Microsemantics: The Nine Sentence Rules That Make Facts Extractable

SENTENCE AS TRIPLEPropXproruns on216 pagesSUBJECTPREDICATEOBJECTPRESENT TENSE, NO MODAL VERBS

Microsemantics is the set of sentence-level rules that make a page’s facts cheap for search and AI engines to extract. This article defines microsemantics, then sets out its nine rules: present indicative modality, subordinate text, first-word sequences, triple verbalisation, bold answer terms, numeric precision, post-noun instances, conditional clause placement, and one value per fact. The article then lists the formatting codes a brief uses to request each answer shape. Four further sections cover sentence perplexity with entity ambiguity, linear proximity, consensus callouts before data tables and groundedness. A checklist for editing a draft and the questions writers ask close the article.

What microsemantics is

Microsemantics is the discipline of word choice, word order and modality inside a sentence, applied so that each sentence yields one extractable fact.

Language models parse prose by word distance and probability. A sentence that states a fact in subject, predicate and object order costs little to extract. A sentence that hedges, delays or wraps the fact in a pronoun costs more, and qualitative re-ranking prefers the cheaper candidate. Microsemantics is where the ranking trade-off between responsiveness and cost is decided, one sentence at a time.

Present indicative modality and no modal verbs

Present indicative modality is the rule that every factual sentence states what is, with no modal verbs: will, should, may, might or could.

VersionSentence
SpeculativeIf you are building a topical map, you might want to start with a keyword export, and it should probably cover your main services.
IndicativeA topical map starts from a query network and covers every attribute of the central entity.

Subordinate text: the first sentence under a heading

Subordinate text is the first sentence under a heading, and it carries the highest passage extraction weight on the page.

Subordinate text answers the heading at once, with no introduction. A heading that asks what a query network is opens with “A query network is” and the definition. A heading that asks a price opens with the price. Every heading on this site follows the rule, which is why each section here opens with a definition.

First-word sequences

A first-word sequence is the opening three to five words of a paragraph, and it carries a concrete entity and attribute, never a pronoun or a transition.

VersionSentence
Pronoun paddingFurthermore, it is also worth remembering that these need to be reviewed before launch.
Entity firstContent briefs pass a microsemantic review before launch.

Triples and data-to-text verbalisation

A triple is a subject, predicate and object statement, and data-to-text verbalisation is the act of writing each fact so the triple reads straight off the sentence.

Search engines decompose documents into subject, predicate and object triples to populate knowledge graphs. Retrieval-augmented generation systems in AI search lift the same triples into answers. A sentence written as its triple needs no interpretation:

  • Unstructured: “PropXpro grew a lot after we rebuilt the site, roughly two hundred pages in total.”
  • Triple: ⟨PropXpro, has architecture size, 216 pages⟩
  • Verbalised: “PropXpro runs on a 216-page semantic content network.”

Bold answer terms

A bold answer term is the exact value a sentence answers with, set in bold so the eye and the parser find the answer, not the question.

Bold marks the value, never the query term or the entity name. “A new domain launches 15 to 20 quality nodes” bolds the answer. Bolding “quality nodes” alone repeats the question back to the reader.

Numeric precision

Numeric precision is the rule that every quantity appears as an exact number with its unit, never as a vague quantifier.

“Many pages” becomes “216 pages”. “A fast turnaround” becomes “five business days”. Each exact number is also one value sitewide: the same fact carries the same number on every page that states it.

Post-noun instances

A post-noun instance is a concrete example placed directly after the plural noun it belongs to.

“AI answer engines such as Google AI Overviews, ChatGPT and Perplexity cite extractable passages” names the instances where the class appears. “Many AI tools cite pages” names none, and the parser learns nothing about which engines.

Conditional clause placement

Conditional clause placement is the rule that the core fact comes first and the condition (if, when, because) comes second.

VersionSentence
Condition firstWhen a domain is older than 12 months, the launch batch grows.
Fact firstThe launch batch grows to 30 to 50 quality nodes when the domain is older than 12 months.

One value per fact

One value per fact is the sitewide rule that each entity, attribute and value pair carries a single value on every page.

Two pages that state different page counts for the same project contradict each other, and a contradiction loses the citation for both. Repeated facts reuse the exact same n-gram as well. Synonym rotation reads as variety to a person and as a second, unconfirmed fact to a parser.

Microsemantic formatting codes

Microsemantic formatting codes are the shorthand a content brief uses to request a specific answer shape for each heading.

CodeAnswer shapeLimit
FSFeatured snippet answer, value in bold within the first five words40 words or 320 characters
PAAPeople Also Ask answer, one declarative sentence30 words or 220 characters
DEFDefinition: [entity] is a [class] that [differentiator]35 words
LSTList preceded by a full definition sentence ending in a colon3 to 7 items
EXAExact definitive answer with every qualifier and metric30 words by default
ANCRAnchor text matching the target page’s title tag vocabulary2 to 5 words, 100 words from the previous link

Where these codes sit inside a brief is set out in how to write a content brief.

Sentence perplexity and entity ambiguity

Sentence perplexity measures the syntactic uncertainty a parser meets while it extracts a triple, and the retrieval cost article places the measure inside one formula. Passive voice, filler phrases and ambiguous modal verbs raise perplexity, and present-tense subject-first statements lower it.

