NLWeb: Microsoft’s open-source bridge to the agentic web

The evolution of search is entering a new phase. After traditional search engines, rich results and AI-generated answers, the idea of the agentic web is starting to gain momentum: a web where AI agents do not simply read pages, but interact directly with websites to answer, compare, recommend and perform tasks on behalf of users.

In this context, NLWeb, Microsoft’s open source initiative, emerges as an important bridge between traditional websites and this new layer of interaction based on natural language. The proposal is easy to understand, but deep in its implications: allowing any website to become more queryable by humans and AI agents.

From a web read by humans to a web queried by agents

For many years, websites were designed mainly for human users. Information architecture, design, menus, content and forms were built for someone to browse, click, read and decide.

With the arrival of AI agents, this model is starting to change. An agent does not only want to find a page. It wants to understand content, validate data, cross-check information and, in many cases, act. It may look for an available product, confirm opening hours, compare prices, select an option or prepare a personalised recommendation.

For this to happen reliably, websites need clear, consistent and structured data. This is where schema markup, Schema.org and formats such as JSON-LD become essential.

Why is schema markup becoming even more important?

Schema markup was already relevant for SEO because it helps search engines understand entities, products, articles, events, organisations, reviews, prices, availability and relationships between pages. This information can contribute to rich results and to a more accurate interpretation of content.

In the agentic web, the role of structured data becomes even more strategic. An AI agent needs to know not only what exists on a page, but also whether that information is complete, up to date, trustworthy and actionable.

A product page, for example, should not only indicate the item name. It should include price, availability, features, reviews, category, image, brand and possible variations. The more complete and coherent the structured data is, the more likely an automated system will be able to interpret it with confidence.

Reading disorganised HTML is also harder and more expensive for AI systems. Structured data reduces ambiguity, makes processing more efficient and helps agents find answers without relying only on free-text interpretation.

What is NLWeb?

NLWeb was presented as Microsoft’s open source initiative to make it easier to create conversational interfaces on websites. In practical terms, it helps a website answer questions in natural language, making content more accessible to users and AI agents.

The difference is relevant. A traditional website is browsed. A website prepared for NLWeb can be queried. Instead of forcing the user, or an agent, to move through menus and pages, the system can receive direct questions and return structured answers.

Imagine the difference between manually checking a list of services and asking directly: «What solutions do you have for international Shopify online stores?». In an environment prepared for the agentic web, the answer should be clear, contextual and based on structured information from the website itself.

Schema is the foundation, NLWeb is the interaction layer

Schema markup works as a layer of meaning. It tells systems what each element represents: a product, an organisation, a person, an article, a review, a service or a location.

NLWeb adds an interaction layer. It allows that information to be queried more directly through natural language and more structured answers. In other words: schema helps the agent understand; NLWeb helps the agent ask and obtain useful answers.

This distinction is essential for marketing, SEO, development and ecommerce teams. The future of digital visibility will not depend only on having indexed pages. It will also depend on the ability of those pages to be understood, queried and used by automated systems.

How to prepare a website for the agentic web?

Preparation starts with a serious audit of structured data. It is not enough to have schema on a few pages. It is necessary to ensure consistency, completeness and alignment between what is visible in the HTML and what is declared in JSON-LD.

Some priorities should be clear:

• Complete critical pages: products, services, articles, events and institutional pages should have robust and updated markup.
• Avoid contradictory data: price, availability, addresses, contacts and reviews should match between visible content and structured data.
• Use JSON-LD: this format makes programmatic reading easier and is a solid option for search engines and AI agents.
• Think in terms of an entity graph: pages, authors, categories, services, products and organisation should be connected coherently.
• Automate whenever possible: online stores and websites with a lot of content need scalable processes to generate and validate schema.

Impact on SEO, GEO and ecommerce

This evolution has a direct impact on SEO and GEO, meaning optimisation for generative engines and AI answers. If agents start preferring sources they can interpret with less effort and greater confidence, websites with complete structured data gain a competitive advantage.

In ecommerce, the impact may be even greater. AI agents may compare products, check availability, recommend alternatives and guide purchase decisions. An online store with incomplete or inconsistent data may lose visibility in this new context, even if it has good design and strong visual content.

That is why brands using Shopify, proprietary platforms or more advanced architectures should start treating schema as infrastructure, not as a technical detail. Optimisation for agents does not replace traditional SEO, but it adds a new requirement: making the website readable, trustworthy and actionable by machines.

A window of opportunity for prepared brands

The agentic web is still under construction, but the direction is clear. AI systems will value websites that provide structured, current and easy-to-process information. Those who prepare early may gain authority before the market starts treating this topic as an obligation.

NLWeb shows that the next stage of the web will not only be more conversational. It will also be more operational. Websites will stop being only navigation destinations and will become active sources of response, decision and action.

At BYDAS, we are following this evolution closely, integrating digital strategy, technical SEO, development and ecommerce. If your brand wants to prepare its website for a more structured, intelligent and agent-oriented web, this is the right time to review data, architecture and technology.