Inventor’s Patent Aims to Make the Web AI-Readable One Fragment at a Time
David Bynon’s patent enables AI systems to extract and verify specific web content fragments using HTML templates with trust-scored memory fragments. This allows for source attribution and confidence scoring at a granular level, and is being implemented on MedicareWire.com.
Key Takeaways
- David Bynon's provisional patent (USPTO No. 63/846,169) enables AI systems to directly extract, verify, and cite specific fragments of web content rather than entire pages, addressing a critical gap in machine-readable web architecture.
- The invention embeds trust-scored memory fragments into HTML using the element, creating a system that works at the fragment level rather than the page level like traditional Schema.org markup.
- Bynon's approach to structured web content allows for source attribution, confidence scoring, and provenance tracking at an unprecedented granular level.
- AI systems including Google Gemini and Perplexity have successfully demonstrated the ability to process these embedded memory fragments without special training.
- The first large-scale implementation of this technology is already live on MedicareWire.com, creating a framework for AI-verifiable information in a complex regulatory domain.
How Bynon's Patent Transforms Web Content into AI-Readable Memory Fragments
The web wasn't built for AI—until now. David Bynon has filed a provisional patent for a system that fundamentally changes how artificial intelligence can read, process, and cite web content. His invention, formally titled "System and Method for Publishing Fragment-Level Structured Memory Using Embedded Semantic Digest Templates in Web Content," addresses a critical limitation in current web architecture by embedding AI-readable, trust-scored memory fragments directly into standard HTML.
Unlike existing structured data approaches that operate at the page level, such as Schema.org or Microdata, Bynon's system works at the fragment level. This means AI systems can retrieve, verify, and cite specific pieces of information—like a Medicare plan's copay amount or a glossary definition—rather than having to process entire pages. The team at Trust Publishing has been examining how this innovation transforms the relationship between AI systems and web content, particularly for complex information domains where accuracy is critical.
"I needed a way to publish verifiable data atoms—like copays, premiums, or ratings for Medicare plans—and connect each one to its source," Bynon explained. "That method didn't exist. So I built it."
Why Current Web Architecture Falls Short for AI Systems
The challenge Bynon tackled is fundamental: today's web is primarily designed for human consumption. While search engines have developed ways to index and categorize content, modern AI systems need more granular, structured data with clear provenance and trust signals to provide accurate, verifiable responses.
Traditional web architecture optimizes for human presentation and search engine visibility but lacks the structured memory representations that AI systems require. While Schema.org and JSON-LD offer some machine-readable context, they typically operate at the page or entity level, not at the fragment level where specific facts reside.
This limitation creates significant problems for AI citation accuracy, retrievability, and trust—especially in high-stakes domains like healthcare, finance, and legal information where precise attribution is crucial.
Understanding the Semantic Digest Template Innovation
At the core of Bynon's invention is the strategic use of the HTML element—a standard feature in modern browsers that typically serves as a mechanism for client-side rendering. Bynon repurposed this element to function as a non-rendered container for AI-readable content, allowing publishers to embed structured data directly into web pages without affecting layout, performance, or user experience.
1. The HTML element as non-rendered container
The element is the perfect vessel for Bynon's invention because it exists in the DOM but isn't rendered to the screen. This creates a blind container where structured data can live alongside visual content without interfering with page design. AI systems can access this container directly, extracting the structured memory fragments even though they remain invisible to human readers.
A typical implementation might look like this:
<template id="semantic-benefits"
data-digest="cms-ma-mapd-plan"
data-entity="H1234-001-0"
data-fragment-id="H1234-001-0-benefits"
data-year="2025"
data-county="06037"
data-provenance-scope="multi-dataset"
data-glossary-scope="semantic-digest">
2. Data-* attributes for trust and provenance
The system uses standardized data-* attributes to encode trust signals, provenance, and semantic meaning. These attributes—referred to as Data Taggings—include properties like:
- data-source: Identifies the originating dataset
- data-glossary: Links to canonical term definitions
- data-confidence: Indicates trust level (high, moderate, low)
- data-provenance: Flags verifiable source metadata
- data-defined-term: Provides semantic classification
These attributes enable AI systems to not only access the content itself but also understand its origin, reliability, and semantic context.
