What is NLWeb? Crystal Carter explains how the Natural Language Web uses the ASK protocol and your structured data to prepare your site for the agentic web.

Click on the whiteboard image above to open a high-resolution version!

Hi, my name is Crystal Carter. I am the Head of AI Search and SEO Communications at Wix and Wix Studio. And today I'm going to talk to you about something new that can make your website better equipped for the agentic web and make use of your structured data, your heading tags, and lots of other great parts of your website.

Today, I'm going to be answering the question, "WTF is NLWeb?" And that's what we're going to talk about today. So if you haven't heard about NLWeb, it stands for Natural Language Web. It is an open-source project from the team over at Microsoft. And it was originally launched back in the spring of 2025.

And in the last few months, it's grown a lot of momentum, and there are some great things that you should know in order to make use of this great technology. 

So let's get into it. Okay, so NLWeb, it's not exactly a protocol. It kind of uses protocols. It uses something called the ASK protocol. But it's basically a sort of system of ways that you can connect with a website, that come together to make sure that your website can be spoken to using natural language. That's what the NL in NLWeb stands for.

So what happens in practice is that an agent, like Roger, he's not exactly an agent, but he's a robot, and so Roger here is standing in for agent. So let's say an agent or a person will come to your website and they'll ask a long-winded question.

They might ask: I am somebody who does tennis every week, and I'm looking for a shoe that will work well for me. And I wear this size. And I play on a sand court, and I do this and that and that. So they might ask something really, really complicated about that.

Now with a classic search and with a classic search on a website, that's not going to get a good response. That's going to get maybe one response that doesn't really make sense. But with NLWeb, you're able to tap into lots of different parts of your website to give a more coherent answer. 

So the human user or the agent user will ask a question. If NLWeb is on your website, then they can pull up the ASK protocol, and what that will do is it will tap into different parts of your data. It will tap into the structured data of your website. 

When I say the "structured data" of your website, I mean, of course, the schema markup on your website. But I mean also all of the other structured data on your website. Some people think that when you say "structured data," you're just talking about schema markup and JSON-LD and all that sort of stuff. But actually, in data science, for instance, anytime you tag or organize data, you are structuring it, and that's the kind of structured data that we're talking about here. 

So, within your website, your schema markup is absolutely structured data. But every time you put a heading tag, a title tag, all of those different elements, that's also your structured data. HTML lists and tables, all of those things that we've been doing for years as good SEOs, those are also structured data. 

Your product feeds, for instance, your RSS feed, which you might use if you have a blog or if you have a podcast, for instance. 

Your sitemaps, as well as your XML sitemaps. And even custom databases. 

So I said I'm from Wix. Within Wix, we have something called CMS pages, and there you can upload dynamic pages and dynamic datasets for lots of different kinds of content. If you have that kind of thing in there, then they can tap into that. 

So when the user asks the question, NLWeb will use all of this data, all of this structured data to give an answer, to ground the answer so that they're not making things up. It helps to reduce hallucinations. We know that AI sometimes makes things up. What NLWeb does is it helps to reduce that by giving it grounding from your actual website. 

So how does that actually translate into something that's useful for a human or for an agentic user? Well, after they check the data via your structured data, they will match up the real data with real content, and they'll use LLMs to help process this. 

So they can use the LLM to help understand and to reply. So I tested this. We're experimenting with this, and I've been looking at some of the tools that we have. Within that, we'll use the LLM to understand the question. So if I ask a really long-winded question as a user, there will be an LLM in the backend that helps to understand. 

Also, within it, there are tools that it can pull on. So it can pull on the MCP (Model Context Protocol), for instance. So it can use the MCP of a website to help make it interactive and interoperable. So if the query calls for that, then it can sort of enlist the MCP as well. 

So all of that will go back and forth, and it will check anything else that it needs to do. 

Then it will go, and it will generate the response. So, back from the other side, it will give a reply that makes sense to that human user. The best thing about this is that it only requires one sort of setup. 

So the thing that's great about this is that the agentic web is very enterprise at the moment. So when we talk about people who are using the agentic web and companies that are adopting the agentic web, we tend to be talking about people like Mastercard. We're talking about Stripe, and we're talking about all these big, huge companies, Walmart, and all of these folks. 

But what NLWeb does is it allows you to create a tool on your website that any website can use to talk to agents and to help them be more compatible. You only need to do one setup to make this work. 

So how might you do this? Well, there are a couple of different ways. So as I said, it's an open-source project. So the person who invented NLWeb is the same person who invented Schema.org. It's also the same person who invented RSS, and that's R.V. Guha. 

R.V. Guha runs GitHub, where you can talk to him about how it all works. People ask questions, and people propose different recommendations and things like that. You can get on there, and you can look at that. If you've got developers in your team, you can use them. You can ask them to go on to that GitHub repo and get all of the documentation and all of the things that they need to set this up on your website. It's one setup that allows you to tap into lots and lots of different resources and lots of different tools that will make your website more interoperable with the agentic web.

Another thing you can do is, for instance, if you are a Wix user, we have an NLWeb integration. This essentially is you tick one button, and it makes your website interoperable with the NLWeb. For this, you have to have a premium website with a custom domain. If you do that, then you can tap into the NLWeb.

There are other NLWeb partners across the web who are making this more and more visible. The more people that we have using NLWeb, the more valuable it will be for the web overall.

But in my opinion, the best thing about NLWeb as a tool and as a resource is that it taps into things that we're already doing. So I think there's that TikTok meme that you've seen, "All that work, and what did it get me?" That's kind of how people are sort of feeling about this movement to sort of LLMs optimization and agentic web and that sort of thing.

Well, all that work, what did it get you? It got you access to the NLWeb. So all the schema markup you've been doing, all the headings you've been doing, all the HTML lists, all of that stuff that you've been doing, it all comes to bear to make your website better prepared for this new era and for the new agentic web.

I think that's, personally, fantastic.