Semantic Keywords: What They Are and How to Find Them

SEMANTIC KEYWORDSKEYWORDVARIANTBROADERATTRIBUTEENTITYQUESTION

Semantic keywords are the words and phrases that share meaning with a target term: its variants, its attributes, the entities around it, and the questions people ask about it. This article defines semantic keywords and separates them from LSI keywords, synonyms, and long-tail keywords. It then explains why a search engine reads those terms as meaning, and sets out the five types of semantic keyword in one table. It follows with six steps for finding semantic keywords, the rules for semantic keyword grouping, which decide the keywords that belong on one page, the places each one is written, five common mistakes, and a checklist for a page. It closes with a worked example from holisticradar.com, the link between semantic keywords and the query network, the tools that return them, and the questions readers ask.

What semantic keywords are

Semantic keywords are the words and phrases that share meaning with a target term: its variants, its attributes, the entities around it, and the questions people ask about it.

The term is an industry label and not a Google term. Google’s documentation describes queries, entities, and intent, and it publishes no list of semantic keywords for any query. A semantic keyword is a working concept: a term a writer includes because the subject requires it, and not because a tool scored it.

The concept exists because of retrieval by meaning. A search engine matches a query to a document by the entities and relations both contain, so a page about knee replacement is expected to contain terms such as implant, recovery, and surgeon without being told. Semantic keywords name those expected terms before the page is written.

Semantic keywords versus LSI keywords, synonyms, and long-tail keywords

Semantic keywords are not LSI keywords, because latent semantic indexing is a retrieval technique described in 1990 for small, closed document sets, and Google’s John Mueller has said there is no such thing as LSI keywords.

TermWhat it isRelation to semantic keywords
KeywordThe string a searcher typesThe input; semantic keywords describe the meaning around it
SynonymA word with the same meaningOne of the five types, the variant
Long-tail keywordA longer, more specific queryA query shape; it contains semantic keywords and is chosen by length
LSI keywordA label for terms that co-occur with a keywordA misnomer; no search engine publishes or uses an LSI keyword list

The practical difference is the source. A synonym comes from a thesaurus, a long-tail keyword comes from a volume report, and a semantic keyword comes from the subject: the entity, its attributes, and the questions the entity raises. The semantic SEO article answers the LSI question in one place, and this article does not repeat it.

Why a search engine reads semantic keywords as meaning

A search engine reads a page as entities and relations, so the terms on a page act as evidence for the entities the page describes.

Google built that reading in stages: the Knowledge Graph in 2012, Hummingbird in 2013, RankBrain in 2015, and BERT, applied to Search on 25 October 2019. BERT made the small words in a query change what the query means, so a page is matched to the whole request and not to its largest words. The sequence is covered in semantic search.

Two consequences follow for keyword work. A page ranks for wordings it never uses when its entities and attributes match the query, so a missing synonym does not cost a ranking. A page loses when its terms point at a different entity, so an ambiguous term is the larger risk than a missing one.

Passage ranking adds a third consequence. Google’s ranking systems guide lists passage ranking as an active system, so a section of a page is scored on its own, and each section carries its own entity and attribute terms in place rather than only in the introduction.

The five types of semantic keyword

There are five types of semantic keyword: variants, hierarchy terms, attributes, related entities, and question forms.

TypeDefinitionExample for an espresso machine
VariantA synonym, plural, abbreviation, or spelling of the termEspresso maker, coffee machine
Hierarchy termA broader or narrower termKitchen appliance; pump espresso machine
AttributeA property the entity hasBoiler type, pump pressure, portafilter size, tank capacity
Related entityAnother entity that occurs with the firstGrinder, barista, crema
Question formA question asked about the entityHow to descale an espresso machine

Attributes carry the most weight, because an attribute is what a page asserts about its entity. The other four types support an attribute: variants widen the wordings it matches, hierarchy terms place it, related entities supply its context, and question forms supply its headings.

