Semantic Extraction

The process of identifying named entities—such as people, organizations, locations, and products—within the DOM and linking them to unique identifiers in a knowledge graph.

• Uses Schema.org @type and @id attributes for deterministic matching
• Resolves ambiguous terms (e.g., 'Apple' the company vs. the fruit) via sameAs references
• Enables AI to build a graph of who, what, and where from unstructured text
• Critical for Knowledge Graph Injection and brand authority signals

The algorithmic analysis of heading elements (h1-h6) and sectioning content to construct a weighted document outline.

• Extracts parent-child relationships between content blocks
• Assigns relative importance scores to text based on heading level
• Uses HTML5 semantic elements (<article>, <section>, <aside>) to identify self-contained content units
• Directly feeds RAG chunking strategies by identifying natural semantic boundaries

The targeted parsing of machine-readable metadata embedded directly within HTML tags to surface high-confidence facts.

• Extracts alt text from <img> elements to understand visual media
• Reads datetime attributes from <time> for temporal reasoning
• Processes itemprop values in Microdata annotations
• Retrieves href and hreflang signals for link relationship mapping
• These attributes form the Accessible Name computation that AI agents rely on

The extraction and parsing of JSON-LD and Microdata blocks that provide explicit, machine-optimized entity definitions separate from visible content.

• Serves as a high-confidence extraction target, bypassing NLP ambiguity
• Defines complex relationships like worksFor, founder, and citation
• Enables Factual Grounding by providing verifiable sameAs links to Wikidata and other authorities
• Processed before raw text analysis due to its deterministic reliability

The practice of parsing the browser's computed Accessibility Tree—the parallel structure exposing roles, names, and states to assistive technologies—as a clean semantic API.

• Surfaces ARIA Landmarks (banner, main, navigation) for page region identification
• Exposes the computed Accessible Name for interactive controls
• Reveals the programmatic state of UI widgets (expanded, selected, disabled)
• Provides a higher-fidelity signal than raw DOM scraping for dynamic applications

The inference of semantic connections between extracted entities and content blocks based on structural proximity and explicit markup.

• Links a <figcaption> to its parent <figure> for media context
• Associates <th> headers with <td> data cells via scope and headers attributes in Data Tables
• Connects a review rating to its itemReviewed entity in Schema.org
• Builds the edge list that populates Enterprise Knowledge Graphs and vector embeddings

The foundational practice of using HTML elements according to their intrinsic, programmatically determined meaning rather than solely for visual presentation. Elements like <article>, <nav>, and <aside> explicitly communicate content roles to AI parsers, enabling accurate extraction of document structure. This contrasts with divitis, the anti-pattern of using only neutral <div> and <span> elements, which produces a flat, meaningless structure that hinders automated interpretation.

Discrete blocks of JSON-LD or Microdata embedded within an HTML document that provide explicit, machine-readable entity definitions. These serve as high-confidence extraction targets for AI-driven search engines, offering unambiguous statements about entities, their attributes, and relationships. Unlike inferred semantics from HTML elements, structured data islands deliver declarative facts that require no probabilistic interpretation by extraction algorithms.

A logical, nested structure of HTML heading elements (h1–h6) that defines the document outline, communicating relative importance and parent-child relationships of content sections. Proper hierarchy enables extraction models to build an accurate table of contents representation, identifying major topics and subtopics. A flat or broken hierarchy—such as skipping from h1 directly to h4—disrupts this structural signal and degrades extraction quality.

A parallel structure generated by the browser from the DOM that exposes semantic information, properties, and relationships of UI elements exclusively to assistive technologies and programmatic agents. AI crawlers traverse this tree to extract the computed accessible name, role, and state of each element. Content hidden from the accessibility tree via aria-hidden="true" or CSS display: none is effectively invisible to semantic extraction pipelines.

The principle that the meaning, state, and value of a user interface component can be reliably interpreted by software, including AI agents, through standardized, machine-readable properties. This concept underpins all semantic extraction: if a document's structure is deterministic, extraction models produce consistent, repeatable results. Non-deterministic patterns—such as JavaScript-injected content without proper ARIA annotations—introduce ambiguity that degrades extraction accuracy.

The formal groupings defined by the HTML specification—Flow, Phrasing, Sectioning, Metadata, Embedded, and Interactive—that dictate where an element can be used and what its semantic purpose is. Understanding these categories is essential for valid document structure. For example, sectioning content elements (<article>, <aside>, <nav>, <section>) each define a distinct scope within the document outline, directly influencing how extraction algorithms partition and classify content blocks.