- A knowledge graph stores entities (people, places, organizations, things) and labeled relationships between them, not indexed pages ranked by keyword relevance. "What is this thing" is a different question from "which pages mention these words," and it's answered by a different kind of system.
- Google's Knowledge Graph launched publicly in May 2012 with more than 500 million objects and 3.5 billion facts. Its last disclosed figure, per Google's own documentation, put that at over 500 billion facts about 5 billion entities, a roughly 140x growth in facts over eight years.
- Wikidata, a separate open database, is widely described as one of the Knowledge Graph's sources, alongside Wikipedia and structured data from individual sites. Google's own documentation doesn't disclose the exact weighting between sources.
- Whether AI chatbots query Google's Knowledge Graph directly is unconfirmed by any primary source found in researching this piece. What is established: Google's own Knowledge Graph Search API uses schema.org vocabulary and JSON-LD, the same structured-data standard this site's entity-resolution work already points toward.
- The practical point isn't the graph's size, it's the mechanism: an AI engine has to resolve "which entity is this" before it can attach a fact, review, or citation to it, which is the whole reason entity consistency work exists.
Most explanations of "the knowledge graph" skip straight to SEO advice without explaining what the thing actually is. Here's the mechanical version: what a knowledge graph stores, how Google's got built, what role Wikidata plays, and what's actually confirmed versus assumed about its connection to AI search.
What does a knowledge graph actually store?
A knowledge graph represents information as entities (nodes) connected by labeled relationships (edges), rather than as documents ranked by keyword match. "Google" the company is a node. "Alphabet Inc." is a node. The relationship "is a subsidiary of" is the labeled edge connecting them. A search index, by contrast, doesn't know that Google and Alphabet are related entities at all; it just knows which pages contain which words.
That structural difference is why a knowledge graph can answer a different category of question than a search index can. A search index can tell you which pages mention "founded in 1998." A knowledge graph can tell you that the entity "Google" has the property "founding date: 1998," directly, as a fact attached to a specific, disambiguated node, without needing to retrieve or rank any page at all.
How was Google's Knowledge Graph actually built, and how big is it?
Google introduced the Knowledge Graph publicly in May 2012, reporting more than 500 million objects and more than 3.5 billion facts about them and their relationships, drawing on sources including Freebase, Wikipedia, and the CIA World Factbook.
The growth since then is the part most SEO content gets wrong by quoting mismatched figures. By October 2016, Google disclosed the graph held over 70 billion facts. Its most recent disclosed figure, stated on Google's own "about knowledge graph and knowledge panels" page, puts the Knowledge Graph at over 500 billion facts about 5 billion entities. That's roughly 140 times the fact count reported at the 2012 launch, over eight years. No more recent official figure turned up in researching this piece, so treat 500 billion facts / 5 billion entities as the latest confirmed disclosure, not a current, continuously-updated number.
| Disclosed point | Objects / entities | Facts |
|---|---|---|
| May 2012 launch | 500+ million objects | 3.5+ billion |
| October 2016 | Not separately disclosed | 70+ billion |
| Most recent disclosure | 5 billion entities | 500+ billion |
Where does Wikidata fit into this?
Wikidata is a separate project, not owned or operated by Google: a free, collaboratively-edited structured database where every item gets a stable identifier and every factual statement is expected to carry a source reference. It maintains its own notability standard, distinct from Wikipedia's, meaning an entity can have a Wikidata item without having a Wikipedia article at all.
Wikidata is widely described, across secondary sources, as one input into Google's Knowledge Graph, alongside Wikipedia and structured data markup publishers add to their own sites. Google's own public documentation doesn't spell out the precise mechanics or weighting of how each source contributes, so treat "Wikidata feeds the Knowledge Graph" as the consistent secondary-source account rather than a claim Google has directly confirmed in writing. What is directly useful regardless of that exact mechanism: a Wikidata item is a real, independently queryable record of your entity's facts, and getting it accurate is a concrete, actionable task in a way "influence Google's internal weighting" isn't. For the practical setup question (is it worth doing, and how), see Wikidata for brands: is it worth setting up?
