Free Semantic SEO learning

Make entity prominence clear without counting mentions

Salience concerns prominence in context. Co-occurrence means things appear together. Proximity concerns how near their mentions are. None of these ideas supplies a universal entity density, sentence distance or writing score that guarantees search performance.

Created by Kevin Maguire.

Two paper compositions contrast a central subject with a clutter of unrelated shapes.

Separate three different observations.

A subject can be prominent because the title names it, the introduction defines the task, and the body explains its relevant properties. That editorial prominence is more useful to a reader than repeating its name without adding information.

Co-occurrence is weaker evidence: two names appear in the same passage or document. They might be compared, confused, denied or mentioned incidentally. Their shared presence does not establish identity, ownership, causation or endorsement.

Proximity describes placement. A qualification placed near a claim can help a reader connect them. That is a communication decision. It should not be converted into an unsupported rule that every entity must occur within a fixed number of words of another.

Three observations, three different limits

Salience

Describes: How prominent a subject is in the passage or document being examined.
Editorial question: Can the reader identify the subject and follow its relevant properties?
Counterexample: A repeated company name can still leave its service unexplained.
Limit: Editorial prominence and a provider’s calculated salience score are different observations. Neither supplies a universal ranking formula.

Co-occurrence

Describes: Two names or concepts appearing in the same input.
Editorial question: What relationship does the text assert between them?
Counterexample: A denial of ownership contains both the company and product names.
Limit: Shared appearance alone establishes neither identity nor ownership.

Proximity

Describes: The placement of mentions relative to one another.
Editorial question: Is the reference or qualification close enough for the reader to connect it?
Counterexample: Two adjacent names can still describe unrelated companies.
Limit: There is no fixed mention distance prescribed by this editorial method.

Read the predicate before inferring a relationship

The predicate tells the reader what is asserted about the subject. Compare these fictional sentences: “Alder Heating owns Northfield Controls”; “Alder Heating does not own Northfield Controls”; and “Alder Heating is sometimes confused with Northfield Controls.” Each contains the same two names, but each asserts a different relationship.

A list of names loses the distinction between ownership, denial and reported confusion. Keep the statement’s date and source attached to the relationship if it describes a real event. If the identities are uncertain, disambiguate the names before drawing a connection. The attributes and predicates explanation covers the relationship record.

Separate navy and blue folders with a loose blank note between them.

Compare equal mentions with different clarity.

The following short examples are fictional writing samples. Both mention Alder Heating twice. The count stays the same while the quality of the explanation changes. They are not a live search experiment or evidence of a ranking effect.

Sample A: “Alder Heating offers heating. Alder Heating has heating services, and those services concern heating.” The company name is repeated, but the passage leaves the service, area and conditions unclear.

Sample B: “Alder Heating offers heat pump maintenance in the fictional Northfield area. Before booking, Alder Heating asks customers for their system model so it can confirm the service is appropriate.” The second passage gives the reader a relationship, location and condition. Those invented details would require verification on a real company page.

The improvement comes from the information and its clarity, not a higher mention count. A tool that only counted the company name would miss that distinction.

Open blank book holding a house and service toolbox miniature.

What the second sample adds

Service and applicability

Service: Heat pump maintenance, rather than an unspecified heating offer.
Place: The fictional Northfield area.
Condition: The system model is needed before suitability can be confirmed.
Evidence boundary: Every company fact in the sample is invented and needs verification before use for a real business.

What remains unmeasured

Name count: Two mentions in each quoted sample.
Observed difference: One passage names the service, area and suitability condition.
Not tested: Rankings, traffic, reader completion and any provider’s salience score.

Keep a relationship and its qualification together.

A sentence such as “The company services this model” needs the correct company and model to be identifiable. A later paragraph about an unrelated product can make a vague pronoun confusing. Use a clear name when the reference would otherwise be uncertain.

Necessary scope belongs close to the statement it limits. If support is available only for selected models or within a defined area, preserve that condition beside the claim. Do not depend on a distant disclaimer to repair a misleading general statement.

