Kevin Sullivan

Case study · Entity architecture

Building and testing an industry knowledge graph for search visibility

Turf Network began as an attempt to organize a fragmented industry.

I built the platform from the ground up as an industry directory, educational resource and experimental environment for SEO, entity architecture, structured information and AI-search research.

2023 – present

Industry organizations in the directory
2,500+
Entity-focused pages
142
Custom structured-data templates
100+
AI-search platforms observed
5

The most important result was not the number of pages created. It was learning when a content strategy I had designed was no longer producing the information clarity I wanted, then rebuilding the system around a different model.

01

At a glance

Project
Turf Network
Industry
Artificial turf
Role
Founder, strategist and builder
Timeline
2023 to present
Project type
Owned research and publishing platform
Platform
WordPress and WooCommerce

What I own:

  • Product and platform strategy
  • Information architecture
  • Entity modeling
  • Website implementation
  • Directory development
  • SEO strategy
  • Content systems
  • Structured data
  • Internal linking
  • Analytics
  • AI-search research
  • Publishing and iteration

02Context

The problem

The artificial-turf industry contains thousands of organizations, products, applications, materials, technologies and local markets.

Information about those entities is often fragmented across:

  • Manufacturer websites
  • Installer websites
  • Supplier websites
  • Product documentation
  • Local business profiles
  • Trade publications
  • Marketing content
  • Technical resources
  • Distributor and dealer pages

The same concept may be described using inconsistent terminology. Companies can operate as manufacturers, suppliers, distributors, dealers or installers, sometimes combining several roles.

My goal was to explore whether that fragmented information could be organized into a clearer industry resource. The initial challenge was broad:

How do you build a useful information system for an industry when there is no existing audience, content library, data model or established authority?

03Execution

Building the platform from zero

I built Turf Network from concept through implementation. The initial platform combined:

  • A national organization directory
  • Educational resources
  • Product and application information
  • Location-based discovery
  • Lead-generation assets
  • Video and visual content
  • Payment functionality
  • Analytics and performance monitoring

I assembled a directory containing more than 2,500 industry organizations using publicly available business information.

The directory created industry breadth. It also introduced questions that later became central to the project:

  1. 01What type of organization is this?
  2. 02Which locations does it serve?
  3. 03What products or services does it offer?
  4. 04Which manufacturers, suppliers or dealers is it connected to?
  5. 05How should those relationships appear in the site architecture?
  6. 06What information belongs on an organization page?
  7. 07What evidence supports each relationship?
Representative view of the Turf Network organization directory, with filters for category, location and application, and a table of selected organizations showing their category and location.
Representative viewPublic listing fields only.

04Hypothesis

The original hypothesis

My original strategy relied on a relatively small number of core entity pages supported by large topical content clusters. The assumption was straightforward:

If the platform published the most comprehensive resources in the industry, that depth would help establish topical authority and improve search visibility.

Some supporting articles grew to approximately 6,000 to 10,000 words. The content addressed subjects such as:

  • Artificial-turf applications
  • Products and materials
  • Installation
  • Maintenance
  • Components
  • Performance features
  • Technical specifications
  • Buyer questions
  • Industry terminology

This approach created depth, but it also created a new problem. The articles contained many distinct concepts without always giving those concepts a clear, persistent place in the information architecture.

A single long article might discuss products, applications, components, features, maintenance requirements and installation methods. The information was present, but the relationships among those entities were often buried inside prose.

Representative long-form guide, “Artificial Turf Infill Materials,” annotated to show that one page carried a component (infill), performance concepts (drainage, heat and odor control), named materials and applications such as pet turf and athletic fields.
Editorial reconstructionOne page carried component, product, performance and application concepts.View full-size image

05Revision

When the evidence challenged the strategy

Publishing more comprehensive content did not automatically create the information structure I wanted. The site had depth, but important concepts were still difficult to isolate, connect, reuse and maintain.

