Field Notes · GEO
Entity SEO Explained
How search engines and AI decide who your brand is. Before Google ranks you or an AI names you, it has to resolve your brand to one unambiguous, corroborated entity, and the brands that lose are the ones machines cannot confidently identify at all.
There is a question search engines and AI systems now answer before they ever decide where to rank you or whether to name you: who, exactly, is this brand? Not which page matches the query, but which real-world thing the query and the page both point to, and whether independent sources agree on what that thing is. This is the shift from keyword matching to entity resolution, and it is the quiet foundation under AI Overviews, Google's AI Mode, and the way ChatGPT and Perplexity choose who to cite.
The stakes became concrete in June 2025. Google did not add to its Knowledge Graph that month, it subtracted: in a single week it deleted more than three billion entities, a contraction of roughly 6.26 percent and the largest cull since around 2015. Search Engine Land framed it as a clarity cleanup, an anti-hoarding move to give AI surfaces high-confidence entities to draw on. The lesson for brands is blunt. Being present in the graph is no longer the bar. Being a clear, corroborated, confidently identified entity is.
What is an entity, and how is it different from a keyword?
An entity is a distinct, real-world thing, a person, organisation, product, place, or topic, that a machine can identify and connect to other things, whereas a keyword is just a string of characters. The classic example is the word "apple". As a keyword it is four letters. As an entity it could be the fruit or the technology company, and Google's natural language processing decides which by reading context and relationships, not the word in isolation. This process, Named Entity Recognition, is how a search engine labels the people, organisations, products and places in your content and works out which known thing each one refers to.
Entities do not replace keywords. The two work as layers. Keyword and topic work tells a system what a page is about; entity work tells it who is saying it and whether that source can be trusted to say it. Treating entity SEO as a replacement for keyword research is one of the most common and most expensive misreadings of the field.
What is the Google Knowledge Graph and the knowledge panel?
The Knowledge Graph is Google's vast internal database of entities and the facts that connect them, and the knowledge panel is the public-facing card it powers in search results. By early to mid-2024 the graph was reported to hold roughly 54 billion entities and around 1.6 trillion facts, up from about 5 billion entities and 500 billion facts in May 2020. Treat those figures as approximate; they come from secondary reporting, not a clean Google number, and the June 2025 cull shows the total moves down as well as up.
What matters more than the size is the direction of travel. Through 2023 and 2024 the graph grew aggressively. Then the posture flipped from maximising volume to maximising confidence, because low-confidence entities are a liability when they feed AI answers. A knowledge panel is Google signalling that it has resolved you to a confident entity. You can claim and verify one: search your entity, click "Claim this knowledge panel", and verify through Google Search Console (the most reliable route, since it proves you own the official site), a connected YouTube channel, a Google Business Profile, or linked official social accounts. Identity claims can require a photo of you holding official photo ID with name and face visible, and verification usually takes a few days to a week.
What are machine identifiers like the KGMID and MID?
A Knowledge Graph Machine ID (KGMID) is the unique internal code Google assigns to a resolved entity, the thing that lets it track "you" as a single object rather than a fuzzy collection of mentions. The format looks like kg:/m/0d7_4z or kg:/g/11b7lsr1sn. The prefix tells you the entity's lineage: a /m/ prefix means the entity originated in Freebase, the old database Google acquired and shut down in 2016, while a /g/ prefix means it was added after that. The practical point is simple. When Google has given your brand a stable machine ID, it has decided you are one identifiable thing. That is the asset you are trying to earn and protect, because everything downstream, ranking confidence, knowledge panels, AI citations, attaches to that resolved identity.
How does corroboration decide your identity?
Search engines and AI confirm who you are by cross-checking your own statements about yourself against independent, authoritative sources, and only an entity that survives that cross-check is treated as high-confidence. Google's documented Knowledge Graph inputs include Wikipedia, Wikidata, Crunchbase, IMDb and MusicBrainz, alongside its own sources and authoritative websites, with Wikidata described as a primary, direct input. This is why the practitioner model of an "Entity Home", popularised by Jason Barnard and Kalicube, is useful: a single owned, canonical URL that acts as your point of reconciliation, the place Google and the models return to and check the wider web against.
In schema terms, you express this with Organization markup and the sameAs property, which links your canonical page to your profiles on Wikidata, LinkedIn, Crunchbase, your Google Business Profile and your official social accounts. Done well, this creates what Barnard calls a self-corroboration loop: every source points to every other, and the machine sees consensus rather than conflict. Note that Entity Home and Barnard's expanded N-E-E-A-T-T framework, which adds Notability and Transparency to E-E-A-T, are respected industry models rather than confirmed Google terminology; Google's own language is closer to "reconciliation".
Does adding sameAs and schema directly boost AI citations?
No, the controlled evidence says schema markup on its own produces no meaningful causal lift in AI citations, and believing otherwise is the single most common mistake in this field. Ahrefs tracked 1,885 pages that added schema, matched each against equivalent control pages without JSON-LD, and measured citation change over the 30 days before and after. The citations barely moved. A separate study attributed to Search Atlas in December 2024 found no correlation between schema coverage and AI citation rates; the variable that predicted citations was content authority and relevance, not the markup payload. Direct retrieval tests further suggest major AI systems read visible HTML and lean on that, so hidden structured data alone is a weak bet.