How to find semantic keywords in six steps

Finding semantic keywords takes six steps, and the order matters because the first two fix what a term means before the later steps supply volume.

  1. Name the entity. Choose the exact entity, not the string, and confirm it against a reference such as Wikipedia or Wikidata.
  2. List its attributes. Read the reference entry’s sections and properties, and the headings of the pages that already rank.
  3. Collect the queries. Pull the queries the site already receives from Search Console, then autocomplete suggestions, People Also Ask questions, and related searches.
  4. Collect the shared terms. Read the top five results and list the entities and attribute words that at least three of them contain.
  5. Group by intent. Put queries that one page answers into one group, and queries that need a different answer into another.
  6. Assign and label. Give each group to one page, and label every term with its type from the table above.

The output of the six steps is a labelled list per page, not a flat list per site. A flat list repeats the keyword problem, because it shows which strings exist and not where each belongs.

Semantic keyword grouping: which keywords belong on one page

Semantic keyword grouping, also called semantic keyword clustering, is the assignment of each semantic keyword to the one page whose entity, heading and search intent it matches. A semantic keyword belongs on a page when it shares the page’s entity, answers one of the page’s headings, and matches the search intent of the page type the results show.

Three tests decide each term. The entity test asks whether the term describes the page’s entity or a different one. The heading test asks whether the term answers a question the page already asks, and a term that needs its own answer needs its own page. The intent test asks whether the searcher expects the same page type: a term that signals buying belongs on a service page, and a term that signals learning belongs on an article.

No fixed count applies. The number of semantic keywords on a page follows from its headings, and a page with eleven attribute headings carries more of them than a page with four. A density target adds nothing, because the search engine reads entities and not repetition.

Ecommerce grouping applies the same three tests to category and product URLs. A term that names a product type and signals buying groups onto its category page, a term that compares or explains groups onto an article, and a term that needs its own answer gets its own URL. Semantic SEO for ecommerce sets out the category and product topical maps that this grouping fills.

Where semantic keywords are placed

Semantic keywords are placed where a reader and a parser both look for the answer: headings, first sentences, bold answer terms, link anchors, and the title.

  • Headings. Each attribute gets one heading, written as the attribute name.
  • First sentence under a heading. The sentence states the attribute value in full, in the present tense.
  • Bold answer terms. The term that answers the heading is bold once, as the microsemantics rules specify.
  • Link anchors. The anchor text names the relation between the two pages and not only the destination.
  • Title tag. One primary term, with the attribute that separates the page from its neighbours.

The same term is written in the same words on every page that touches it. Consistency keeps the entity single, and a second wording for the same attribute splits it.

Semantic keyword mistakes

Five mistakes account for most failed semantic keyword work, and each has a test.

MistakeWhy it failsTest
Accepting a tool’s synonym list as the answerThe list holds variants only, one of five typesEvery heading has an attribute behind it
Chasing LSI keywordsNo search engine uses such a listEach term traces to an entity or attribute in a reference
Setting a density targetThe engine reads entities, not repetitionA repeated term that adds no new fact is cut
Mixing intents on one pageThe page matches no single results typeEvery heading answers a question from one group
Writing one attribute two waysThe entity splits across wordings and pagesA site search for the second wording returns nothing

The tests are run on the draft, before publication. A page passes when all five return the expected result, and a failed test is repaired in the brief and not in the prose, because the brief is where the entity, the intent, and the headings are fixed.

A checklist for a page’s semantic keywords

  • The page names one entity, and the name matches the reference entry.
  • Each heading is an attribute or a question form of that entity.
  • Each term is labelled with one of the five types.
  • Every term shares one search intent with the page’s head term.
  • The same attribute is written the same way on every page of the site.

Semantic keywords on holisticradar.com

Holistic Radar treats the central entity Semantic SEO as a set of attributes, and each attribute that earns a page is a semantic keyword with a named owner.