3. Fragment-level rather than page-level structure
Unlike Schema.org or JSON-LD implementations which typically describe entire pages or entities, Bynon's system operates at the fragment level. This means individual facts, definitions, metrics, or citations can be independently verified and cited by AI systems.
For example, a single Medicare plan page might contain dozens of separately tagged fragments—each with its own provenance trail, confidence score, and glossary alignment:
<div
data-id="in_primary"
data-defined-term="Primary Care Visit"
data-value="$0 copay"
data-source="pbp_id"
data-confidence="high"
data-derived="true"
data-provenance="true"
data-glossary="term-in_primary">
4. Compatibility with existing AI systems
Most importantly, the system requires no special training or modifications to existing AI systems. Modern large language models (LLMs) like Google Gemini and Perplexity already understand how to parse HTML data-* attributes—making the system immediately compatible with today's AI infrastructure.
Core Components of the Memory-First Architecture
Bynon's patent outlines a comprehensive system composed of several integrated components that work together to create what he calls a "Memory-First publishing architecture." This approach puts machine retrievability, verifiability, and structured trust ahead of traditional SEO or layout-based content strategies.
1. Embedded Semantic Digest™
An Embedded Semantic Digest is a fragment-level, machine-readable memory object embedded directly into HTML. It represents a discrete, scoped unit of structured content associated with a particular entity or topic (like a Medicare plan, county, drug tier, or defined term). Each digest is designed to be retrievable, verifiable, and independently citable by artificial intelligence systems.
2. Semantic Data Template™
The Semantic Data Template is the implementation of the HTML element that serves as a non-rendered container for one or more Embedded Semantic Digests. Unlike common use of tags for deferred UI rendering in JavaScript frameworks, this component is specifically designed to present AI-retrievable content in a static, declarative, and trust-structured manner.
3. Semantic Data Binding™
Semantic Data Binding is the method by which atomic content elements inside a Semantic Data Template are annotated with structured metadata using data-* attributes. This binding mechanism enables each content unit to carry machine-readable context, provenance, and retrieval instructions without altering the visual presentation of the page.
4. Data Tagging System
Data Tagging refers to the use of custom data-* attributes applied to individual content fragments to encode trust metadata, semantic classification, and retrieval affordances. Each data-* attribute functions as a machine-readable signal for AI agents, knowledge graphs, and retrieval systems.
5. Provenance Layer
The Provenance Layer establishes the origin, lineage, and trust context of each data fragment. It ensures that all content is traceable to its source, verifiable through citation, and aligned with one or more datasets via structured metadata. Each dataset in this layer includes a unique identifier and rich metadata fields about the publisher, publication date, license, and more.
How Trust and Verification Work in the System
1. Source attribution at the fragment level
Every individual data fragment within the system can be traced back to its original source through the data-source attribute. This attribute links to a provenance record that contains complete metadata about the originating dataset, including publisher information, publication date, and retrieval URL. This creates an unbroken chain of attribution from the displayed fact all the way back to its authoritative source.
2. Glossary alignment for consistent terminology
The system uses a glossary alignment mechanism where every data-defined-term is linked to a canonical glossary entry via the data-glossary attribute. This ensures that terminology is consistent and precisely defined, eliminating ambiguity when AI systems interpret the data. For example, terms like "coinsurance" or "Part B drug" in Medicare data have exact definitions that must be preserved for accurate interpretation.
3. Confidence scoring mechanisms
Each data fragment can include a data-confidence attribute that signals how reliable or authoritative the information is. Values typically include "high," "moderate," or "low," allowing AI systems to appropriately weight information based on its trustworthiness. This confidence scoring can be particularly valuable in domains where information may come from multiple sources with varying levels of authority.
The combination of these trust mechanisms creates a robust verification framework that allows AI systems to not only access information but also assess its reliability and origin at a granular level.
