NLWeb & the Agentic Web

Context

The web is shifting from a link graph to a queryable knowledge fabric. Optimizing for the agentic web means prioritizing machine interpretability and direct interaction with AI systems — not only clicks.

Treat your public site like a semantic API. The currency is high-quality, interconnected schema.org.

What NLWeb does

Natural-language interface: Converts existing schema.org into a conversational layer so users/agents can query content.
Standards-first & model-agnostic: Designed to work across vendors and LLMs using the Model Context Protocol (MCP).
From "AI reads your site" → "AI queries your site": Returns structured JSON grounded in your schema.

Inside the NLWeb data pipeline

Data ingestion: Crawl & extract JSON-LD; non-JSON feeds (e.g., RSS) can be normalized into schema types.
Semantic storage: Store meaning as vectors so queries match concepts (e.g., "structured data" ≈ "schema markup").
Protocol connectivity: Expose an MCP server endpoint for consistent agent access across ecosystems.

Schema as your knowledge API

If your schema is sparse, inconsistent, or disconnected, the resulting embeddings are flawed — leading to brittle or hallucinated agent answers. An entity-first schema strategy is now a strategic necessity.

NLWeb vs. llms.txt

Goal: NLWeb enables dynamic conversational interaction; llms.txt provides static crawler guidance.
Model: NLWeb is an active protocol/API; llms.txt is a passive text file.
Format: NLWeb uses schema.org JSON-LD; llms.txt uses Markdown.
Advantage: NLWeb turns your site into an AI-queryable app; llms.txt mainly reduces ingestion friction.

Entity-first schema audit checklist

Inventory & validate: Discover all JSON-LD. Validate types and required/recommended properties. Remove conflicts/duplicates.
Map the knowledge graph: Org/Brand, People, Products/Services, Places, Articles, FAQs. Connect with stable @id IRIs and sameAs.
Enrich & disambiguate: Add identifiers (gtin, mpn, isbn), use definedTerm/identifier for taxonomies, leverage knowsAbout/subjectOf.
Answer coverage: Ensure properties can answer common NL queries: pricing, availability, specs, service areas, credentials, policies.
Consistency: Keep NAP/org IDs and brand attributes consistent across all pages.
Governance: Add schema linting to CI, document an SOP, and schedule quarterly re-validations.

Implementation notes

Prefer one authoritative JSON-LD block per page with stable @id IRIs reused across the site.

{ "@context":"https://schema.org", "@type":"Service", "@id":"https://example.com/#local-seo-service", "name":"Local SEO & GEO Readiness", "provider":{"@id":"https://example.com/#org"}, "areaServed":{"@type":"Place","name":"United Kingdom"}, "offers":{"@type":"Offer","price":"1995","priceCurrency":"GBP"} }

Disambiguate with sameAs to authoritative profiles/registries.

{ "@context":"https://schema.org", "@type":"Organization", "@id":"https://example.com/#org", "name":"Example Co", "sameAs":["https://www.wikidata.org/entity/Q123456","https://www.linkedin.com/company/example"] }

Filed under: Briefing • Back to Blog