NLWeb: Natural Language Interface for Your Website
TL;DR
NLWeb is an emerging protocol that lets AI agents query your website using natural language. Instead of crawling HTML, AI can ask your site questions and get structured answers.
Last updated: 2026-03-09
What Is NLWeb?#
NLWeb (Natural Language Web) is an open protocol developed by Microsoft that gives AI agents a way to talk to your website directly. Instead of parsing HTML, scraping text, and guessing at meaning, an AI agent can send a natural-language question to your NLWeb endpoint and receive a clean, structured answer.
Traditional web crawling is like reading a book by scanning every page. NLWeb is like asking the author a question and getting a direct answer. The protocol defines a standard endpoint at
/.well-known/nlweb where AI agents can send queries and receive JSON responses.
Microsoft released the specification in early 2025, and it represents a significant shift in how websites communicate with AI. Rather than waiting for crawlers to discover and interpret your content, NLWeb lets you serve answers on demand.How It Works#
NLWeb uses a well-known URL pattern that AI agents can discover automatically. Here is the basic flow.
First, an AI agent checks whether your site supports NLWeb by looking for
/.well-known/nlweb. If the endpoint exists, the agent knows it can send natural-language queries. Second, the agent sends an HTTP request with a question in plain English (or any supported language). Third, your NLWeb endpoint processes the question, searches your content, and returns a structured JSON response with the answer, source URLs, and confidence scores.
Behind the scenes, your NLWeb implementation typically uses a vector database or search index to match questions against your content. The protocol specifies the request and response format, but the retrieval method is up to you.Example NLWeb request and response
// Request: GET /.well-known/nlweb?q=What+are+your+shipping+rates
// Response:
{
"query": "What are your shipping rates?",
"answers": [
{
"text": "We offer free standard shipping on orders over $50. Standard shipping is $5.99 and express shipping is $12.99.",
"confidence": 0.95,
"source": "https://example.com/shipping",
"title": "Shipping Policy"
}
],
"site": "example.com",
"protocol_version": "0.1"
}json
Why NLWeb Matters for the Agentic Web#
The web is shifting from a place where humans browse pages to a place where AI agents act on behalf of humans. In this new world, your website needs to communicate with machines as effectively as it communicates with people.
NLWeb matters because it bridges the gap between static web content and dynamic AI interaction. When a user asks an AI assistant to find a product, compare prices, or check business hours, the assistant can query your NLWeb endpoint directly instead of guessing from scraped HTML.
This means faster answers, more accurate citations, and a better experience for users who interact with your brand through AI. Sites that adopt NLWeb early will have a meaningful advantage as AI agents become the primary way people discover and interact with businesses online.
Current Adoption#
NLWeb is in its early stages. Microsoft published the spec and reference implementation as open source. Adoption is growing among forward-thinking companies, especially in e-commerce, publishing, and SaaS.
As of early 2026, most major AI platforms are aware of the protocol and some have started querying NLWeb endpoints when they are available. Early adopters report that having an NLWeb endpoint increases the accuracy of AI-generated answers about their products and services.
The AgentReady™ scanner checks for the presence and proper configuration of NLWeb endpoints as part of the AI Protocols factor. Even basic implementations can improve your score.
- Microsoft published the spec and reference implementation as open source
- Growing adoption in e-commerce, publishing, and SaaS sectors
- AI platforms are beginning to query NLWeb endpoints when available
- Early adopters see improved accuracy in AI-generated answers about their brand
Implementation Overview#
Setting up NLWeb requires a server-side endpoint that can process natural-language queries. The simplest approach uses a pre-built library or framework.
For most sites, the implementation involves three steps. First, index your content into a searchable format (a vector database or full-text search engine). Second, create an API endpoint at
/.well-known/nlweb that accepts query parameters. Third, wire up the endpoint to search your index and return structured JSON responses.
For detailed, step-by-step instructions, see our NLWeb implementation guide. If you are evaluating whether NLWeb is right for your site, focus on whether your content is frequently queried by AI assistants and whether you have the technical resources to maintain an endpoint.Related Pages
Frequently Asked Questions
Do I need NLWeb if I already have llms.txt?
They serve different purposes. llms.txt provides a static summary of your site. NLWeb provides dynamic, real-time answers to specific questions. For maximum AI readiness, use both.
Is NLWeb hard to implement?
It depends on your technical setup. If you have a developer and a content database, a basic implementation can be done in a day. See our step-by-step guide at /docs/fixes/add-nlweb.
Was this page helpful?