I concluded that several assumptions behind the original strategy needed to change:

  • A long page is not necessarily a clear page.
  • Mentioning an entity is not the same as defining it.
  • Topic coverage does not automatically explain relationships.
  • More words can increase ambiguity when several entities share one page.
  • Internal links are more useful when both ends of the relationship have clear roles.
  • Search visibility alone does not prove that the information model is working.

Rather than continue expanding the content clusters, I changed the strategy. I removed oversized content and shifted investment toward:

  • Explicit entity pages
  • Smaller entity clusters
  • Clearer page purposes
  • More intentional internal linking
  • Structured relationships
  • Reusable page frameworks
  • More consistent terminology

This produced the principle that now guides much of my work:

Content depth is not the same thing as information clarity.
Before-and-after content architecture. Before: topical hubs for components, applications and buying guides, each made of long, cross-linked articles. After: defined entities and relationships—an organization serves a location and offers a product; the product is used for an application and has a feature enabled by a component.
SchematicBroad topics became explicit, reusable relationships.View full-size image

06Decision

Redesigning the site around entities

The revised architecture treats important concepts as entities that deserve clear definitions and relationships. The current model includes entity types such as:

Organizations

  • Manufacturers
  • Suppliers
  • Distributors
  • Dealers
  • Installers
  • Industry service providers

Products and systems

  • Turf products
  • Product lines
  • Installation systems
  • Accessories
  • Supporting materials

Applications

  • Residential lawns
  • Pet areas
  • Putting greens
  • Playgrounds
  • Sports fields
  • Commercial spaces
  • Rooftops and specialty installations

Features and components

  • Fiber shapes
  • Face weight
  • Pile height
  • Backing systems
  • Infill
  • Drainage
  • Cooling technologies
  • Antimicrobial features

Geographic entities

  • States
  • Metropolitan areas
  • Cities
  • Service areas
  • Organization locations

The site then connects these entities through navigation, contextual links, taxonomies and structured data.

Entity model with example relationships: an organization serves a location and has a role such as installer or supplier; a product is made of a component, such as fibers, which is equipped with a feature, such as antimicrobial protection, suited for an application, such as pet areas.
SchematicPrimary entity types and example relationships.

07Execution

Creating reusable page standards

Moving toward entity pages required more than changing URLs. I developed repeatable standards for new page types. Depending on the entity, those standards define:

  • The page’s primary intent
  • The entity being represented
  • Required supporting information
  • Related entities
  • Internal-link destinations
  • Structured-data requirements
  • Source and validation expectations
  • Publication checks
  • Post-launch monitoring

I also created more than 100 custom schema templates using Rank Math to apply structured data across recurring page types.

The templates were intended to improve consistency and make page relationships more explicit. They were not treated as a guarantee of rankings or AI-search inclusion.

Illustrative field map for an organization-profile template—display name, business role, service region, related offering, use case, performance trait and system material—beside a rendered example profile for “Example Turf Company,” an installer and supplier in Texas.
Illustrative dataOne structured field pattern can populate many profiles.View full-size image

09Research

Turf Network as an AI-search research environment

As generative search became more prominent, Turf Network provided an environment for studying how search systems interpreted structured industry information. I used the project to investigate questions such as:

  • Which organizations are included in AI-generated recommendations?
  • Which sources are cited?
  • How are manufacturers, suppliers, dealers and installers classified?
  • Do AI systems distinguish products from applications and features?
  • Which pages are retrieved for complex industry questions?
  • How does inclusion differ across query types?
  • What happens when entity definitions or relationships become clearer?
  • Where does external validation appear to matter?

Research included observations across:

  • Google AI Overviews
  • ChatGPT
  • Claude
  • Gemini
  • Perplexity

The project contributed to the development of my broader AI Visibility Diagnostic methodology. It helped separate three related questions:

  1. 01IdentityCan the system understand what the entity is?
  2. 02ArchitectureCan it understand how that entity relates to other information?
  3. 03ValidationCan it corroborate the claims being made?
The three diagnostic lenses applied to the Turf Network entity model. Identity: define who the organization is, by role and location. Architecture: connect product, component, feature and application. Validation: check the published evidence, then retest.
SchematicClear identity, connected structure and verifiable evidence.

