Skip to content

Semantic Relevance Keyword Scorer

Source: Bill Slawski's analysis of Google's NLP and Semantic Understanding patents, SEO by the Sea Foundation: Post-Hummingbird (2013) — Google's shift from keyword matching to meaning-based indexing

Why Frequency Metrics Are Wrong

Traditional keyword optimization asks: "How many times does this keyword appear?"

Google's semantic understanding system asks: "Does the use of this keyword signal genuine topical expertise?"

A keyword used once in a precisely correct semantic context contributes more ranking signal than the same keyword repeated 10 times in shallow or off-topic contexts. This is the central insight from the patent methodology — semantic weight over frequency.

The 6 Scoring Principles

Principle 1: Semantic Weight Over Frequency

The model: Google assigns semantic weight based on the topical relationship strength between the keyword and its surrounding context — not based on count.

A keyword that appears in a sentence that defines, contextualizes, and extends the concept (high semantic weight per occurrence) outperforms a keyword inserted into a sentence that doesn't develop the concept (low semantic weight, regardless of frequency).

Scoring factor: Does each keyword appearance contribute semantic signal, or is it a low-information insertion?


Principle 2: Disambiguation (Non-Negotiable)

The problem: Polysemous words have multiple meanings. Google's system must determine which meaning applies to your content before it can associate your content with the correct queries.

Common polysemous terms that require disambiguation:

  • bank — financial institution / river bank
  • python — programming language / snake
  • mercury — element / planet / car brand / Roman god
  • apple — fruit / technology company
  • cloud — weather phenomenon / cloud computing
  • cookies — food / browser cookies / tracking cookies
  • crawling — physical movement / web crawling
  • anchor — nautical / text anchor / anchor text
  • lead — to guide / a prospect / the metal
  • spring — season / spring (mechanical) / to jump

Disambiguation method:

  • Add type assertion: "Python programming language" vs. just "Python"
  • Add co-occurring terms that only appear with the correct meaning: "Flask," "Django," "pip" → programming language
  • Use explicit category context: "in the context of SEO..." before using ambiguous terms

Scoring factor: Are all polysemous terms clearly anchored to their intended meaning by surrounding context?


Principle 3: Paraphrase Equivalence

The mechanism: Google's indexing system links multiple phrasings to the same concept. Content that expresses a concept through multiple natural phrasings gets indexed under all of them.

What this means in practice: If you only write "search engine optimization," you miss users searching for:

  • "SEO"
  • "organic search optimization"
  • "website optimization for search"
  • "ranking in Google"
  • "improve search rankings"

These are all semantically equivalent — Google knows they refer to the same concept. Content that uses multiple natural phrasings of the same concept is indexed more broadly than content that uses only one phrasing.

The rule: For every primary concept in your content, use at least 3 natural phrasings without forcing them.

Natural paraphrase example:

  • "keyword research" → "finding search terms" → "discovering what people search for" → "search query analysis"

Forced (bad) paraphrase:

  • "keyword research" → "keyword researching" → "keywords research process" (these aren't natural phrasings — they're manufactured variations)

Principle 4: Query Rewriting Awareness

Google's query rewriting behavior: When a user searches, Google internally rewrites the query through:

  • Specialization: Adding specificity ("coffee" → "arabica coffee beans")
  • Generalization: Broadening scope ("Miami plumber" → "plumbers in south florida")
  • Reformulation: Different phrasing for same intent ("how to fix a leaky faucet" → "leaky faucet repair")
  • Stemming: Word form variants ("optimize" → "optimizing" → "optimization")

Your content must be semantically aligned to the meaning behind the query, not just the exact string.

Scoring factor: If Google rewrote your target query 5 different ways (more specific, more general, different phrasing, different form), would your content still be the best match for all 5 rewrites?


Principle 5: Topical Depth

The signal: Shallow keyword usage signals thin content. Deep semantic fields signal genuine expertise.

A "topically deep" document doesn't just cover the surface-level definition of a topic — it extends into:

  • Subtopics (related sub-concepts)
  • Related concepts (neighboring topics that experts associate with this topic)
  • Prerequisite knowledge (what must be true for this topic to make sense)
  • Downstream applications (what people do with this knowledge)
  • Edge cases and exceptions (where the normal rules don't apply)

Measurement: Count the number of distinct conceptual territory areas the content covers. Shallow content covers 1-3 areas. Deep content covers 7-10+ areas.

Example — shallow "SEO" content: Covers: what SEO is, on-page optimization, backlinks.

Example — deep "SEO" content: Covers: what SEO is, on-page optimization, technical SEO, entity optimization, semantic search, user behavior signals, E-E-A-T, link building strategy, local SEO considerations, content freshness, crawl budget, indexation strategy.


Principle 6: Opinion & Sentiment Context

The patent: Google's opinion analysis system extracts opinions from content and attributes them to specific entities. This affects how keyword usage is interpreted.

A keyword appearing in a factual context carries different semantic weight than the same keyword in an opinion context.

Examples:

  • "The Python library reduces processing time by 40%" — factual claim, high semantic weight
  • "I think Python is probably easier than JavaScript" — hedged opinion, lower semantic weight
  • "Python is the worst choice for real-time applications" — strong opinion, Google attributes this stance to the author

Implications for keyword scoring:

  • Factual, authoritative uses of keywords score higher than hedged opinions
  • Expert opinions (attributed to a credentialed source) score higher than unattributed opinions
  • Negative claims about entities can affect how Google associates your content with those entities

Scoring System: 0-100 Per Keyword

For each target keyword, score it across all 6 principles:

PrincipleWeightScore (0-10)Weighted Score
Semantic Weight20%/10/20
Disambiguation20%/10/20
Paraphrase Equivalence15%/10/15
Query Rewriting Alignment20%/10/20
Topical Depth15%/10/15
Opinion/Sentiment Context10%/10/10
TOTAL/100

Score Interpretation

ScoreVerdictAction
80-100Strong semantic alignmentMaintain current usage
60-79Good alignment, minor gapsAdd paraphrase variants, check disambiguation
40-59Shallow usageRewrite with deeper semantic development
Below 40Keyword mismatch or stuffing riskEither remove or fully rewrite keyword strategy

Semantic Scoring Workflow

Step 1: List all target keywords for the content piece Step 2: For each keyword: read every sentence it appears in Step 3: Score each keyword against the 6 principles Step 4: Flag keywords scoring below 60 Step 5: For each flagged keyword: diagnose which principle is failing Step 6: Rewrite based on diagnosis — don't just insert more keywords

Common Failure Patterns

Pattern: Exact-match insertion Keyword appears in sentences that don't develop the concept. Sentence could be deleted without losing meaning. Fix: Add a sentence that defines, contextualizes, or extends the keyword before/after each appearance.

Pattern: Missing disambiguators Polysemous keyword appears without context anchors. Fix: Add type assertion or co-occurring context words on first mention.

Pattern: Single phrasing Only one phrasing of the core concept appears throughout. Fix: Add 2-3 natural synonymous phrasings distributed through the content.

Pattern: Surface-level topical coverage Keyword appears but surrounding content only covers the definition, not subtopics and applications. Fix: Expand to cover at least 5-7 distinct aspects of the topic.

Pattern: Hedged factual claims "X might help..." "Some experts say X could..." "You may want to consider X..." Fix: Where you can verify claims, make them factual. Where you can't, attribute the opinion to a named source.

Grounded in Bill Slawski's SEO by the Sea patent research