A plain database stores facts in rows, but much of what we know is really about relationships: who made what, which thing is part of another, how events connect. A knowledge graph represents information as a web of entities joined by named relationships, so the connections are data in their own right.
This lets a system answer questions that span several links, such as tracing an indirect connection, and gives other software a structured map of a domain. Paired with language models, graphs supply grounded facts and clear relationships that free-form text alone struggles to keep straight.