Skip to content

On-Page Signals & Text Analysis Patents Reference

150+ patents covering phrase-based indexing, entity salience, BERT query understanding, featured snippets, and content quality signals.


Phrase-Based Indexing (Anna Patterson — 15 patents)

PatentDescription
US7536408Phrase-based indexing — "Documents with MORE RELATED phrases rank higher"
US9990421B1Phrase indexing continuation
US12013887B2Contextual link information gain

Core principle: Google indexes by MEANINGFUL PHRASES, not isolated keywords.

Good phrase: "affordable roofing contractor"
  → Natural, strong signal (common co-occurrence)

Bad phrase: "roofing contractor affordable services"
  → Unnatural, weak signal (forced construction)

Phrase frequency is measured against the full corpus to detect over-optimization. Pages with unnaturally high phrase repetition are flagged.


Phrase Architecture Distribution

Per Patterson patent + NLP principles — recommended distribution for a page:

LocationUsage
H11× exact primary phrase
First 100 words1× exact phrase + 1× near-variant
H2 headingsPhrase variants (NOT exact primary phrase repeats)
Body text1× per 200-300 words for primary phrase (natural distribution)
Image alt-text1× descriptive phrase (not stuffed)

Information Gain (WO2020081082)

Measures how much UNIQUE information a page provides compared to existing content on the same topic.

What scores high:

  • Original research and proprietary data
  • Unique expert perspectives not found elsewhere
  • Specific numeric values and measurements
  • First-hand experience accounts
  • Information not yet covered in the top-ranking pages

What scores low:

  • Restating information already in top-ranking pages
  • Generic overviews without new insights
  • AI-generated summaries of existing content
  • Thin rewrites of competitor articles

Entity Salience (US20150127617A1)

4 factors determine how important an entity is to a document:

FactorDescriptionSignal Strength
PositionEarlier in document = higher salienceHigh
FrequencyMore mentions = higher salienceMedium
CentralityMore connections to other entities = higher salienceHigh
Co-occurrenceAppears near related entities = stronger signalMedium

Implication: The central entity of your content should appear in the first paragraph, be mentioned multiple times naturally, and be surrounded by related entities that create a semantic context cluster.


BERT Query Understanding (US10452978B2, US20230334045A1)

BERT processes the FULL query context, not just individual keywords:

What BERT understands that keyword matching missed:

  • Prepositions ("for", "to", "without", "near") — changes intent completely
  • Pronoun references — "it", "they", "their" resolved in context
  • Query intent from surrounding context words
  • Negations — "shoes without heel" correctly parsed

Query-dependent ranking (US9218397B1): The weight of each ranking factor changes based on the query type. Information-heavy queries weight content depth. Navigational queries weight brand signals. Local queries weight proximity and prominence.


Content Quality Signals

PatentDescription
US8898296B2Boilerplate detection — DOM tree shape + text ratio analysis
US9959315B1Passage quality — standalone answer passages in each section
US8707459B2Content originality — original-to-copied ratio
US9767157B2N-gram quality — writing quality via n-gram pattern analysis
US6424983B1Grammar/spelling — lexicon finite state machine analysis
US20070067294A1Readability — reading level match to target audience
US8458207B2Heading structure — anchor/heading context analysis
WO2014209758A1Above-the-fold — content visibility, scroll distance from top

Boilerplate detection (US8898296B2): Google compares the DOM tree shape (structural pattern) of a page against the ratio of meaningful content to repeated template elements. Sites with high boilerplate-to-content ratios score lower.

Passage quality (US9959315B1): Each section (under a heading) should contain a standalone, extractable answer. This enables passage indexing and featured snippet selection.


TF-IDF and Relevance

PatentDescription
US20130346424A1Term frequency, inverse document frequency

Modern application: Raw TF-IDF has been largely replaced by BERT-era semantic matching, but topic modeling across the document corpus still applies. Pages with the right DISTRIBUTION of topically relevant terms (not just the target phrase) score better for relevance.


Pages eligible for featured snippets MUST have:

[ ] Direct, concise answer to the query (extractable paragraph)
[ ] Proper heading structure (H2/H3 over the answer)
[ ] Information in extractable format:
    - Paragraph: 40-60 words answering the question directly
    - List: Numbered or bulleted steps/items
    - Table: Data in rows and columns with headers
[ ] Query intent match (the page must be ABOUT this query, not just mention it)
[ ] Sufficient page authority to compete for the feature

On-Page Optimization Checklist (Patent-Grounded)

Phrase Architecture:
[ ] H1: 1x exact primary phrase
[ ] First paragraph: entity + primary phrase + near-variant
[ ] H2s: phrase variants (not primary phrase repeats)
[ ] Body: natural phrase distribution (1x per 200-300 words)
[ ] Alt text: descriptive, 1x relevant phrase

Entity Coverage:
[ ] Central entity in first paragraph (position = high salience)
[ ] Related entities present (co-occurrence signals)
[ ] Entity mentioned multiple times naturally (frequency)
[ ] Entity connects to related entities in context (centrality)

Information Gain:
[ ] Unique data, research, or perspective not in top-10
[ ] Specific numeric values and measurements used
[ ] Original examples or case studies included

Passage Quality:
[ ] Each H2 section answers a standalone question
[ ] Featured snippet format for key questions (paragraph/list/table)
[ ] Above-fold content addresses primary query immediately

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