Nearby words still do not prove a relationship. A sentence explaining that two companies are unrelated contains both names. A draft review must read the predicate and the surrounding qualification rather than treating mention distance as factual evidence.

For example, “Alder Heating supports model N within Northfield” joins the subject, supported model and service area in one statement. This is another fictional sentence. A later reference to “it” must still identify the same company or model clearly. The choice to keep these facts together concerns reading clarity, not a rule about how many words may separate entity names.

House miniature and location pin together on an open blank book beside a magnifying lens.

Use tool outputs as prompts for review.

A language model or analysis service may produce its own salience score. The score depends on that tool’s method, input and configuration. Record those details if the score informs a comparison. Do not present it as a universal measure or an exposed search engine signal.

A list of co-occurring entities can help identify questions the draft has not considered. It can also contain irrelevant or misleading suggestions. Include a subject only when it serves the reader’s task and reliable evidence supports its treatment.

Google’s helpful content guidance focuses on useful, reliable explanations for people. Our clarity checks apply that editorial aim; they are not a published equation for prominence or proximity.

Interpret a salience score within its own method

Before comparing outputs, record the provider, model or version, exact input passage, preprocessing, scoring scale and observation time. Retain what was removed or normalized, such as navigation, punctuation or repeated template text. A comparison of a clean passage with an entire page can describe a change in input rather than a change in the writing.

An open blank folio beside a blue booklet, colored record tabs and an amber clip.

A comparison record without invented scores

T1: review requirements

Provider and version: Enter the observed service and its documented version; leave unknown details open.
Input: Save the exact passage and identify the document version.
Preprocessing: State text extraction and normalization choices.
Scale: Record the provider’s definition and permitted range.
Observation time: Record when the request was made.
Comparability: Compare like inputs under the same method; explain any difference.
Result: No tool request or numeric result is supplied by this teaching record.

Read about contextual language and embeddings for the limits of model representations. Choose the main subject from the reader task and evidence, rather than from a score alone.

Review the passage before adding another mention.

Ask a reader to identify the subject, the relationship and the condition from the passage alone. If the answer is unclear, repair the explanation. Adding a name again may solve a reference problem, but it cannot supply missing evidence or a missing fact.

Read the page as a whole after local edits. A useful passage can still belong under the wrong heading or distract from the main task. Preserve a clear route from the title through the explanation to the next relevant question.

Pen beside a page draft and a revised page with highlighted content blocks.

Completed fictional edit record

ER1: equal counts

Original passage: Sample A above, with two company name mentions.
Confusion: The service and its applicability are unspecified.
Changed wording: Sample B above, also with two company name mentions.
Retained limit: Suitability depends on the customer’s system model.
Evidence: Invented teaching text; company facts require confirmation.
Review check: A reader can identify who offers which service, where, and under which condition.
Next question: Check aliases and names if naming itself creates confusion.

Repair an ambiguous reference

Fictional draft: “Alder Heating contacted Northfield Controls after it changed the service terms.” The pronoun could refer to either company. If the verified event concerns Northfield Controls, write “Alder Heating contacted Northfield Controls after Northfield Controls changed its service terms.” The repeated name resolves the reference; it does not add proof that the change happened.

If the event is unverified, keep it as an open input instead of choosing a subject for convenience. After the local repair, check the surrounding headings and task. A precise sentence still needs to belong on the page.

An ivory sheet retained in a navy sleeve beside a separate white sheet, pencil and amber clip.

Sources and the next useful question.

Check Google’s helpful content guidance. External documentation supports the stated model or editorial boundary; the fictional examples and record formats are Semantec SEO teaching material.

Continue with aliases, names and terminology. Return to the entity SEO overview or the semantic SEO learning overview to choose another task.

Choose the next question

Return to the Semantic SEO learning library to choose another explanation. If you want help applying this work to a website, send Kevin the site and the task you need to address.