6 August 2026

    Keywords are not entities. How to build content around what users actually mean?

    Keywords reveal how people search. Entities, attributes and relationships explain what content needs to clarify. Learn how to connect both layers in a practical content model.

    Compass and measuring tool on a map, illustrating the process of connecting keywords, entities and search intent
    photo by Alexander Andrews | Unsplash

    Keywords show how people ask. Entities identify what, whom or which relationship an answer must explain. If a strategy treats those layers as identical, a page may match a query while leaving its subject and practical significance unclear.

    A stronger approach does not replace keyword research with “entity optimisation”. It uses keywords as evidence of demand, then connects them to intent, entities, relationships, claims and evidence. The result is a content model, rather than a list of terms to repeat.

    Keywords and entities describe different layers

    A keyword is the textual form of a query, or part of one. It can reveal a topic, audience language, selection criteria and decision stage. Queries such as “CRM for a small business”, “affordable CRM” and “CRM with sales automation” express related needs, but do not yet define a content model.

    An entity is an identifiable thing or concept about which something can be said and which can be distinguished from other things. It may be a person, organisation, product, place, event, technology or defined concept. An entity has attributes and relationships with other entities.

    One expression can refer to several entities, while one entity can have many names. Research into entity linking and disambiguation treats spotting a mention and assigning it to the correct object in a knowledge base as separate tasks. Context and relationships help resolve ambiguity.

    Do not assume that every entity has a universal identifier shared by Google, search engines, knowledge bases and language models. Systems can use different representations and recognition processes. Google Cloud Natural Language documentation illustrates one approach, not Google Search ranking systems.

    A keyword list is not yet a content model

    Keyword research remains essential: it reveals audience language, variations of a problem and likely demand. On its own, it does not answer:

    • Which entity is the page’s central subject?

    • What does the user need to establish about it?

    • Which attributes affect understanding, evaluation or choice?

    • Which relationships need explaining?

    • Which claims require evidence?

    • Should related queries lead to one page or several specialised URLs?

    Consider “best CRM for a small sales team”. It does not contain one entity. It combines a product category, an audience type, a use case and a recommendation criterion. “Best” is not an entity either. It signals comparative intent and a need for explicit criteria.

    A plan based only on similar phrases may repeat “CRM”, “small business” and “sales”. A plan that includes entities and relationships asks about users, pricing, automation, integrations, implementation, security and product limits. Those details make an answer usable.

    The keyword-to-entity binding matrix

    A practical tool is a keyword-to-entity binding matrix. It is not a list of names to insert into copy. It is a decision model that links search language with meaning, information structure and evidence.

    The order is deliberate: identify the scenario, the central entity, then its attributes, relationships and claims. Only then decide the page structure.

    Example: from “AI visibility audit” to an entity map

    Imagine a cluster including AI visibility audit, AI brand audit and how to measure visibility in ChatGPT. It may conceal different needs: a definition, research method, tool selection or service purchase.

    The table shows why a query cannot be the sole planning unit. A page about an AI visibility audit should explain what it measures, what material it uses and by which criteria. A separate guide can cover how to run an AI visibility audit.

    Semantio is an instrumental entity, not the article’s subject. It becomes relevant when a one-off map turns into repeatable research across scenarios, responses, sources and time. This makes monitoring brand representation in AI responses a useful next step without turning the article into a sales page.

    Turning the matrix into page architecture

    Group queries by scenario, not only word similarity

    Two phrases can look similar while serving different tasks. “What is an AI visibility audit?” calls for a definition and scope. “AI visibility audit tool” may signal tool selection. “AI visibility audit agency” is more transactional. “How to measure visibility in ChatGPT” may need a methodology for one specific product surface.

    This does not automatically justify four pages. Assess whether intents can be served on one URL and whether each proposed page has distinct value.

    Choose one central entity

    A page can include many entities, but needs a clear subject. A text covering an audit, tool, platform and consultancy at once has no clear purpose.

    The central entity need not be the most popular expression in keyword research. It should be the object whose explanation best resolves the dominant user need.

    Select supporting entities for usefulness

    Supporting entities do not belong in a text merely because competitors use them or an NLP tool detected them. Each must define, distinguish, explain, support a decision or reduce misunderstanding.

    For an AI visibility audit, useful concepts may include scenario, prompt, run, response, mention, recommendation, citation, source, claim and stability. They describe different parts of the research. Collapsing them into one “AI visibility” category makes results harder to interpret.

    Assign claims to sections and evidence

    An entity map tells you what needs discussing. A claim map tells you what will be said and on what basis. The distinction is developed in entity mapping, claim mapping and source alignment.

