Entity SEO: optimising entities for AI search | Ighenatt

Search engines do not read keywords. They model the world. When someone searches for “SEO agency Barcelona”, Google is not looking for pages that contain those three words: it evaluates which entities exist in that category, what relationships they have with the geographic location, and how confident it is about each one. This shift — which Amit Singhal described in 2012 with the phrase “from strings to things” — has direct consequences for how generative AI models decide which brands to mention and which to ignore.

An Ahrefs study of 75,000 brands, published in November 2025, quantified the scale of the problem: brand mentions on the web correlate with visibility in Google AI Overviews at a coefficient of 0.664, while traditional backlinks correlate at only 0.218. A 3x gap that inverts the logic of classic link building. AI models learn from textual co-occurrences, not from link graphs. Before an LLM can cite you, Google needs to recognise you as an entity. That is where Entity SEO begins.

What entities are in SEO — and why they outperform keywords

An entity, in the context of Google’s Knowledge Graph, is anything that can be unambiguously identified: a company, a person, a place, a concept, a product. What distinguishes an entity from a keyword is disambiguation. “Barça” can refer to the football club, to the city of Barcelona, or to a colloquial nickname. As an entity, “FC Barcelona” has a unique identifier in the Knowledge Graph and a set of verified attributes — founding year, stadium, president, trophy record — that allow Google to speak about it with confidence in any context.

Google’s Knowledge Graph stores more than 500 billion facts about approximately 5 billion entities. When your brand does not exist in that knowledge base, generative AI models — ChatGPT, Perplexity, Google AI Overviews — cannot cite it precisely, regardless of how many pages you have indexed. For an LLM to mention you in response to “best SEO agencies in Barcelona”, it needs to know you are an entity of type “organisation”, with the category “SEO agency”, located in “Barcelona”, with verifiable credentials. That cannot be achieved with keywords: it is built with entity signals.

The practical distinction matters for understanding where to invest. Keyword SEO competes on the results page. Entity SEO builds the brand’s reputation in the knowledge base that feeds both that results page and the AI systems progressively replacing it.

Knowledge Graph: how it works and why your brand might not be in it

Google does not build the Knowledge Graph by reading web pages in isolation. It builds it by triangulating between authoritative sources: Wikipedia, Wikidata, DBpedia, Google Business Profile data, Common Crawl, and structured data published by websites themselves. When these sources are coherent and mutually reinforcing, Google can assert with high confidence that entity X exists, belongs to category Y, and has attributes Z.

The disambiguation process is the most critical. Google assigns each recognised entity a unique identifier called a KGMID (Knowledge Graph Machine ID). When multiple sources consistently mention a brand in the context of “SEO agency Barcelona”, Google unifies those mentions under the corresponding KGMID. When sources are inconsistent — different names, varying descriptions, contradictory data — Google may create duplicate entities or simply fail to recognise the entity at all.

In June 2025, Google removed millions of entities from the Knowledge Graph in a clean-up operation that prioritised quality and consistency over volume. Ambiguous, outdated, or contradictory entities were the first candidates for removal. For a brand that has built its digital presence haphazardly — with different names across channels, no structured data declaring its identity, no Wikidata entry — the practical consequence is invisibility to LLMs.

sameAs and Wikidata: declaring your identity to Google

The sameAs property in the Organisation schema is the most direct identity declaration you can make to Google. It asserts that the entity described on your website is the same as the one registered in Wikidata, LinkedIn, Wikipedia, or other authoritative sources. Google uses those links to unify your brand’s mentions scattered across the web under a single Knowledge Graph node. To integrate this schema markup on your site, the implementation is an array of URLs in the JSON-LD on your homepage:

{
  "@type": "Organization",
  "@id": "https://ighenatt.es/#organization",
  "name": "Ighenatt",
  "url": "https://ighenatt.es",
  "sameAs": [
    "https://www.wikidata.org/wiki/Q[YOUR_ID]",
    "https://www.linkedin.com/company/ighenatt",
    "https://www.crunchbase.com/organization/ighenatt"
  ]
}

The URLs to include in sameAs follow a priority order by authority: Wikidata (highest priority, primary KG source), Wikipedia (where an article exists), LinkedIn (for organisations), Crunchbase (for startups and tech companies), Google Business Profile URL, and sector reference directories.

Wikidata is the most important step and the most commonly overlooked. Unlike Wikipedia, creating a Wikidata entry requires no prior media coverage: any verifiable entity can have one. The process takes under an hour: you create the item with a name and basic identifiers (P856 for official URL, P17 for country, P571 for founding date, P452 for industry), publish it, and link it to your schema via sameAs. That single act transforms your brand into an entity that Google can process with verifiable structured data.

