Google's Knowledge Graph: what the record holds, how to look up your own entity, and which sources move it

Google's Knowledge Graph is a database of things, each stored as a record with its own identifier, a type, a set of attributes and links to other records. It is the reason a search for a company returns a box of facts rather than only a list of pages, and the reason an answer about your business can be stated in a sentence Google wrote. The record is assembled from sources Google trusts and agrees with, which is why the useful questions are practical ones: does a record for my business exist, what identifier does it carry, what does it currently say, and which sources is it saying it from.

What a record in the graph actually holds

Part of the record	What it is	Example for a company
Identifier	A stable internal id for the thing itself, independent of its name	A value beginning /m/ or /g/
Type	The class the thing belongs to	Organization, LocalBusiness, Person, Place, Product
Attributes	The facts held about it	Founding year, founders, headquarters, industry, parent company
Relationships	Links to other records	Founded by, subsidiary of, headquartered in, employs
Description	A short written account, usually sourced rather than written by Google	A paragraph often taken from an encyclopaedia entry
Images and links	A picture and the official profiles associated with the thing	Logo, website, social and video profiles

The identifier matters more than it looks. A name is a label and can be shared by three companies; the identifier is the thing. When two businesses with similar names get understood as one, it is because one identifier is carrying both sets of attributes.

What Google has published about it, with dates

• July 2010. Google acquired Metaweb, the company behind Freebase, an open database of structured facts about people, places and things. That acquisition is where much of the early content came from.
• 16 May 2012. Google announced the Knowledge Graph publicly, describing it at the time as around 500 million objects and more than 3.5 billion facts and relationships between them. Both figures are from the announcement and are a 2012 snapshot rather than a current size.
• December 2014. Google said Freebase would close and that it would help move the data to Wikidata, the Wikimedia community's structured database. Freebase went read only the following year and the site closed in 2016. This is why Wikidata is now the open database most often named in entity work.
• Since then. Google's own documentation describes knowledge panels, the boxes of facts beside search results, as generated from the Knowledge Graph rather than submitted by the business.

How to look up your own entity

Two methods, and they answer slightly different questions.

• The search URL. Search your company name, click the result that opens a panel, and read the address bar. When Google has resolved the query to an entity, the URL carries a kgmid parameter holding the identifier. Copy it somewhere permanent, because it is the one handle for your business that does not change when your name or your website does.
• The Knowledge Graph Search API. Google publishes an API that takes a name and returns matching entities with their identifiers, types, a short description and, where one exists, a longer description with the source it came from. It needs a Google Cloud API key and returns JSON. Searching your own name there shows you which entities Google holds for that string, which is the fastest way to find out that two of them exist.

Two identifier shapes come back. Values beginning /m/ are the older ones inherited from Freebase. Values beginning /g/ are the newer ones. Either is a valid handle, and a business created as an entity in the last few years normally has the second kind.

The three positions a business can be in

Position	How you can tell	The work that follows
No record	No panel, and the API returns nothing for your exact name	Make the business identifiable: one name, one home page that stands for the company, Organization markup, and listings that agree
A record with no panel	The API returns an entity but no box appears in search	Corroboration. Independent sources that describe the company, not only mention it
A record and a panel	A box appears with a description and facts	Claim it as a verified representative, then fix wrong facts at the source first
Two records	The API returns two entities with your name, or the panel mixes two companies	Name discipline and distinct markup, held until the records separate

Where the facts in a record come from

Google treats your own site as the one source that speaks for you, and everything else as a check on it. In practice the record is built from agreement.

• Your own entity home. One URL that stands for the company, carrying the name, the description, the location and the founding facts, with Organization markup and a sameAs list pointing at every official profile you hold. The mechanics of that markup are set out in entity based SEO.
• Open structured databases. Wikidata most of all, plus the industry registers and regulator databases that hold entries for companies in your sector.
• Sources that describe rather than mention. Trade press, industry bodies, award and conference pages, reputable company databases. A paragraph saying what the company is does more work than a passing name check.
• Listings and profiles. The same name, address and phone across directories, which is what stops two half built records forming.

Changing what a record says is a matter of changing what the sources say and then waiting for the refresh. The claim and correction route for the visible box, including what a verified representative can submit, is covered in Google knowledge panels for a business.

A monthly reading worth keeping

Record five things on the same date each month in a spreadsheet: the date, whether a panel appears, the identifier, the first sentence of the description, and the image being used. Four readings in a row make a trend, and a changed sentence tells you a source changed upstream. Without the log you are comparing today against your memory of a box you last looked at properly in March.

Questions about Google's Knowledge Graph

Is the Knowledge Graph the same thing as a knowledge panel?

No. The graph is the database. The panel is one place the database gets displayed. A record can exist in the graph with no panel shown for it, which is the most common position for a small company.

Can I add my business to the Knowledge Graph directly?

There is no submission form. Google builds records from sources it reads. What you can do is make the business easy to identify and get independent sources describing it the same way, which is the work that produces a record.

What is a kgmid?

It is the Knowledge Graph identifier for an entity, visible as a parameter in the Google search URL when a query resolves to one. It is worth storing alongside your other brand assets, since it names the thing rather than the words used for it.

Do I need a Wikipedia article?

Companies hold records and panels without one. Wikipedia and Wikidata are source based projects with their own inclusion rules, so they follow coverage rather than create it. Industry registers and trade coverage are the realistic anchors for most businesses.

Does a Knowledge Graph record help with anything other than the panel?

It is the reason Google can answer a question about your company in a sentence, pick the right company when the name is ambiguous, and connect you to the people and products associated with you. All three of those run on the record rather than on any single page.

How long does a change take?

It follows the sources. A fact that already agrees across several places moves quickly once your own page matches; a fact only your footer has ever stated waits until something independent repeats it.

The same record, read by something that answers in sentences

A buyer who used to type a category into a search box now often asks an assistant who to use. The answer comes back as two or three named companies with a line of description each, assembled from text about those companies published across the web. That is the same material a graph record is built from, which means the entity work has a second audience and a second output: not a box of facts, but a sentence recommending somebody. It is worth reading the sentence currently written about you. AI Knows Us asks those buying questions across the main assistants and records which companies get named and what the answers were built from.