Entity ambiguity measures unanchored references, such as the pronouns it and they or the phrase this platform. A pronoun forces the parser to resolve a reference before the parser stores the fact, and an explicit noun removes the step. Sentence perplexity and entity ambiguity sit beside URL complexity and contextual drift as the four factors of retrieval cost.

Weak sentenceRewritten sentenceExtracted triple
Semantic HTML might help crawlers understand a page.Semantic HTML names the role of each page part through its element.Semantic HTML, names, page part roles
It is important that a topical map should be built before writing.A topical map assigns every entity-attribute pair to a URL before writing begins.Topical map, assigns, pairs to URLs
A content brief could include the queries.A content brief states the URL, title tag, central entity and heading hierarchy.Content brief, states, page decisions

Each rewrite moves the entity to the first words, replaces the modal verb with an indicative verb and replaces a vague noun with a concrete one. The third column shows the triple that a parser stores.

Linear proximity between entity and attribute

Linear proximity sets the distance between a target entity and its attribute noun inside one sentence. A distance of fewer than 5 words gives transformer models, such as BERT, a stronger vector match between the two terms.

Term co-occurrence in one paragraph adds to the match. Dense retrieval matches a query to a document by vector proximity, so a paragraph that places related terms beside the entity matches queries that share none of the paragraph’s keywords. The sentence a topical map holds 15 fields per row keeps the entity and the attribute value within 4 words of each other.

The editing rule is short. Place the attribute noun directly after the entity, and place related terms for the entity in the same paragraph block.

Consensus callouts before data tables

A consensus callout is a summary passage that aligns key facts with the established consensus of authoritative sources. The framework places the callout immediately before a data table, so the table inherits the verified framing.

Data-to-text verbalisation converts the table rows into present-tense prose. The rows of a knowledge base, such as a country, its currency and its central bank, become two declarative sentences that name each entity once. Verbalisation and callout work as a pair: the callout states the agreed fact, and the table lists the detail.

Groundedness: the check no sentence rule replaces

Groundedness measures the verifiable alignment between claims in prose and established factual knowledge. Sentence rules lower extraction cost, and groundedness decides whether the extracted fact is correct.

Google research treats the same property as a quality signal. The Knowledge-Based Trust paper of 2015 scores a source by the share of its extracted facts that are correct, applied to 2.8 billion extracted facts from 119 million webpages. The LaMDA paper of 2022 sets groundedness as a target for dialog models, which consult an information retrieval system to back replies with known sources.

The editing step is verification. Check every number, currency, legal code and date against a primary source before publication, and delete any claim that lacks one.

Microsemantics before and after

Each rule changes a sentence in a way a reader can check. The pairs below show one rewrite per rule.

RuleBeforeAfter
Indicative modalitySemantic SEO can help to improve how a site is read.Semantic SEO lowers retrieval cost by giving each attribute one owner page.
First-word sequenceThere are several methods that measure topical coverage.Topical coverage equals published rows divided by total rows.
Numeric precisionThe map has lots of outer pages.The map has 46 outer nodes.
One value per factThe map has roughly 30 core rows or so.The map has 30 core rows.
Conditional placementA page ranks if the query network is covered completely.When the query network is covered completely, a page ranks across its phrasings.

Microsemantics by page predicate

The predicate of a row decides which microsemantic rules carry the most weight.

  • Know pages lean on the first-word sequence and the definitional sentence under each heading, so the engine lifts the answer whole.
  • Do pages lean on numeric precision and one value per fact, because a buyer compares offers on exact terms.
  • Learn pages lean on ordered lists and present indicative verbs, so each step reads as an instruction with no hedging.

The editing checklist below applies to all three. The weight of each rule shifts with the predicate.

A microsemantic editing checklist

A microsemantic edit checks every draft against eight questions, in this order:

  1. Does the first sentence under each heading answer the heading?
  2. Does any sentence contain will, should, may, might or could?
  3. Does any paragraph open with a pronoun or a transition word?
  4. Does each sentence state one fact as a readable triple?
  5. Is every answer value in bold, and nothing else?
  6. Is every quantity an exact number with its unit?
  7. Do instances follow their plural nouns, and conditions follow their facts?
  8. Does every fact match its value on every other page?

Sentence rules only pay off when every page of the network follows them, and the copywriting service applies them to every brief. → Microsemantic copywriting

Microsemantics questions

Does microsemantic writing read as robotic?

Microsemantic writing reads as plain and direct. Rhythm comes from varied sentence length and concrete examples, not from hedges and transitions.

Is “can” a modal verb under these rules?

“Can” states capability, not speculation, so the rule allows it sparingly. Will, should, may, might and could stay out of factual sentences.

How long is a sentence under microsemantic rules?

A sentence under microsemantic rules runs 22 words at most. Longer sentences split into two facts.

Microsemantics matter more for AI search, because answer engines lift individual sentences. A sentence that carries its own triple is quoted intact.

Microsemantics in the semantic SEO topical map

Microsemantics is the sentence rules 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 Fact-first writing, since sentence rules are applied by the semantic copywriting service.

→ How to Write a Content Brief: the brief that carries the rules to a writer

→ Semantic HTML: the markup counterpart at page level

Get my free Radar ScanFree Radar Scan