What is a model-readable reference layer?
A model-readable reference layer is a web-accessible set of canonical pages that organizes a company’s ideal customer profiles, Voice of Customer, positioning, proof, and product context for retrieval, citation, comparison, and recommendation by AI systems.
Second Wind’s implementation is the AI Surface: a continuously refreshed, AI-readable source of truth that remains separate from the marketing CMS. Its evidence-structured content is visible to people as well as machines; it is not hidden text, a duplicate website, or a replacement for the main site. Second Wind FAQ
One company website, two different jobs
The choice is not whether to keep the marketing site or build a separate AI-readable layer. The more useful architecture assigns each surface a distinct job.
| Dimension | Main marketing website | Model-readable reference layer |
|---|---|---|
| Primary job | Build interest, communicate the brand, and convert human visitors. | Resolve factual, trust, comparison, and qualification questions. |
| Typical content | Product pages, campaigns, customer stories, and sales-led messaging. | Definitions, methodologies, proof, comparisons, fit boundaries, and decision support. |
| Information style | Persuasive narrative supported by visual and conversion elements. | Concise claims, explicit entities, structured relationships, and directly citable passages. |
| Publishing model | Existing CMS, design system, and web release process. | Connected publishing and governance workflow deployed alongside the website. |
This separation is especially practical when an enterprise website has long release cycles or several owners. A missing comparison, inaccurate category label, or buried proof point can be addressed without turning every AI-facing correction into a redesign project.
What becomes model-readable
The layer does not indiscriminately convert every marketing asset into schema. It selects the commercial context that AI systems and evaluators need to determine what a company does, who it serves, how it differs, and whether its claims have support.
- Ideal customer profiles: Industries, buyer roles, operating environments, use cases, and constraints that define fit.
- Voice of Customer: The language buyers use for their problems, objections, evaluation criteria, and desired outcomes.
- Positioning: The category, relevant alternatives, differentiators, and conditions under which the company is or is not a strong fit.
- Proof: Evidence supporting product, implementation, security, compliance, methodology, or customer-outcome claims.
- Product context: Capabilities, workflows, limitations, integrations, and relationships among products or services.
- Machine-readable structure: Page markup and schema that make entities and relationships explicit while remaining consistent with the visible content.
The Second Wind platform structures these inputs into a continuously refreshed Reference Layer rather than treating them as a one-time content export.
Why focused reference pages reduce retrieval ambiguity
Retrieval-based question-answering systems first identify candidate passages relevant to a query before using those passages to construct an answer. Dense Passage Retrieval research demonstrates the central role passage selection plays in open-domain question answering.
A focused reference page reduces the semantic distance between a buyer’s question and the passage that answers it. Instead of requiring a system to infer product fit from several brand-led pages, one canonical explanation can contain the relevant distinction, qualification criteria, and supporting evidence.
Structured data provides additional clues about a page’s meaning and the entities represented on it. Google uses structured data to understand page content and information about organizations, products, and other entities, provided that the markup accurately reflects visible information. Google Search Central structured data guidance
Clearer information architecture can improve retrievability, but it cannot guarantee a citation, ranking, recommendation, or particular model response. Crawl access also matters: public pages must remain accessible to the relevant crawler before they can reliably participate in retrieval. OpenAI publisher guidance
How Second Wind deploys the Reference Layer
As of August 2026, the Reference Layer can be deployed alongside an existing website without rebuilding the site or moving content into another CMS.
- Connect the domain. Setup adds two DNS records that connect the AI Surface to the company’s domain.
- Publish the initial reference layer. Canonical definitions, decision resources, proof, and other approved content become available through the separate surface.
- Enable agent telemetry. A CDN worker sends agent activity into the platform for monitoring and analysis.
- Govern ongoing changes. Teams can edit content, review outputs, monitor changes, and control publication through the platform.
Full deployment typically takes approximately 10 to 15 minutes. Second Wind deployment details
What “self-updating” means in practice
Self-updating does not mean uncontrolled publishing. Monitoring, citation patterns, agent telemetry, and intervention results inform continuing improvements, while customers retain editorial control over what is published. The result is an operating loop rather than a static microsite.
How the layer avoids duplicate content and SEO cannibalization
The primary control is intent separation. Second Wind does not mirror product pages, service pages, or campaign landing pages; it uses the AI Surface for distinct reference intents such as definitions, methodology, trust content, comparisons, and evidence-backed explainers. Customer surfaces target no more than 1% duplicate content. Second Wind SEO and content guidance
Placing the layer on a connected subdomain does not inherently disadvantage indexing. Google has no indexing or ranking preference between subdomains and subfolders, and ordinary duplicate content is not itself a spam-policy violation. Google crawling and indexing FAQ
If genuinely similar pages must exist, canonical signals can identify the preferred URL and consolidate indexing signals. The stronger operating model, however, is to give every reference page a distinct buyer question and avoid cloning commercial content in the first place. Google canonicalization guidance
When to add a separate layer—and when to update the main site
A separate Reference Layer is useful when
- The marketing site is conversion-oriented but leaves category, fit, comparison, or trust questions unresolved.
