The End of the Click? NLWeb and the Quiet War on Search

Author(s): Pawel Rzeszucinski, PhD

Originally published on Towards AI.

NLWeb is born

Last week at Microsoft’s Build conference, many exciting announcements were made, from open-sourcing GitHub Copilot to major upgrades for Microsoft Copilot, but one that could prove the most disruptive went largely unnoticed: the introduction of a new standard called NLWeb, or Natural Language Web. It’s a technical standard with the potential to radically reshape the way we access, discover, and monetize information online. At first glance, NLWeb is just an open-source framework for making websites easier to query via LLMs. But in my opinion behind the code, and a fashionable attempt to introduce their own standard (vide Anthropic’s MCP) lies something far more ambitious: a strategic play to redefine the interface between users and the internet, challenging the foundations of how Google search, web advertising, and digital publishing currently operate.

The Search

NLWeb allows websites to expose their content in a structured, conversationally accessible format. LLMs like Copilot, ChatGPT, or Gemini can already “talk to” websites directly although today it’s more of a monologue rather than a dialogue. Nevertheless studies show AI summaries can reduce organic clicks by up to 70%. Websites no longer need to be visited in the traditional sense to be useful — they just need to be readable by AI. This marks a shift from web browsing to conversational retrieval, from linking to answering, and from pageviews to presence in a model’s response.

What NLWeb brings to the table is standardization. No wonder it came from Microsoft as it aligns perfectly with it’s broader ambitions. For decades, Google has dominated the web economy by owning the starting point: the search bar. Users input queries, click results, and see ads, generating revenue at pretty much each step of the funnel. Microsoft’s Bing has always trailed far behind. But the rise of LLMs offers a reset. If users begin their journey by “asking” Copilot instead of “searching” Google, the power dynamic shifts. NLWeb is a vehicle to accelerate that transition. It allows LLMs to be more useful, more accurate, and less reliant on directing users to external links. But that’s nothing new…

The Ads

What I think is slowly taking place is laying the groundwork for something bigger: advertising within LLM interfaces. Right now, platforms like ChatGPT and Copilot are heavily subsidized, with millions of users accessing powerful models for free or at minimal cost. Today we see turf wars, to take as large piece of the market share as possible, but this can’t last forever. These platforms need a business model that scales, and embedding intent-aware, native ads directly into AI answers seems like a clear path forward. NLWeb makes that possible. By structuring web content for easy extraction, ranking, and synthesis, it enables models to serve not just answers, but recommendations, comparisons, and — it had to come at some point — sponsored results. The experience will look less like Google search results and more like having an informed assistant that just happens to suggest a particular hotel, product, or service.

The split

Of course, not all web publishers will welcome this future. In fact, the response to NLWeb will likely divide the publishing world in two, similarily to what we’re seeing today with regards to LLM crawlers. On one side, we’ll find those who thrive on visibility. E-commerce platforms, product aggregators, recipe blogs, review sites, and public institutions — all of them benefit from being discoverable and accessible through new channels. These sites don’t need users to stay and browse; they need to be surfaced at the right moment in a user’s journey. For them, being part of an LLM’s answer is just as good — if not better — than earning a click.

On the other side, however, are the content creators who depend on session control i.e. keeping the user within their own ecosystem, to survive. Media outlets, subscription newsletters, streaming services, and scientific publishers make money by owning the entire user experience — through ads, paywalls, or premium memberships. If LLMs start summarizing their content without driving traffic, these models collapse. For these players, NLWeb could feel less like an opportunity and more like disintermediation. Unless a revenue sharing program is introduced that benefits both parties…which today is not on the table. NLWeb seems to offer no built-in monetization for those who share their content. Unlike Google’s search ecosystem, where publishers could earn traffic and ad revenue, NLWeb’s structure shifts value upstream, toward the platforms that operate the LLMs. Whether this imbalance is addressed through rev-share agreements, metadata-based tracking, or affiliate mechanisms remains to be seen. Without such incentives, major publishers may choose to wall off their content or pursue legal challenges, especially in regions where copyright enforcement is tightening.

I’m surprised Microsoft has not made a bigger deal out of this announcement. At the strategic level, NLWeb is Microsoft’s attempt to leapfrog Google — not by beating it at traditional search, but by changing the game entirely. Google has naturally been embedding Gemini deeper into its products, transforming Search into a more AI-native experience. But Microsoft’s approach is more radical: turn the open web into a structured, conversational layer that can be queried by any model, anywhere. This isn’t just about search. It’s about redefining what it means to “visit” a website, what it means to “rank,” and who controls the digital pathways between questions and answers. Soon ‘owning a website’ might mean more having a well structured JSON file, rather than a flashy and beautiful interface.

Ultimately, NLWeb could become the new SEO frontier. Just as companies once scrambled to optimize their websites for Google’s crawler, we may soon see businesses optimizing their content for LLM visibility, using schemas, protocols, and interfaces that help models extract and repackage their information. Those who adapt early may gain disproportionate exposure in this new AI-native landscape. Those who wait may find themselves invisible, not just in rankings, but in relevance.

The standard

New standards, especially in fast-moving fields like AI and the web, tend to follow one of three paths. Sometimes, a standard “sticks” — it gains traction quickly, achieves widespread adoption, and becomes the foundation others build upon. This is what Microsoft is hoping for. Other times, the initial release sparks interest but is quickly overtaken by a superior or more developer-friendly alternative that captures the momentum and effectively replaces the original. This is likely what Google, Anthropic and others are currently working on. And in many cases, a standard simply fades into obscurity, never achieving critical mass, often because the timing was off, the implementation was too cumbersome, or the ecosystem wasn’t ready to adopt it. This seems unlikely give how embedded LLMs are today.

We are to see how this one plays out, but the web is no longer just for humans to browse. It’s becoming a space for machines to read, reason, and respond on our behalf. And NLWeb may be the blueprint for that transition.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI