The Entity Before
the Keyword

Quick answer

Entity SEO helps search systems identify the people, organizations and concepts described by your content. I use Entity Density as an editorial framework for maintaining consistent facts, connected identifiers and verifiable corroboration. It is not a ranking metric or a guarantee of AI citations.

The identity mistake on my own site

My earlier disambiguation copy referred to an Ecuadorian footballer and a Colombian boxer. It appeared in visible text, structured data and llms.txt. Checking the identities behind those descriptions exposed the weakness: I had repeated a distinction without verifying its targets.

The references used in the correction were Fernando Angulo, the Spanish basketball player, Fernando Angulo, the Peruvian ornithologist, and my own entry. These entries are useful reference points, not an exhaustive list of everyone with the name. An absent search result does not prove that a person does not exist.

This is a case of correcting published identity information. I do not have a retained before-and-after answer panel that demonstrates a resulting increase in AI accuracy or citations.

Diagnose before you optimize

Entity Density: three maintenance practices

I call this maintenance framework Entity Density. The term describes my approach; it has no standardized score or validated causal weight. Work through the following three practices to make your published identity easier to inspect and maintain.

1. Keep the facts consistent

Maintain a source-of-truth record for name, current role, organization, canonical website and verified profiles. Assign an owner and record when each fact was checked. A shortened biography or a native Spanish description can express the same identity without copying every character. Investigate contradictory facts, not harmless wording differences.

When a role changes, update current profiles and date historical biographies rather than rewriting past affiliations. Record legitimate aliases and former names where they help readers understand continuity.

2. Connect identifiers to the right entity

Use a stable @id for the person or organization in your structured data, and reuse it when another page refers to that same entity. The example below shows a person linked to a profile that identifies that person. It is an illustrative pattern, not extra live markup for this article.

{
  "@context": "https://schema.org",
  "@type": "Person",
  "@id": "https://fernandoangulo.com/#fernando-angulo",
  "name": "Fernando Angulo",
  "url": "https://fernandoangulo.com/",
  "sameAs": [
    "https://www.wikidata.org/wiki/Q138975073"
  ]
}

@id identifies the person within the graph; url points to the website. sameAs links to a page that unambiguously identifies the same entity. Do not put an employer, a namesake or a loosely related article into that list. An organization should have its own identifier and appropriate type.

Check each URL manually, ensure the facts agree with visible content, and validate the markup with the Schema.org validator. Validation checks the markup, not whether an AI system will use it. Google’s AI-search guidance says there is no special schema or new AI text file required for its AI features.

3. Find corroboration you can verify

Look for relevant organizer biographies, editorial profiles, research bylines and professional records that identify the subject clearly. Record who controls each source: your own profile, an employer’s page and an independent editorial reference provide different kinds of evidence. A frequently cited domain is not automatically an authoritative source for your particular identity.

No Wikipedia or Wikidata entry? Start with an accurate official biography, verified professional profiles and relevant external references. You can implement this maintenance process without an encyclopedia entry. Add an external identifier only when it exists and matches the subject; do not manufacture an entry to complete a checklist.

Measure accuracy separately from citations

Before changing a source, save a small fixed set of natural questions, such as “Who is [name] at [organization]?” and “What does [name] research?” These are proposed branded diagnostic prompts, not evidence of search demand. Use separate, unbranded topic questions to investigate discovery.

1. Capture a baseline. Save the exact prompt, answer, cited URLs, platform, model or mode, search setting, language, market and date. Record unavailable fields as unknown.
2. Check facts against sources. Label correct, incorrect and unverifiable statements. Record the specific error rather than assigning a general visibility score.
3. Log the intervention. Keep the old text, new text, source URL, owner and publication date. Save snapshots where possible.
4. Repeat the same panel. Use fresh conversations and the same settings. A practical starting protocol is three runs per prompt before and after the update; this is a consistency check, not a statistically conclusive experiment.

Track four separate outcomes: identity accuracy, brand mention, domain citation and exact-page citation. Report accuracy as correct checkable claims divided by all checkable claims, and report unverifiable claims separately. For mentions and citations, retain the count of successful responses as the denominator. A failed request is missing data, not a negative result.

Choose and record the follow-up interval before collecting results. Source refreshes, changing retrieval and model updates can affect answers. A change between rounds is an observation; it does not isolate the effect of your edit. For the broader distinction, see citation authority.

The idea predates AI search

Google’s 2012 Knowledge Graph announcement made “things, not strings” a useful public explanation of entity-based search. It is a historical milestone, not the birth of all work on entities or semantic search.

Generative engine optimization (GEO) concerns visibility in AI-generated answers. My practical argument is that incorrect identity information deserves attention alongside content quality and discoverability. It does not depend on a universal claim that every engine resolves identity before relevance. Google describes different models and techniques, including related searches across subtopics, in its AI features documentation.

Limits and multilingual maintenance

Clear identity cannot make weak research worth citing, and valid markup cannot guarantee retrieval. An inaccurate answer may also involve stale sources, retrieval choices or unsupported generation. Investigate those alternatives before calling it an entity-resolution failure.

Keep the underlying identity stable across English and Spanish while writing naturally in each language. Audit role descriptions, linked profiles and dated affiliations in both. Test each language independently; a result in one market does not establish performance in another. The wider question belongs in Cross-Lingual Citation Authority.

Your next move: audit one spokesperson or brand, document one verified discrepancy, and establish the answer baseline before claiming an improvement.