NLWeb
NLWeb is an open project, started by Microsoft, that lets a website answer questions in plain language, for people and for AI agents.
NLWeb, short for Natural Language Web, is an open project that lets a website answer questions in plain language. Instead of clicking through menus, a visitor, or an AI agent, asks the site a question and gets an answer built from the site's own content.
Microsoft announced it in May 2025, and the code is on GitHub. It matters to marketers because it's one of the first serious attempts to make websites talk directly to AI agents, rather than wait for an agent to read and interpret their pages.
How it works
NLWeb builds on data many sites already publish, such as Schema.org markup and RSS feeds. It loads that data into a search index, and a language model of the site owner's choice uses it to answer questions.
Each NLWeb site also works as an MCP server. That means an AI agent can query the site in a standard way and get structured answers back, if the site owner allows it.
For example, a coffee roaster like Fernhill Coffee could set up NLWeb over its product feed. A shopper could ask the site "Which of your beans work best in a cafetière?" and get an answer from Fernhill's own catalogue. An agent comparing roasters for someone could ask the same question and get the same facts, straight from the source.
Should you care yet?
It's early. The project describes itself as a starting point for others to build on, and the AI platforms people use every day still learn about you mostly by reading your pages and other people's.
What NLWeb does show is where things may be heading: sites that answer agents directly. The groundwork is the same either way. Clean, accurate structured data and up-to-date feeds help a platform read you today, and they're what a project like NLWeb would use tomorrow.