This does not make schema pointless, and the debate is genuinely contested, with vendors like Schema App arguing sameAs still materially aids disambiguation. The honest framing keeps two claims separate. Schema helps machines parse and disambiguate your entity, which supports resolution. Schema is not a direct ranking or citation lever, which is a different and unsupported claim. Markup that helps a machine confirm who you are is worth shipping. Markup bolted on in the hope of a citation bump is not.
How does NAP and name consistency lock your entity down?
Consistent Name, Address and Phone details across every place your brand appears are the cheapest and most reliable corroboration signal you have, because contradictions force a machine to lower its confidence in your identity. If your business name is spelled three ways, your address differs between your site and your Google Business Profile, and your phone number varies across directories, you are handing the system reasons to doubt that all those mentions are the same entity. The fix is unglamorous and absolute: pick one canonical form of every fact and make the entire web agree.
This is sharpest for the same-name disambiguation problem. If another organisation shares your name, or your founder shares a name with a more notable person, the machine has to split a tangled cluster of mentions into separate entities. You win that split by being relentlessly specific and consistent: a stable location, a distinct description, unique linked profiles, and an Entity Home that ties it all to one identifier. Ambiguity is not neutral. It is the condition under which a low-confidence entity gets pruned.
How do LLMs resolve entities, and why does it matter now?
Large language models resolve entities by leaning on knowledge graphs to disambiguate candidates, then enriching their answer with descriptions of the entity they have settled on. 2025 academic work, including arXiv paper 2505.02737 on knowledge graphs for entity disambiguation, shows LLMs using a graph's class and subclass hierarchies to prune the field of possible matches and entity descriptions to enrich the prompt, with GraphRAG and KG-RAG approaches adding multi-hop reasoning across the graph. In plain terms, the same corroborated, well-described, clearly-classified entity that Google trusts is the one an LLM can confidently pick out and name. The disciplines have converged. A clean entity is now a single asset that pays off across classic search and AI answers at once, which is why entity SEO has been reframed as core GEO work rather than a niche.
There is one more reason the timing matters. A 2024 update tied entity recognition more tightly to credibility signals, so identity and trust are increasingly read as one thing. The job is not to demonstrate expertise only to human raters. It is to make that expertise legible to machines, as structured, corroborated, consistent facts they can verify without guessing.
A practical checklist to lock your entity down
Locking an entity down is a corroboration exercise, not a markup exercise: you are making one identity true everywhere a machine looks. Work through these in order.
- Establish an Entity Home: one canonical, owned URL (usually an About or brand page) that states clearly who you are, what you do, and where, in plain visible text.
- Standardise NAP: choose one exact form of your name, address and phone, and reconcile it across your site, Google Business Profile and every directory.
- Add Organization schema with a complete sameAs array pointing to your authoritative profiles, so machines can parse and disambiguate, while remembering markup confirms identity rather than buys citations.
- Build the corroboration set: Wikidata, LinkedIn, Crunchbase and a Google Business Profile, since Wikidata in particular is a direct Knowledge Graph input.
- Make every external source point back to the Entity Home and to each other, closing the self-corroboration loop.
- Claim and verify your knowledge panel through Search Console once Google resolves you, and correct any wrong facts it shows.
- Resolve same-name ambiguity with specific, distinguishing details so your cluster of mentions cannot be confused with another entity's.
- Lead with authority and relevance in the visible content itself, because that, not the JSON-LD, is what the studies show actually drives citations.
Entity SEO is not a markup trick, it is an identity discipline. The brands that get ranked with confidence and named in AI answers are the ones a machine can resolve to a single, corroborated, unambiguous thing, with their own statements and the independent web telling the same story. Get that right and the rest of your SEO and GEO work compounds on a foundation that holds. Building that foundation, the Entity Home, the corroboration set, the consistent facts that survive a clarity cull, is exactly the work we do at DarkHorse Creative.
Questions people ask
What is the difference between an entity and a keyword?
A keyword is a string of characters a system matches against a query, while an entity is a distinct real-world thing, a person, company, product or place, that a machine identifies and connects to other things. The word "apple" is one keyword but two possible entities, the fruit and the company, and Google decides which one you mean from context and relationships. Entity SEO layers on top of keyword work, it does not replace it.
Does adding sameAs schema improve my AI citations?
Not directly. Controlled studies, including an Ahrefs test across 1,885 pages and a separate Search Atlas analysis, found no meaningful causal lift in AI citations from schema markup alone; content authority and relevance were the variables that predicted citations. Schema still helps machines parse and disambiguate your entity, which supports identity resolution, so it is worth shipping, just not as a citation lever.
What is a knowledge panel and how do I claim one?
A knowledge panel is the information card Google shows for an entity it has confidently resolved in its Knowledge Graph. To claim yours, search the entity, click "Claim this knowledge panel", and verify through Google Search Console (the most reliable route since it proves you own the official site), a connected YouTube channel, a Google Business Profile, or linked official social accounts. Verification usually takes a few days to a week, and identity claims can require a photo of you holding official photo ID.
Why did Google delete billions of entities from its Knowledge Graph?
In June 2025 Google removed more than three billion entities in a single week, a roughly 6.26 percent contraction and its largest cull since around 2015. Search Engine Land reported it as a clarity cleanup, an anti-hoarding move to give AI surfaces like AI Overviews and AI Mode high-confidence entities to rely on. The takeaway is that simply existing in the graph is no longer enough; you have to be a clearly corroborated, high-confidence entity to survive and benefit.