Knowledge Graph, Wikidata and Entity Reconciliation
Search engines and AI assistants describe your brand using entity databases such as Google's Knowledge Graph and Wikidata. Entity reconciliation means making every source agree on who you are, so those descriptions are accurate.
  • The Knowledge Graph stores entities with machine IDs. Check whether you have one with the Knowledge Graph Search API.
  • If you have a knowledge panel, claim it through Google's verification process and use suggested edits to fix errors.
  • Wikidata is open, widely reused and has a lower notability bar than Wikipedia, but items still need independent references.
  • Pick one entity home page, keep a canonical facts sheet and make your site, schema and third-party profiles match it.
  • Fixing facts at the source corrects many downstream surfaces at once, including AI answers.

When someone searches your brand, the panel on the right, the summary in an AI Overview and the answer ChatGPT gives about your company often trace back to the same few sources. Google's Knowledge Graph and the open Wikidata project sit near the centre of that web. Entity reconciliation is the work of making those sources, your own site and your third-party profiles all agree, so every system that describes you starts from the same facts.

What the Knowledge Graph is

The Knowledge Graph is Google's database of things rather than pages: people, companies, places, products, creative works and concepts, each with attributes and relationships. Every entity has a machine identifier. Older ones start /m/ and date from Google's Freebase acquisition, while newer ones start /g/.

Google assembles these entities from many places, including Wikipedia, Wikidata, licensed data, official websites, structured data and, for local businesses, Google Business Profile. A knowledge panel is simply one visible output of the graph. Unlike a rich result, you cannot switch one on with markup. It appears when Google is confident about who an entity is and thinks the panel helps the searcher.

Find your entity ID

The Knowledge Graph Search API lets you look up entities by name. It needs a free API key and returns results in JSON-LD, each with an @id such as kg:/m/0dl567 , a type, a short description and a result score.

 https://kgsearch.googleapis.com/v1/entities:search?query=example+ltd&key=API_KEY&limit=5

Search for your brand, its founders and its main products. No result, or a result describing someone else, tells you Google has not confidently matched your name to you. You will also see the ID in Google URLs as the kgmid parameter when you click through from a knowledge panel.

Claiming and correcting a knowledge panel

If a panel already exists, look for the "Claim this knowledge panel" link at the bottom of it. Google's get verified guide walks through confirming that you represent the entity, usually by signing in to an official account linked to it. Verified representatives can suggest edits, propose a featured image and add other people as managers or contributors.

Suggestions are reviewed, not applied automatically, and some elements are not editable at all. Back each suggestion with a public URL that confirms the fact. If your panel mixes you up with someone of the same name, Google's help pages have a specific route for that problem. Local businesses should manage their details through Google Business Profile instead.

Wikidata: the hub everyone reads

Wikidata is a free, openly licensed knowledge base run by the Wikimedia Foundation. Each item has a Q number, a label, a description, aliases and statements, and each statement can carry references. Because the data is open and machine-readable, it is reused by search engines, voice assistants and a great many AI datasets.

Its notability policy is far more generous than Wikipedia's. An item is acceptable if it links to a Wikimedia page, describes a clearly identifiable entity that serious and publicly available references can describe, or fills a structural need. In practice the second route is the one most organisations use, and the word that matters is references. Items supported only by the organisation's own website are regularly deleted.

Creating an item properly

  • Search first. Duplicates get merged and look careless.
  • Write a neutral label and a short description that distinguishes you from namesakes, such as "UK technical SEO consultancy".
  • Add core statements: instance of, official website, inception date, country, founders and official identifiers such as a company registration number.
  • Reference statements with independent sources: company registries, press coverage, industry bodies.
  • Check Wikimedia's current rules on paid contributions before editing for a client, and be open about the relationship.
  • Once the item exists, add its URL to the sameAs list in your Organization or Person markup.

Reconciliation: making the sources agree

  • Choose an entity home. One page, usually About, that states the canonical facts in plain text and carries your main structured data.
  • Keep a facts sheet. Legal name, trading name, founding date, founders, headquarters, company number, a one-line description and the current logo.
  • Match everything to it. Your site, schema, LinkedIn, Companies House, Crunchbase, directories, speaker bios and press boilerplate.
  • Link both ways. sameAs from your markup to the profiles, and links from the profiles back to your site.
  • Disambiguate namesakes with distinctive descriptions and hard identifiers.
  • Handle rebrands deliberately. Keep the old name as an alternate name, update profiles and add dated statements on Wikidata.

Common mistakes

  • Creating a Wikidata item with no independent references.
  • Writing your own Wikipedia article, which usually ends in deletion and a conflict of interest note.
  • Different founding dates or addresses across directories.
  • sameAs links that point at a namesake.
  • Expecting a knowledge panel to appear on request.

How to test

  • Query the Knowledge Graph Search API for your brand and people every quarter and log the IDs and descriptions.
  • Search your brand in Google and check the panel, if there is one.
  • Ask several AI assistants who you are and what you do, and note errors and the sources they cite.
  • Review your Wikidata item's statements and references.

The GEO angle

Language models learn a great deal about entities from reference sources such as Wikipedia and Wikidata, and live AI search leans on the same sources when it describes a brand. Errors there spread and persist. Correcting the upstream facts once, with good references, tends to fix far more than chasing each assistant's answer individually.

Further reading

Related guides