The potential for a natural-language based web

With new language models launching seemingly every other week and new protocols attempting to standardise the Wild West that is the AI-age internet, keeping up with every advancement in the artificial intelligence space can be overwhelming. But when Microsoft unveils a shiny new AI project, itʼs worth paying attention.

On May 19th, 2025, Microsoft introduced NLWeb (Natural Language Web), which aims to make natural language interfaces a native part of the web. Much like how HTML became the standard for building websites, NLWeb strives to become the internet-standard for making websites ready for the AI era. In Microsoftʼs own words, their goal is to make NLWeb the “fastest and easiest way to effectively turn your website into an AI app.ˮ (Source)

A key thing to note: NLWeb itself is not a protocol, product, or tool, but rather a collection of existing protocols and open-source tools. When combined, this infrastructure will allow humans and web-surfing AIs to engage with websites using simple, conversational language.

Donʼt just visit websites, talk to them!

Think of a time that you had to scour a poorly designed website to answer a seemingly simple question. Maybe you were deciphering your tax situation with fifty government webpage tabs open, or shopping for a niche gift for a friend on some obscure online marketplace. The internet has seen many technological advancements over the years; however, the struggle of navigating content-dense webpages or sites with subpar UX still persists.

NLWeb aims to remedy this by offering a framework that exposes a natural language input on your webpage — allowing users (and AIs) to simply ask for what they want. Ideally, this input would be visible on your homepage or, even better, present across the entire interface.

How does it work?

On the official project page, Microsoft clarifies that NLWeb is not a final solution but a proof of concept, encouraging the community to iterate on it. Still, the core idea can be broken into a few components:

Inputs

For humans, this might be a simple text box or a full-featured chat interface. For AIs, NLWeb supports the Model Context Protocol (MCP) — a promising standard for agent-to-agent communication.

Not familiar with MCP? Check out our blog post: https://www.timeundertension.ai/imagine/model-contextprotocol-what-australian-marketing-leaders-should-know

Behind the scenes

No matter how you interact with an NLWeb site, natural language requests generally follow the same pattern. The key assumption is that all of the content on your webpage has been processed, vectorised, and stored in a vector database.

When you ask a question to the website, your query is optimised by an LLM to increase the likelihood that relevant results are returned from vector searching. Maybe the LLM knows exactly what youʼre asking, but needs to expand on your question by using more specific keywords. The vector search then surfaces relevant site content, and the results are displayed however you choose.

One benefit of this approach is that NLWeb is technology agnostic, meaning that it doesnʼt rely on one single LLM provider or database hosting service.

Value Proposition

For Users

Visitors to an NLWeb-enabled site can find information easily using plain language. No more keyword guesswork, filter toggling, or endless clicking — just ask and get results. The heavy lifting is handled by embeddings and LLMs.

For Creators

The opportunities for what an NLWeb experience could look like on your website is groundbreaking. At minimum, you can improve the overall user experience of your site and make your page more accessible to the ever-growing number of AIbased web-surfers.

For instance, imagine a recipe blog. Visitors could input ingredients they have at home, and your NLWeb interface could instantly suggest relevant meals by skimming through your recipe archive.

Thereʼs no one-size-fits-all implementation; the framework remains the same, but the execution is entirely use-case dependent.

For AIs

Letʼs imagine for a moment that 90% of the internet has adopted NLWeb.

You tell your AI assistant that you need black running shoes, size 10, available for same-day pickup. The AI queries nearby sporting goods stores via NLWeb interfaces. Each store responds with options. The AI compares prices, checks hours, confirms availability, and evaluates return policies. It then presents the optimal choice — say, a store 15 minutes away with your shoes in stock.

This might sound achievable today — and to some extent, it is — but most of todayʼs systems rely on scraping. NLWeb eliminates scraping by enabling direct conversations about website content.

Sounds great, whatʼs the catch?

Given that this project is in its infancy stage, itʼs not exactly a no-code setup. If you want to NLWeb-ify your site, it will require some programming experience. Even though there is a sample project you can setup, the creators encourage integrating the code into an existing application environment instead of running a standalone NLWeb server. If this project gains traction, there will likely be some providers in the near future that will offer prepping your website for the NL-age as a service.

There are some bigger names that have become early adopters, namely Eventbrite, OʼReilly Media, Shopify, and Tripadvisor to name a few. For the time being though, there is some upfront technical investment if you wish to hop on the NLWeb train. It is worth considering that the success of this framework depends almost entirely on how widely it is adopted by websites themselves.

Another item to consider is the impact that this could have on your websiteʼs ability to generate ad revenue. If users ask a question and quickly get what they need, they may leave before clicking around—reducing page views. Plus, integrating LLMs and vector databases introduces infrastructure and cost considerations. This opens the door for innovation in ad delivery via chat interfaces.

Final Thoughts

Proposing new standards is always welcome—especially given how sprawling and sophisticated the internet has become. The key takeaway is that NLWeb is not a one-size-fits-all solution but rather a framework for ensuring your webpage is ready for the AI phase of the internet.

Curious to learn more? Reach out to anyone at Time Under Tension, weʼd love to chat NLWeb with you!