Unlinked brand mentions: link building for the AI era

Language models learn about brands through the co-occurrence of words in the texts they are trained on, not through the web’s link structure. This principle explains the central finding from the Ahrefs study: brand search volume — how many people search for your name — predicts LLM citations with a correlation of 0.334. Media mention volume scores 0.664. Backlinks, 0.218.

A mention in a reference digital publication, even with no link to your website, contributes more to ChatGPT citing you than a backlink from a low-authority directory. The reason is that LLMs are trained on text, and text featuring your name in relevant contexts — SEO agency, client results, tool comparisons, sector articles — is what builds your entity’s reputation in training data.

The practical strategy has three dimensions. First: digital PR in sector media with articles that name the brand in a relevant context (interviews, mentions in rankings, co-authored pieces in reference publications). Second: registration in authoritative sector directories (agency guides, comparison platforms, professional associations). Third: co-citation with high-authority entities — when your brand appears in the same text as recognised names in the sector, Google infers relationship and relevance through semantic proximity.

To audit existing mentions, tools such as Google Alerts, Mention, or the search operator "brand name" -site:yourdomain.com in Google will give you a list of current co-occurrences. Unlinked mentions in high-DA media are candidates for classic link-building outreach, but their value as entity signals exists independently of whether you secure the link.

NAP consistency and entity consolidation for local businesses

For businesses with multiple locations or branches, NAP (Name, Address, Phone) consistency across all digital touchpoints is Entity SEO applied at local scale. Brightview Senior Living, with 47 locations across the United States, documented this problem: near-me searches returned inconsistent results because Google held multiple partial entities for the same sites, with slightly different names depending on the directory.

The solution was to link each location entity to Wikidata via sameAs and implement a LocalBusiness schema with unique identifiers per site. The result was the consolidation of duplicate entities and consistent visibility in high-intent local searches, without any additional link-building campaign.

To implement this: each location needs its own unique @id in the JSON-LD, exact name consistency across Google Business Profile, local directories, and the website schema, and a sameAs pointing to the same identifier in Wikidata or Google Business Profile. Tools such as Moz Local or Semrush Listing Management automate consistent NAP distribution across directories, reducing the risk of duplicate entities.

How to measure your entity presence in the Knowledge Graph

The Knowledge Panel is the most visible indicator of entity recognition. Search for your brand name: if the side panel appears with organisation information, Google recognises you as an entity. If it does not, it does not recognise you — regardless of your ranking for other queries.

Beyond the panel, there are three proxy metrics that track entity-building progress. Branded search volume, available in Google Search Console under “brand queries”, shows how many people search for you by name: it correlates directly with LLM citations according to the Ahrefs study. The number of indexed press mentions, measurable with tools like Ahrefs Content Explorer, reflects the volume of co-occurrences that AI models can process. The rate of appearances in AI Overviews for brand keywords, visible in Search Console from December 2025, is the most direct measurement of Entity SEO’s effect within the AI ecosystem.

Google’s Knowledge Graph Search API lets you query directly whether your organisation has an assigned KGMID. An empty response indicates absence as a recognised entity; a response with @id and a high resultScore confirms recognition and allows comparison against competitors.

From entity to citation: how LLMs decide whom to mention

The mechanism by which an LLM decides to cite a brand has more in common with reputation than with ranking. Models are trained on text in which certain brands appear in authoritative contexts: “best agencies” lists, sector reference articles, founder interviews, case studies. That training text determines which brands are in the model’s vocabulary for a given category.

For real-time retrieval systems — Google AI Overviews, Perplexity, ChatGPT’s search mode — the relative weight of live sources is higher. This is where Entity SEO connects directly with the strategy for appearing in AI Overviews: pages that Google indexes as part of the entity (official website, articles authored by people within the organisation, verified profiles) are given preference as answer sources.

Only 11% of domains are cited simultaneously by ChatGPT and Perplexity, according to TheDigitalBloom’s AI Visibility Report (2025). This means that building entity presence on a single engine is not enough: strategy must be diversified across training sources (Wikipedia, reference media, open datasets) and retrieval sources (regularly updated content on the official site, fresh structured data schema, presence on platforms that LLMs use as real-time sources).

The analysis of E-E-A-T factors in content converges with Entity SEO at a fundamental point: Google trusts content produced by recognised entities more. Once your brand exists as a verified node in the Knowledge Graph, every article, study, or guide you publish inherits part of that accumulated trust. And to also understand how AI bots crawl your site before indexing that content, analysing AI bot logs such as GPTBot and ClaudeBot is the natural next technical step.