    For example, “a mention of a brand is not the same as a recommendation” is a claim. Support may be a methodological definition, documentation, response set, study, experiment or first-party data. Readers should see the claim’s status and where evidence ends.

    Treat internal links as records of relationships

    An internal link should develop a specific relationship or answer the reader’s next question. A definition can link to a procedure; a procedure to a tool; a case study to its method and data.

    This is better than mechanically matching anchors. Google’s guidance on crawlable links recommends concise, natural, descriptive anchor text placed in context that makes the destination’s relevance clear.

    Writing content with semantic clarity

    Semantic clarity comes from defining objects, distinguishing them from similar concepts and explaining relationships in complete, verifiable sentences.

    Four editorial principles help:

    1. Define. At first mention, name an entity precisely enough for readers to know what it is. Definitions matter most when a term is new, ambiguous or used in several ways.

    2. Distinguish. State the boundary: a mention is not a recommendation; a citation is not proof that an entire source was used correctly; a query cluster is not an entity map; structured data is not user-visible content.

    3. Connect. State relationships explicitly: a product serves a group; a feature addresses a problem; an audit assesses selected systems; a claim requires evidence.

    4. Evidence. Match support to the statement. Documentation can confirm a feature; research can report a sample; first-party data can document an observation. Do not automatically generalise across systems, languages or markets.

    This approach aligns with Google’s guidance on helpful, reliable, people-first content, which asks for original information, analysis or value beyond a simple rewrite of existing results. An entity list cannot replace that value.

    Where structured data helps, and where it does not

    Structured data can signal the type, properties and relationships of objects described on a page. The schema.org data model provides types and properties for representing entities and connecting them.

    Useful properties include mainEntity, for an entity principally described by a page; about, for its subject; sameAs, for an unambiguous identity reference; and @id, for consistent reference inside a JSON-LD graph.

    Markup should describe content visible on the page, not add hidden facts to “enrich” an entity. Google’s structured data policies require it to be current, accurate and representative of the main content. Valid markup does not guarantee a rich result.

    Schema.org is not a guarantee of entity recognition, ranking, AI citation or accurate brand representation. Structured data is descriptive; it does not replace content, a coherent source ecosystem or outcome measurement.

    What this does not mean

    Connecting keyword research with an entity map does not mean that:

    • Entity SEO replaces keyword research.

    • Every keyword is an entity, or every entity needs its own page.

    • A page should include every entity found on competitor sites.

    • Entity-name frequency is a simple measure of quality or ranking potential.

    • A salience score from one NLP tool reveals how Google Search evaluates a page.

    • A term appearing in a patent proves it is used in a current ranking system.

    • Extensive schema markup creates topical authority or guarantees AI visibility.

    • A mention, citation and recommendation are equivalent outcomes.

    • One prompt can assess brand representation across an entire platform.

    In AI search, separate what can be controlled from what can only be influenced and observed. A brand controls its own pages, definitions, data and structure, but not final synthesis, recommendation or citation. The control, influence and observation model for GEO explains that boundary.

    A short operational checklist

    • Do we know which user scenario the page serves?

    • Can we name its central entity in one sentence?

    • Have we separated the entity from its names, synonyms and abbreviations?

    • Have we covered the attributes needed for understanding or a decision?

    • Are relationships explicit and factually correct?

    • Does every section perform a function rather than merely add terms?

    • Do important claims have appropriate evidence or a stated status?

    • Does the page have a clear role among related URLs?

    • Do internal links develop genuine topical relationships?

    • Does structured data reflect visible content?

    • After publication, do we measure not only traffic and rankings, but also the accuracy of representation in AI responses?

    From queries to organised knowledge

    Keywords remain valuable evidence of demand and audience language. Entities should not replace them. They move us from query wording to the things, products, problems and relationships a user is trying to understand.

    The best result is not a page with more proper nouns. It is one with an unambiguous subject, recognised intent, necessary relationships, verifiable claims and useful onward links.

    The keyword-to-entity binding matrix structures that path: from query language, through intent and entities, to page structure, evidence and later measurement. That is how entity SEO becomes a method for designing semantically clear, useful information, rather than an exercise in adding terms.



    Michał Grzebyk
    Michał Grzebyk
    COO Brand Semantics

    Co-founder of Brand Semantics. Engaged in marketing since 2009. Trainer. Strategist. Explorer of new frontiers in modern marketing. Integrates knowledge from diverse fields to deliver innovative business solutions for clients.