same database.
Knowledge graph Q&A: perspectives from across the web
Where does the information in Google's Knowledge Graph come from?
Semrush's overview states that Google obtains information from public sources, licensed data, and content owners, then stores it in graph format. The data does not necessarily all come from a single website; public sources may include Wikipedia, for example. When someone searches for information about a particular entity, some of that content may appear in the search results, but whether it appears still depends on the query and the system's assessment.
Is a knowledge graph built only when a user asks a question?
The author of a Search Engine Journal article argues that Google's Knowledge Graph is built and updated through established processes, rather than assembled in real time after a user enters a question. The reasoning is that the data involves identifiers, relationship rules, and alignment with existing entities. Matching existing information at query time is a different task from building an entire graph on the spot. This is the author's analysis of Google's graph and chat products, and should not be generalized into a universal rule for all systems.
Does exploring related topics in a knowledge graph lead to a deeper understanding of a subject?
Not necessarily. A Semrush article argues that knowledge panels can provide background information and link to items such as works associated with an entity. Following these links mostly means learning about other related entities, which does not necessarily deepen your understanding of the same subject. This kind of exploration is useful for getting an initial sense of context and related entities. For more in-depth content, you can still visit relevant web pages to read further.
References
- Google Knowledge Graph: What it is & why it matterswww.semrush.com · 2021-03-31T09:35:00Z — Discusses: Where does the information in Google's Knowledge Graph come from?
- 知識圖譜是什麼?知識圖譜運作原理、4大優化策略一次看welly.tw · 2025-01-22T07:54:44.039Z — Source used in the article above.
- Schema, LLMs & The Low Bar For ‘Evidence’ In GEOwww.searchenginejournal.com · 2026-06-16T11:00:57+00:00 — Discusses: Is a knowledge graph built only when a user asks a question?
- Why Google’s 10 Blue Links Will Have to Go (Eventually)www.semrush.com · 2022-01-24T09:05:00Z — Discusses: Does exploring related topics in a knowledge graph lead to a deeper understanding of a subject?