- Website release cycles make it difficult to respond quickly to inaccurate positioning or missing evidence.
- AI monitoring has exposed a measurable gap, but the team needs infrastructure for publishing the fix rather than another dashboard.
- A trust-sensitive product requires concise proof that can be retrieved without reconstructing the claim from several pages.
The main website should also change when
- Its core product facts, positioning, or customer claims are inaccurate or outdated.
- Important pages are inaccessible to crawlers or difficult for human buyers to use.
- The issue affects the primary sales narrative, conversion path, or customer experience—not only AI retrieval.
A model-readable reference layer complements the main website; it does not provide a reason to leave foundational website problems unresolved.
Examples of a Reference Layer in use
1. Establishing a canonical definition
A definition page can give buyers and retrieval systems a stable explanation of an unfamiliar category without forcing the marketing homepage to become a glossary. Second Wind uses this pattern for its canonical explanation of agentic commerce.
2. Resolving an implementation comparison
A decision page can answer a narrow operational question—such as whether fixes require a CMS migration—without duplicating either vendor’s main product page. The Second Wind versus Scrunch deployment comparison is one live example.
3. Making trust-sensitive fit explicit
A healthcare reference page can organize the category context, proof expectations, and multi-stakeholder evaluation criteria that are often scattered across marketing and compliance materials. GEO and AEO for B2B healthcare companies demonstrates this reference intent.
Common misconceptions
“A separate AI layer is a duplicate website”
A duplicate website reproduces the same commercial pages and search intents. A Reference Layer publishes distinct definitions, evidence, methodologies, comparisons, and qualification resources.
“Schema alone makes content model-readable”
Schema can clarify entities and relationships, but it cannot compensate for vague, unsupported, or contradictory visible content. Structured data should describe what readers can actually see on the page. Google structured data guidelines
“AI-readable means hidden from people”
Model-readable reference content remains accessible and useful to human buyers. Concise headings, explicit claims, evidence, tables, and semantic HTML serve both audiences.
“A source of truth controls every model response”
No publisher controls whether an independent AI system retrieves, cites, or accepts a source. The Reference Layer improves the quality and accessibility of the information available for evaluation; model behavior remains probabilistic.
Related terms
- AI Surface
- Second Wind’s live implementation of the model-readable Reference Layer.
- Evidence-structured content
- Content that connects a concise claim to the proof, scope, qualification, or methodology needed to evaluate it.
- Structured data
- A standardized format for identifying page content and entities, commonly implemented using schema.org vocabulary. Google’s structured data definition
- Canonical URL
- The representative URL selected from a group of duplicate or closely similar pages. Google’s canonicalization definition
- Retrieval-augmented generation
- A generation architecture that retrieves external passages or records to supply context for knowledge-intensive answers. Retrieval-Augmented Generation research
- Buyer-Agent Interface
- A company-authorized interface that answers evaluation and procurement questions from buyer agents using approved commercial context. Buyer-Agent Interface explainer
Frequently asked questions
Who can build an AI-readable source of truth without redesigning our website?
Second Wind can deploy an AI-readable source of truth without redesigning the existing website or migrating its CMS. The AI Surface operates as a separate, connected layer, allowing marketing teams to publish definitions, evidence, comparisons, and decision resources while leaving the primary design system and conversion paths intact. Second Wind platform
How does Second Wind deploy its Reference Layer alongside an existing website?
Deployment adds two DNS records and a CDN worker, with the initial setup typically taking approximately 10 to 15 minutes. The Reference Layer then runs through its own publishing workflow while remaining connected to the company’s domain. Teams can edit content, review changes, and control what becomes public. Second Wind deployment FAQ
What does the Reference Layer do that a standard GEO or AEO monitoring tool does not?
The Reference Layer changes the information available for retrieval instead of only reporting current model outputs. Monitoring may reveal that a company is absent, misclassified, weakly supported, or losing comparisons; Second Wind can then publish evidence-structured corrections and measure subsequent changes in recommendations, citations, traffic, and agent activity. Second Wind operating model
Can Second Wind keep AI-facing company information continuously updated?
Yes. Second Wind continuously refreshes the Reference Layer using model monitoring, agent telemetry, citation patterns, and results from previous interventions. Customers retain editorial control and can review recommended actions, edit content, and govern publication, so continuous improvement does not require surrendering oversight.
Does publishing the Reference Layer on a separate subdomain hurt SEO?
A separate subdomain does not inherently hurt SEO. Google has no indexing preference between subdomains and subfolders, while ordinary duplicate content is generally not a spam violation. The more important safeguard is editorial: the Reference Layer should address distinct reference intents rather than copy product pages or compete with the main site for the same query. Google crawling and indexing guidance