10Observation

Observed search visibility

Turf Network pages have appeared in traditional organic results and Google AI Overviews for relevant industry searches. These observations demonstrate that the platform’s information has been discovered and used by search systems.

They do not prove that a specific architecture, schema implementation or content change caused inclusion. AI-search results change over time and can vary based on:

  • Query wording
  • Location
  • Personalization
  • Platform updates
  • Available sources
  • Indexing state
  • Competing evidence

For that reason, screenshots are presented as point-in-time observations rather than permanent outcomes or proven ranking relationships.

Google AI Overview for the query “ai visibility for artificial turf,” whose response describes Turf Network as an industry resource and cites a turfnetwork.org page titled “TURF NETWORK // AI Knowledge Graph of the Turf Industry.”
Observed responseQuery: “ai visibility for artificial turf.” The capture date was not recorded.View full-size image
Google organic results for “artificial grass system features,” with a turfnetwork.org page, “Turf Features: Drainage, UV Protection, Antimicrobial & More,” listed on the first page among other artificial-grass sites.
Observed resultQuery: “artificial grass system features.” The capture date was not recorded.View full-size image

11Output

What the project produced

An owned industry platform

A functioning platform that combines directory records, educational information, entity pages, structured data, internal relationships and commercial functionality.

A dataset of more than 2,500 organizations

The directory provides a base for investigating organization types, geographic coverage, market relationships and industry structure.

142 entity-focused pages

The revised architecture gives important industry concepts clearer definitions and dedicated places within the site. This count is a point-in-time measure because the platform continues to evolve.

A repeatable publishing system

New page types can follow documented standards for intent, relationships, internal links, structured data, publication and monitoring.

A research environment

Because I control the platform, I can change its structure, document the change and observe what happens afterward. That makes Turf Network different from a client engagement, where experimentation may be constrained by scope, approvals or business risk.

12Limitations

Research limitations

Turf Network is an active project, not a controlled laboratory. Several limitations affect what can be concluded:

  • Search systems change continuously.
  • AI responses can vary between tests.
  • The platform has changed substantially over time.
  • Multiple changes may occur within the same period.
  • Inclusion does not prove causation.
  • Third-party traffic estimates are not first-party analytics.
  • A single industry cannot represent all search environments.
  • Some entity relationships remain incomplete or provisional.

The purpose of the project is not to claim universal answers. It is to create a disciplined environment for forming hypotheses, testing implementations, recording observations and improving the next version of the system.

13

What I learned

Content depth is not the same thing as information clarity

A comprehensive article can still obscure the individual entities and relationships a user or retrieval system needs.

Being willing to remove work is part of strategy

I had invested significant time in the original content model. When the results and structure did not support the original hypothesis, I removed content and changed direction instead of defending the existing approach.

Entity pages require relationships

Creating one page per concept is not enough. The value comes from defining how organizations, products, features, applications, components and locations connect.

Structure and validation solve different problems

Clear architecture can make information easier to interpret. It does not replace reviews, citations, technical documentation, brand mentions or other forms of external validation.

A system needs operating standards

Scalable publishing depends on rules for page intent, evidence, relationships, internal linking, structured data and quality control. Without those standards, the architecture will gradually become inconsistent.

14

What this project demonstrates

Turf Network shows my ability to:

  • Take a search product from concept to implementation.
  • Build and manage a large industry dataset.
  • Design entity-focused information architecture.
  • Create scalable page and publishing systems.
  • Connect SEO, content, structured data and internal linking.
  • Conduct hands-on AI-search research.
  • Distinguish observations from causal claims.
  • Document limitations and uncertainty.
  • Change strategy when evidence challenges the original hypothesis.
  • Turn one-off discoveries into reusable frameworks and standards.

The project is still evolving. That is part of its value. Turf Network is both a working industry resource and a record of how my thinking about content, entities, architecture and AI retrieval has changed through implementation.

Explore additional work across search strategy and systems.