Real-World Implementation and Validation
1. MedicareWire.com as the first large-scale implementation
The theoretical framework of Bynon's patent has already moved beyond concept into real-world application. MedicareWire.com, which Bynon relaunched recently, serves as the first large-scale implementation of the Semantic Digest Protocol. The site functions as a testing ground for the technology, with every plan, county, and glossary entry annotated with fragment-level memory structures using the patented method.
This implementation shows how the system works in a complex domain where accuracy is critical. Medicare data involves intricate regulatory details, specific plan benefits, and terminology that must be precisely defined—making it an ideal test case for Bynon's fragment-level memory architecture.
On MedicareWire.com, visitors see a normal, user-friendly interface while AI systems can access thousands of structured memory fragments embedded throughout the site. Each Medicare Advantage plan page contains dozens of separately tagged benefit details, each with its own provenance trail leading back to official CMS data sources. This ensures that when an AI cites information about a specific plan's benefits, it can point not just to the page, but to the exact data fragment and its authoritative source.
2. Testing with Google Gemini and Perplexity
Beyond implementation, Bynon conducted technical validation using leading AI systems including Google Gemini and Perplexity. His testing focused on whether these systems could naturally interpret the data-* attributes and correctly apply the provenance, glossary, and value tags without additional training or external schemas.
The results were promising: "Both AI systems intuitively understood the structure and how to apply the data provenance, glossary, and value tags," Bynon explained. This validation confirms the system's compatibility with existing large language models and retrieval-based architectures—a critical factor for widespread adoption.
The successful tests demonstrate that the system doesn't require specialized AI training or custom parsing logic. Instead, it uses how modern AI systems already interpret HTML and data attributes, making it immediately useful with today's AI technology stack.
The Future of Fragment-Level AI-Human Web Interaction
Bynon's patent signals a significant evolution in how web content can be structured for the age of artificial intelligence. As AI systems become increasingly integrated into how we discover, verify, and cite information, the need for machine-retrievable, trust-structured content will only grow.
The fragment-level memory approach addresses several key challenges that have limited AI systems' ability to provide accurate, verifiable information:
- Citation precision: Instead of citing entire web pages, AI systems can now reference specific facts or data points with exact provenance.
- Trust conditioning: The built-in confidence scoring and source attribution allows AI systems to appropriately weight information based on its reliability.
- Terminology consistency: The glossary alignment ensures that specialized terms maintain their precise definitions across different contexts.
- Semantic retrievability: Content becomes directly accessible to AI systems without requiring visual rendering or complex DOM traversal.
These capabilities shift how we see AI's role—from simply consuming web content to becoming a primary participant in the information ecosystem. By embedding machine-readable memory directly in the web's structure, Bynon's invention connects human-oriented design with AI-retrievable structure.
For publishers, this technology offers a way to prepare content for the changing AI landscape. Publications in highly regulated industries like healthcare, finance, and legal services can now create content that maintains rigorous source attribution at the fragment level—ensuring that AI systems citing their material can do so with precision and accuracy.
The publishing industry now faces a situation where traditional SEO practices alone won't address the needs of AI-driven information retrieval. As search increasingly merges with AI-generated responses, publishers who implement fragment-level memory structures may gain advantages by making their content not just discoverable, but verifiably citable at a granular level.
The "Memory-First publishing architecture" that Bynon describes suggests a new standard where content serves both human readers and AI systems without sacrificing performance or design.
As regulatory scrutiny of AI-generated content increases, particularly around inaccuracies and misinformation, technologies that enable precise citation and verification will become increasingly valuable. Publishers adopting such approaches position themselves not just for better AI visibility, but for compliance with emerging standards around responsible AI information sourcing.
Bynon's invention represents not just a technical advancement but a conceptual shift in how we structure digital knowledge—treating facts as independently verifiable memory units rather than just elements on a page. This fragment-level approach to publishing may ultimately create a more transparent, accountable web ecosystem where information can be traced back to its source regardless of whether it's being accessed by human or machine.