Semantic keywordTypeOwning page
Semantic searchAttribute: the search engine side/learn/semantic-search/
Entity SEOAttribute: entity resolution/learn/entity-seo/
Query networkRelated entity: the query side/learn/query-network/
Topical authorityRelated entity: the outcome/learn/topical-authority/
Structured dataAttribute: the technical layer/learn/structured-data/
LSI keywordsVariant that is a misnomerAnswered inside /learn/semantic-seo/

The table is the method applied to one site. A term with its own question and its own page type gets its own owner, and a term that only supports an answer stays as a mention inside the owner. The last row shows the second case: the LSI question is answered in one section of the semantic SEO article, and no separate page exists for it.

Semantic keywords and the query network

Semantic keywords are the output of query research, and the query network is the structure that groups them.

A keyword list says which strings exist. A query network says which strings belong on one page, which come first, and how the groups relate, because it records the behavioural relationships between queries. The same semantic keyword appears in the list once and in the network with a position: head term, supporting term, or question form.

That difference changes what is published. A site working from a list publishes one page per string and splits one question across several URLs. A site working from a network publishes one page per group and links the groups in the order readers move through them.

Semantic keyword tools

Semantic keyword tools return one of four kinds of output, and a tool is only as useful as the kind it returns.

Tool kindExamplesReturnsDoes not return
Engine surfacesSearch Console, autocomplete, People Also Ask, related searchesReal queries and question formsAttributes or entity relations
Reference sourcesWikipedia, WikidataAttributes and related entitiesSearch volume
Entity analysisGoogle Cloud Natural Language API, open-source named entity recognitionEntities found in a text, with salience scoresWhat the page is missing
Keyword research suitesThird-party volume and difficulty toolsVolume, difficulty, variantsIntent grouping or page assignment

A synonym generator returns variants only, which is one of the five types. Entity analysis shows which entities a draft contains and so checks the result of the six steps. The types of SEO tooling are compared in the article on semantic SEO tools, and entity analysis with working code is in semantic analysis for SEO.

Semantic keyword questions

Are semantic keywords a ranking factor?

Semantic keywords are not a ranking factor, because Google lists no such signal. They matter as evidence: a page that contains the expected attributes and related entities shows the search engine that it covers the subject.

How many semantic keywords does a page need?

A page needs as many semantic keywords as its headings require. The count follows the attributes the entity has and the questions the intent raises, and no number is correct for every page.

Are semantic keywords the same as LSI keywords?

Semantic keywords are not LSI keywords. Latent semantic indexing is an older retrieval technique, and Google has said it has no LSI keywords.

Do semantic keywords replace the primary keyword?

Semantic keywords do not replace the primary keyword. The primary keyword names the page’s head intent, and the semantic keywords extend it with the attributes, entities, and questions that intent contains.

Semantic keywords matter for AI search, because answer engines extract entities and attributes from passages, and the terms that name them decide whether a passage is liftable as an answer.

Semantic Keywords in the semantic SEO topical map

Semantic Keywords is the keyword attribute of semantic SEO, the central entity of the Holistic Radar topical map, and sits in the outer section under the Know predicate.

The outer section holds the page because the page teaches the concept without a direct sale, so internal link equity flows from it to Query network research, since keywords are the output of query research and the network groups them.

→ Query Network: how keywords are grouped by shared intent

→ Semantic Search: why meaning, not word overlap, matches a query to a page

Sources

The engine history comes from Google’s own announcements and ranking systems guide; the worked example comes from holisticradar.com’s own pages.

  1. Google, Understanding searches better than ever before (2019)
  2. Google Search Central, A guide to Google Search ranking systems
  3. Google Search Central, Creating helpful, reliable, people-first content
  4. Google Cloud, Natural Language API: analyzing entities
  5. John Mueller, post on X about LSI keywords (2019)
Get my free Radar ScanFree Radar Scan