Google knowledge graph

Google has never been transparent about its knowledge graph (KG) infrastructure and that silence is itself revealing.

A 2018 paper by Niel Chah (arxiv.org/abs/1805.03885) tried to reverse-engineer Google’s KG by reconstructing the Freebase dump: 1.9 billion RDF triples. Google acquired Freebase in 2010, quietly shut it down in 2016, and absorbed its ontology into a proprietary blackbox that nobody outside Google can inspect yet it answers millions of queries every day.

Fast forward to 2025: Google’s KG now reportedly holds ~54 billion entities and 1.6 trillion facts. Last June they pruned over 3 billion entities to make it leaner as a backend for Gemini and AI Overviews. The ontology remains opaque. It’s just a faster, more AI-optimized blackbox.

Google IO made the shift from SEO to GEO (generated engine optimization) explicit. That transition is structurally dependent on schema and ontologies yet Google didn’t say so. Neither does Microsoft, despite CosmosDB being a capable multi-modal store. The knowledge layer is load-bearing infrastructure that neither company wants to discuss publicly.