1. Introduction

In today’s AI-driven era, the way users interact with the web is undergoing a profound transformation. Search engines and chatbots have made natural language the preferred interface for information retrieval, but websites themselves remain largely static and form-based. That, when put into the context of the modern digital experience, presents growing challenges:

• Extensive product & service offerings: Businesses now provide a broader range of products and services, making it harder for users to locate what they need quickly.
• Demand for complete & rich information: Users expect comprehensive, accurate, and contextually relevant information before making decisions.
• Limited screen space & high navigation overload: With finite space to present increasingly complex information, websites rely on deeper navigation layers, often leading to frustration, confusion, and high bounce rates.

NLWeb, a project launched by Microsoft, seeks to resolve those challenges. By enabling websites to expose content and services directly through natural language interfaces, NLWeb represents a major step toward bridging conversational AI and the open web.This article will explain what NLWeb is, highlight its key benefits, and explore the main obstacles for companies adopting it. More importantly, it will present FPT Software’s unique perspective: NLWeb is not just about making websites conversational — it has the potential to redefine how digital experiences are orchestrated.

2. What is NLWeb and How does it work?

NLWeb definition
The Natural Language Web (NLWeb) is a new open standard introduced by Microsoft in 2025, designed to make websites natively accessible to large language models (LLMs) and AI agents. At its core, NLWeb enables AI systems to query structured information directly from websites — such as product catalogs, event listings, or recipe databases — in a way that is both machine-readable and semantically rich.
Traditional websites are optimized for human readers but not for AI agents. LLMs often need to crawl and parse raw text, a process that is resource-intensive and error-prone. NLWeb addresses this by providing a protocol for sites to publish structured content into a vector database, enabling AI systems to understand, search, and retrieve information through natural language queries. 
How NLWeb works 
Technically, NLWeb uses common web standards such as JSON-LD and Schema.org markup to structure data. Once published, this content is ingested into a vector database, where it can be semantically queried. AI models then connect through Microsoft’s Model Context Protocol (MCP), which ensures interoperability between LLMs and the NLWeb-enabled sites. In other words, NLWeb transforms static websites into conversationally interactive knowledge bases.
An example of early NLWeb adopter would be TripAdvisor – where NLWeb is integrated to allow AI assistants to answer nuanced travel queries such as “family-friendly hotels near Rome with free breakfast.”
While NLWeb is broadly recognized as a framework for enabling natural language interactions on the web, here at FPT, we see its potential extending even further - redefining not only how users search, but how the entire digital experiences are orchestrated. By combining NLWeb with AI-driven assistants, websites can move beyond static responses to actively guiding users — determining what information should be presented, when, and in what format — to support decision-making throughout their journey. This can be considered as “agentic-powered dynamism”: interfaces where AI agents adapt dynamically to the conversational context and semantic queries of each user.
For example, consider an e-commerce scenario, specifically for PC Hardware. When a user asks, “What hardware upgrades do I need to play Black Myth: Wukong smoothly?", NLWeb would not just return a list of compatible hardware components. In our vision, the system could curate a set of optimal upgrade components displayed in a purpose-built template, highlight the key performance attributes that make them suitable, explain the reasoning behind each recommendation, and even guide the user through selection, checkout and installation instructions. In this way, NLWeb transforms websites from passive catalogues into proactive assistants, helping users reach decisions more effectively and confidently.

3. Key Benefits & Opportunities of NLWeb

NLWeb offers significant advantages for businesses, developers, and users. Its promise lies not only in technical efficiency but also in the strategic opportunities it unlocks.
3.1. Improved user experience
For end-users, the benefit is clear: faster, more intuitive, and convenient interactions, coupled with richer, higher-quality information.

• Instead of navigating multiple menus and complicated sitemap on a website, users can simply ask an AI agent a question and be redirected to their desired content or receive relevant, structured answers immediately.
• For E-commerce sites, this means no more applying multiple, rigid filters that oftentimes cannot provide a high level of personalization (due to large search language gap), users can just describe their desired product in natural language and have the suitable items filtered out for them.
• For websites that provide large volumes of information — such as news platforms, academic research portals, and corporate knowledge bases — NLWeb significantly enhances the user experience by enabling direct, content-level querying. Instead of manually scanning through lengthy articles or datasets, users can interact with the content conversationally and receive precise, context-driven questions, such as:
  • “What are the sources supporting the statistics cited in this article?”
  • “Summarize the key findings of this research paper for me.”

