A Deep Dive Into NLWeb

For as long as there have been Web sites, there have been search boxes to dig through the contents. Those search boxes tend to be hit or miss when it comes to accuracy, and they only have two methods of sorting: by date and by relevance.

Microsoft is aiming to make the search process more natural with the introduction of NLWeb, or natural language Web. NLWeb enables Web sites to easily add a conversational interface to the site, essentially turning a Web site into an AI app where users can search for content using natural language rather than keywords.

"The idea behind NLWeb is it is a way for anyone who has a Web site or an API already to very easily make their website or their API an agentic application," Microsoft CTO Kevin Scott said during his keynote at the recently-held Build 2025 conference. "You really can think about it a little bit like HTML for the agentic Web."

NLWeb builds on existing web infrastructure while using modern AI technologies. So NLWeb doesn't compete with or replace other protocols; rather, it builds on top of them. The new protocol uses existing structured data formats like RSS, Atom and JSON Feed, which Web sites already use, combined with LLM-powered tools to create natural language interfaces usable by both humans and AI agents.  This means publishers don't need to rebuild their content infrastructure completely.

NLWeb includes tools for adding this structured data to vector databases, a specialized database designed to store, index and search high-dimensional vector embeddings -- numerical representations of data like text, images, audio or video. These vectors are often generated by machine learning models to search and retrieve semantic models.

LLMs then augment this stored data with added external knowledge and context. For example, if a user makes a query about restaurants, the system automatically adds on things like reviews, local information, and other related information to provide a full, comprehensive response rather than just providing a link.

By offering search via a natural language interface rather than searching by keywords, visitors can ask questions in plain English and receive conversational responses. AI systems can programmatically access and query the site's information through the Model Context Protocol (MCP) framework.

This approach allows site owners to turn their existing Web site to a generative AI-drive chatbot without requiring extensive technical overhauls. It makes AI-powered search and interaction as accessible as creating a basic webpage was in the early days of the internet.

Examples of NLWeb in Action
NLWeb is intended to offer a far more natural experience. Instead of searching by keywords and getting links or static Web pages, the user searches with a natural, conversational query and the answer is dynamically generated from scratch.

Here are some real-world usage examples of Microsoft NLWeb in action:

Generating recipes. A user may query a recipe site with “Create an authentic Tuscan recipe using chicken as the main ingredient.” Or you can get very specific and request “create a spicy Sichuan recipe” and then we'll use those specific ingredients.

E‑commerce and Retail. A user may ask, "show me appropriate attire for winter in Calgary" or, "suggest appropriate gifts for new parents of a baby girl." And instead of clicking check boxes on a search page, you simply ask, "show me business casual dress shoes for under $100 in a size 9."

Another example is product comparisons. You can do a query such as, "What smartphones released in the past five years offer 500 gigabytes of storage?"    

Event and Experiences. NLWeb can make real life experiences more enjoyable. You can build a query such as, "Where is a good place outside of Boston to go on a first date that doesn't involve alcohol?"

Travel and Local Discovery. Going on a vacation and want to avoid the tourist traps? Query a travel site with, "Where can a couple with no children go in Hawaii that is not overrun with tourists?"

Currently, Microsoft NLWeb is not available as a public API or developer tool. It's a research project, primarily used to advance natural language understanding over Web tables and is part of broader work on semantic parsing, Web-scale question answering, and the WebQA/NQ datasets. But many Microsoft products have started out as research projects and were eventually taken to market. Hopefully this will be the case as well.

Posted by Andy Patrizio on 06/23/2025