Imagine you’re on your favorite recipe site, typing into a search bar, “I need a vegan dessert for a summer barbecue.” A moment later, the site doesn’t just spit out a list of links—it hands you a curated selection of plant-based sweets, factoring in seasonal ingredients and even suggesting pairings for your grill party. Or maybe you’re shopping for hiking gear and ask, “What’s the best tent for windy Colorado trails?” The retailer’s site doesn’t make you sift through pages of products; it delivers a tailored list, complete with specs, photos, and weather data for those gusty peaks. This isn’t a pipe dream—it’s the kind of experience Microsoft’s new NLWeb protocol aims to bring to every corner of the internet.

Announced at the Build 2025 developer conference, NLWeb is Microsoft’s latest attempt to reshape how we interact with the web. It’s not a product, an app, or even a single piece of software. It’s an open protocol—a set of rules and tools that lets any website or app developer add sophisticated, AI-powered, natural-language search and interaction features with minimal effort. Think ChatGPT-style conversations, but baked directly into your favorite sites, powered by their own data and expertise, and dirt cheap to run. To understand why this matters, and what NLWeb really is, let’s dive into the vision, the tech, and the challenges behind Microsoft’s big bet on the future of the web.

The problem with today’s web

The internet is a miracle of human ingenuity, but it’s got a problem: it’s getting harder for websites to stand out. Search engines like Google have long dominated how we find information, funneling users through their algorithms and often prioritizing paid ads or popular sites. Now, AI chatbots like ChatGPT and Claude are taking it a step further, answering questions directly and sometimes bypassing the original sources altogether. For publishers, bloggers, and small businesses, this is a nightmare. Their content fuels these AI models, but they get little traffic or credit in return.

Ramanathan V. Guha, a Microsoft technical fellow with a storied career in shaping the web, sees this as a dangerous trend toward centralization. Guha’s been around the block—having co-created RSS feeds and Schema.org, two foundational standards that made the web more accessible and structured. He’s not just some tech exec spinning a tale; he’s got the receipts to prove he knows how to build tools that stick. With NLWeb, he’s aiming to give the web’s creators—its publishers, developers, and small businesses—a way to fight back.

What is NLWeb?

So, what exactly is NLWeb? At its core, it’s an open protocol that lets any website or app add natural-language search and interaction features. If you’ve ever chatted with a bot like ChatGPT, you know the vibe: you ask a question in plain English (or any language, really), and it responds with a clear, relevant answer. NLWeb makes it possible for any site to offer that kind of experience, without needing to reinvent the wheel or pay through the nose for a third-party AI service.

Here’s how it works. A developer adds a few lines of NLWeb code to their site, connects it to an AI model of their choice (like OpenAI’s GPT-4o mini or even cheaper alternatives), and feeds it their site’s data—think an RSS feed, a product catalog, or a database of articles. NLWeb handles the logistics: it takes a user’s free-form question, processes it through the AI, and returns a structured answer that the site can display however it wants. The magic is in the simplicity. You don’t need a PhD in machine learning or a massive server farm to make it work.

To see it in action, Guha shows a demo of Serious Eats, the popular food site, with NLWeb integrated. He types into the search bar: “Give me spicy, crunchy appetizers for a party.” In seconds, the site returns a handful of recipes, each with images and links, all meeting the criteria. Then he refines it: “Actually, make them vegetarian and good for Diwali.” The system remembers the vegetarian preference and pulls up dishes like spicy chickpea fritters, perfectly suited for the festival. Next, he jumps to an outdoor retailer’s site and asks, “Find me a jacket warm enough for Quebec winters.” The results include product listings, complete with insulation ratings and local weather data, presented in a clean, shoppable format.

What’s striking is how natural it feels. Unlike traditional search, where you’re guessing keywords and wading through irrelevant results, NLWeb lets you talk to a site like you’d talk to a knowledgeable friend. And because it’s limited to the site’s own data, plus some general knowledge from the AI model, the answers are focused and relevant—no tangents about unrelated topics.

NLWeb’s biggest selling point is its accessibility. Traditional web search is a beast to maintain. Companies like Google spend billions crawling and indexing the internet, and even smaller players running site-specific search (like Google Custom Search, which Guha also helped build) face steep costs. Websites, meanwhile, have to optimize their pages for SEO or pay for ads to stay visible. NLWeb flips this model on its head.

How cheap? Guha says NLWeb can run on lightweight models like GPT-4o mini, which costs a fraction of what bigger models like GPT-4 demand. For small businesses or independent publishers, this is a game-changer. Instead of signing expensive deals with AI providers like OpenAI (as Shopify has done), they can roll out their own conversational search for pennies. Microsoft is already working with partners like O’Reilly Media, TripAdvisor, Eventbrite, and Shopify to integrate NLWeb, and the early demos suggest it’s flexible enough to work across industries—food, travel, e-commerce, you name it.

But it’s not just about cost. NLWeb gives sites control over their own destiny. Right now, if you ask ChatGPT for a recipe, it might summarize content from Serious Eats without sending you there. With NLWeb, Serious Eats can keep you on its site, serving up answers in its own voice and style, complete with its own ads or sponsored links. It’s a way to reclaim traffic and revenue from the AI middlemen.

Guha’s passion for NLWeb comes from a career spent making the web more open and usable. In the early 2000s, he co-created RSS, which let users pull content from sites into feed readers, laying the groundwork for everything from podcasts to news aggregators. Later, at Google, he spearheaded Schema.org, a project that structured the web’s data so search engines could better understand it. Both are now de facto standards, used by millions of sites worldwide.

Of course, Microsoft’s not bankrolling this out of charity. Guha’s candid about the business angle: NLWeb ties into the company’s broader push for conversational AI, which could drive demand for Azure-based tools like Copilot or the Model Context Protocol (MCP), an Anthropic-led open project that NLWeb integrates with. If more sites adopt conversational interfaces, Microsoft’s cloud and AI infrastructure stand to benefit. But Guha insists the protocol itself is agnostic—it works with any AI model, not just Microsoft’s, and it’s open for anyone to use.

The hurdles

For all its promise, NLWeb faces an uphill climb. First, it needs buy-in. An open protocol is only as good as the number of people using it, and convincing big players like Google or Meta to support it could be tough. Both have their own AI agendas and might see NLWeb as a threat to their control. Smaller publishers, meanwhile, might stick with lucrative licensing deals from companies like OpenAI rather than invest in a new system, even a cheap one.

Then there’s the web’s stubborn history of centralization. From Yahoo’s directories in the ‘90s to Google’s search dominance in the 2000s to today’s AI chatbots, the internet loves to consolidate power. NLWeb’s vision of a decentralized web, where every site runs its own AI-powered search, is ambitious but goes against the grain. “The aggregators have always had the better tech,” Guha admits. His bet is that NLWeb can level the playing field by giving everyone access to that tech.

There are also thorny questions about data and privacy. If a site like Serious Eats learns you’re vegan, should it share that with other NLWeb-powered sites for a seamless experience? What if you don’t want it to? And what about “agentic” AI—systems that don’t just answer questions but take actions, like booking a trip or buying that Quebec-ready jacket? Guha says these are open questions, and NLWeb’s success will depend on the web’s community—developers, publishers, and users—working together to find answers.

What’s next?

If NLWeb catches on, it could redefine how we interact with the web. Instead of one-size-fits-all chatbots, we’d have a constellation of specialized, context-aware search tools. A news site could become your personal briefing assistant, remembering your interests. A travel platform could plan your itinerary based on a single question. A blog could feel like a conversation with an expert. Each would leverage AI to deliver precise, relevant answers without handing your attention to a third-party bot.

For now, NLWeb is in its infancy. Microsoft is rolling out demos, encouraging developers to experiment, and building partnerships with major players. Whether it becomes the next RSS or fades away depends on whether the web’s stakeholders see the same potential Guha does.

In a world where AI is reshaping how we find and use information, NLWeb is a bold attempt to keep the web’s soul intact. It’s not just a protocol—it’s a vision for a future where the internet belongs to its creators and users, not just its gatekeepers. Whether that vision becomes reality is anyone’s guess, but Guha’s track record suggests it’s a bet worth watching.