3.2. Enhanced business visibility through AI-readiness and agentic web integration
For businesses, AEO (Answer Engine Optimization) is similar to what SEO (Search Engine Optimization) did for the early internet. With NLWeb, websites can provide structured, high-quality data that these agents can directly access, positioning NLWeb as a key enabler of AI-driven discoverability.
As the internet evolves toward an agentic web — where AI assistants act as intermediaries between users and services — NLWeb allows websites to become active participants rather than passive data sources. Instead of being scraped for information, websites can proactively serve intelligent, context-aware responses tailored to user intent.
3.3. Improved accuracy and relevance
Currently, LLMs like GPT or Gemini rely heavily on probabilistic reasoning over raw text. When asked a question such as “What’s the difference between this laptop and the previous model?”, the AI must parse product pages designed for human reading — often cluttered with marketing copy, images, and inconsistent formats. This parsing is error-prone, leading to incomplete or even hallucinated answers. 
NLWeb mitigates hallucination, error-prone or incomplete answers by allowing businesses to publish structured data, rather than raw text, directly into a semantic framework. This structured information enables AI models to generate answers that are not only accurate but contextually relevant to the query.
3.4. Saving resources for fine-tuning AI data
One of the biggest bottlenecks in scaling large language models is data acquisition and processing costs. Today, most LLMs rely on scraping vast portions of the open web, cleaning noisy data, and then repeatedly re-training or fine-tuning on these massive datasets. By enabling websites to publish clean, structured, and query-ready data directly into vector databases, AI companies no longer need to expend resources on indiscriminate crawling and fine-tuning. This makes the entire AI ecosystem more sustainable.
3.5. Enabling adaptive and agentic interfaces
Beyond these immediate advantages lies the opportunity that FPT emphasizes: agentic-powered dynamism. With NLWeb as the foundation, websites can evolve into dynamic, adaptive systems that anticipate user needs and guide them toward outcomes. This transforms the user experience from passive browsing to active collaboration with an intelligent agent. Companies that embrace this shift will position themselves as leaders in customer-centric digital experiences.

4. Main Obstacles for Companies When Adopting NLWeb

4.1. Integration complexity
Implementing NLWeb requires technical expertise in structured data (e.g., JSON-LD, Schema.org), vector databases, and secure API design. Many organizations — especially smaller businesses — may lack in-house capacity. Without strong implementation partners, adoption could be slow.
4.2. Governance and standardization
Because NLWeb is a rather new technology, there are few established governance mechanisms. Who decides the rules for data structuring, security, and compliance? Until a formal standards body is in place, businesses may be concerned about long-term stability.
4.3. Security and privacy risks
In May 2025, security researchers disclosed a critical flaw in the NLWeb reference implementation that exposed websites to potential misuse as it allows remote users access to sensitive files including system configuration and cloud credentials.  Although Microsoft quickly patched the issue, the company has not issued a formal CVE (Common Vulnerabilities and Exposures) entry. This has sparked debates about whether NLWeb security practices are mature enough for widespread adoption.
Privacy is another concern. By making structured data more accessible, companies risk exposing sensitive or proprietary information if implementations are not carefully scoped.
4.4. Organizational readiness
Adopting NLWeb is not just a technical decision but also a strategic one. Businesses must align their marketing, IT, and compliance teams to ensure that the data they expose via NLWeb is accurate, sufficient, up to date, and compliant with regulations. This requires process changes that some organizations may resist.
4.5. Lack of industry best practice
The last significant challenge in adopting NLWeb today is the absence of well-defined industry standards or implementation methodologies. Unlike mature technologies such as traditional web frameworks or cloud-native architectures, NLWeb is still relatively new and rapidly evolving. At present, there is no widely accepted framework for defining NLWeb integration scope, designing industry-specific conversational flows, or setting metrics to measure its effectiveness.

5. Conclusion

NLWeb is more than a new web standard; it is a gateway to the next era of digital experiences. By enabling websites to expose their content and services through natural language, it lowers costs, improves accuracy, and expands accessibility. But its true promise lies in what comes next: the evolution of agentic-powered dynamism, where AI agents actively guide and support users throughout their journeys.
At FPT, we are pioneering the development of Adaptive and Agentic Interfaces, redefining how users interact with digital experiences. Through our initiative at ON.E - a platform showcasing a variety of advanced digital experience services, businesses can explore our vision of the future of NLWeb — where user journeys are intelligently personalized and dynamically tailored through conversational interactions.
We are also proactively developing a structured methodology to guide NLWeb adoption. Our goal is to create comprehensive frameworks and reference architectures for common NLWeb-powered scenarios that are industry-specific, enabling scalable solutions that can be applied across multiple industries.