Knowledge Graph

What Google said it was

The announcement is from 16 May 2012, written by Amit Singhal, and the sentence everyone half-remembers is worth having in full:

"It's why we've been working on an intelligent model—in geek-speak, a “graph”—that understands real-world entities and their relationships to one another: things, not strings."

The example in the post is [taj mahal], which can mean a monument, a Grammy-winning musician, or a casino. Under keyword matching those are one query. Under an entity model they are three things, and the search engine can ask which one you meant.

Google also named its starting material:

"Google's Knowledge Graph isn't just rooted in public sources such as Freebase, Wikipedia and the CIA World Factbook. It's also augmented at a much larger scale"

The numbers you have read are from 2012

This is the part worth knowing, because the figures circulate in decks and articles as though they described the present:

"It currently contains more than 500 million objects, as well as more than 3.5 billion facts about and relationships between these different objects."

The word is "currently", and the current in question was May 2012. Google has published no updated figure in that post, and the post has not been revised. Anyone quoting 500 million objects today is quoting a fourteen-year-old snapshot of a system that has been rebuilt around it.

The same date stamp applies to the sources named above. Google now describes Freebase, listed there as a root, in the past tense on its own developer page: "The Freebase API has been shut down, but the last available data dump is accessible", with the dumps kept for citation in studies rather than for use. Treat the 2012 paragraph as a historical statement rather than a description of what feeds the graph now.

The API, and what it does not promise

Google publishes a Knowledge Graph Search API, and its documentation is unusually frank about its limits.

"The Knowledge Graph Search API lets you find entities in the Google Knowledge Graph. The API uses standard schema.org types and is compliant with the JSON-LD specification."

Three warnings sit on that page, and each one matters if you were planning to build on it:

The read-only line is the one that settles a recurring question. There is no submission form, no feed, and no endpoint through which a brand adds itself.

What it is not

It is not the knowledge panel. The panel is one surface where graph data can appear. The graph is the underlying store, and entities exist in it without ever producing a panel.

Structured data is not documented as a way in. Google's introduction to structured data runs to about nineteen thousand characters and mentions the Knowledge Graph zero times, and the knowledge panel zero times. It talks about rich results (13 mentions), schema.org (6) and JSON-LD (7). We counted rather than assumed, with "structured data" (64 occurrences) as the control confirming the page was read.

Read that for exactly what it is. Structured data may well help Google understand an entity, and Google does not present it as a route into the Knowledge Graph on the page where it explains what structured data is for.

What to do with it

Never quote the 2012 figures as current. If a number is needed, quote it with its date and say it is from the announcement.

Do not promise a client entry into the graph. The API is read-only, and no documented submission path exists.

Separate the graph from the panel when reporting. A missing panel is not proof of a missing entity, and a present panel is not proof that everything in it came from your site.

Build new integrations on the successor product. The documentation says so on the old one's front page.