Can you query the Knowledge Graph directly?
Yes. Google publishes a Knowledge Graph Search API that lets you look up entities directly, independent of a regular search query. It's built on schema.org vocabulary and is compliant with the JSON-LD specification, the same structured-data standard behind the markup discussed in structured data for AI search. This is useful in a very concrete way: it lets you check whether your own brand resolves as a single entity with a stable identifier, rather than guessing at how Google's systems see you.
Does this connect directly to how AI chatbots answer questions?
This is the part that gets overstated most often in marketing content, and it's worth being precise about what's actually confirmed. Google's own AI Overviews and AI Mode are Google products, and it's plausible they draw on Google's broader internal infrastructure, which likely includes Knowledge Graph data in some form. But Google hasn't published the specific architecture connecting AI Overviews' answer generation to Knowledge Graph lookups, so claims about exactly how or how much it's used there are inference, not disclosed fact.
What's unambiguous: ChatGPT, Claude, and Perplexity have no access to Google's Knowledge Graph at all. They're built and operated by different companies with their own training data and retrieval systems. Any "entity resolution" those models do runs on an entirely separate mechanism, not a shared graph. If a vendor pitch implies that optimizing for "Google's Knowledge Graph" is a universal AI-visibility tactic across all chatbots, that's conflating one company's internal system with a problem every model actually has to solve independently.
Why does any of this matter if you're not optimizing for Google specifically?
Because the underlying problem the Knowledge Graph solves for Google (confidently identifying which entity a piece of text is actually about, before attaching any fact to it) is a problem every AI system has to solve in some form, on its own infrastructure, before it can cite you confidently. Google's Knowledge Graph is the most-documented, most-queryable example of that mechanism in action, which makes it a useful model for understanding entity resolution generally, even on platforms that have nothing to do with Google. The practical takeaway isn't "get into Google's Knowledge Graph specifically." It's "make sure every system resolving your brand as an entity, Google's included, has consistent, unambiguous facts to resolve it with," which is exactly the problem entity consistency addresses.
Frequently asked questions
What is a knowledge graph?
A structured database of entities (people, places, organizations, products, concepts) connected by labeled relationships, rather than a collection of indexed web pages ranked by relevance. Where a search index answers "which pages match these keywords," a knowledge graph answers "what is this thing, and what else is true about it."
How big is Google's Knowledge Graph?
At its May 2012 public launch, Google reported more than 500 million objects and more than 3.5 billion facts. By its last disclosed figure, cited on Google's own "about knowledge graph and knowledge panels" page, that had grown to over 500 billion facts about 5 billion entities. Google has not published a more recent official figure found in researching this piece.
Where does Wikidata fit in?
Wikidata is a separate, open, collaboratively-edited structured database (not owned by Google) that assigns every item a stable identifier and sources its facts to citable references. It's widely described as one of the sources Google's Knowledge Graph draws on, alongside Wikipedia, licensed data, and structured data markup on individual sites, though Google's own documentation does not spell out the exact weighting of each source.
Do AI chatbots use Google's Knowledge Graph directly?
That's not confirmed. Google's own AI Overviews and AI Mode plausibly draw on Google's internal infrastructure, which likely includes the Knowledge Graph in some form, but Google hasn't published the specific architecture. Chatbots from other companies (ChatGPT, Claude, Perplexity) have no access to Google's Knowledge Graph at all; any entity resolution they do runs on their own training data and retrieval systems, which is a different, non-interchangeable process.
Sources referenced
- Google: about knowledge graph and knowledge panels
- Google Official Blog, "Introducing the Knowledge Graph: things, not strings" (May 2012)
- Google Knowledge Graph Search API documentation
- Wikipedia: Knowledge Graph (Google)
For what resolving an entity actually requires and why it's a precondition for citation, see what is entity consistency, and why AI engines need it. For the practical Wikidata question, see Wikidata for brands: is it worth setting up?.
