Most brands are still catching up to AI Overviews. A quieter, more significant shift is already underway. NLWeb - the Natural Language Web - introduces a protocol that allows AI agents to query websites directly, using structured data as the interface. It is not a distant concept. It is the next layer of infrastructure being built on top of the web right now.
Crystal Carter's breakdown on Moz puts it plainly: NLWeb uses something called the ASK protocol alongside your existing structured data to make your site legible to the agentic web. The implications for how brands get found, recommended, and acted upon are substantial. If you are serious about AI visibility, this is the mechanism you need to understand.
What NLWeb Actually Is
The Natural Language Web is a framework designed to let AI agents communicate with websites conversationally. Rather than a human typing a query into a search box, an AI agent sends a natural language question directly to a site and receives a structured, machine-readable answer. The ASK protocol is the communication layer that makes this possible.
This matters because it shifts the relationship between AI systems and websites from passive indexing to active querying. Current AI search - Google AI Overviews, Perplexity, ChatGPT browsing - works by pulling from pre-indexed content. NLWeb points toward a model where agents interrogate sites in real time. Your structured data is no longer just a signal to a crawler. It becomes the actual data layer that answers the question.
The ASK Protocol and Your Structured Data
The ASK protocol is designed to work with the structured data you already have - or should have - on your site. Schema.org markup, properly implemented, gives AI agents something coherent to interrogate. If your structured data is incomplete, inconsistent, or missing entirely, an agent using the ASK protocol has nothing reliable to work with. It cannot give a confident answer about your products, services, prices, or availability.
This is where brands face a practical problem. Most structured data implementations were built to satisfy Google's rich result requirements - a defensible starting point, but not sufficient for what NLWeb demands. Rich results care about whether the markup validates. Agentic queries care about whether the markup is accurate, complete, and semantically meaningful. Those are different standards.
Consider a product page. A typical Schema.org Product implementation might include a name, image, and price. An AI agent asked 'does this come in a medium and can I get it delivered by Thursday?' needs availability, size variants, fulfilment options, and delivery timeframes - all marked up and current. Most ecommerce sites are nowhere near that level of structured data fidelity.
Why This Is an AI Visibility Problem, Not Just a Technical SEO Problem
There is a temptation to frame NLWeb as a developer concern - something for the technical team to handle while marketing focuses elsewhere. That framing will cost you. The brands that get recommended by AI agents in an agentic web are the ones whose data is queryable, accurate, and authoritative. That is a marketing and commercial decision as much as a technical one.
Think about what AI agents are being asked to do on behalf of users right now - comparing insurance policies, booking services, finding local suppliers, shortlisting software. If your site cannot answer an agent's structured query reliably, you are not in the consideration set. You do not get a second chance via a click. The agent moves to the next source that can answer the question properly.
This is the commercial reality of the agentic web. GEO and AEO have focused on getting your content cited in AI-generated answers. NLWeb pushes the requirement further: your site must be able to function as a direct data source for AI decision-making. Citation is the floor. Queryability is the ceiling brands need to be building toward.
What Preparation Looks Like in Practice
Start with a structured data audit - not just a validation check, but a content completeness review. For each major page type (product, service, FAQ, location, team), ask: if an AI agent queried this page to answer a specific customer question, would the structured data contain enough to give a confident, accurate response? In most cases, the honest answer is no.
Priority areas for most businesses: Product and Offer schema with real-time or regularly updated availability and pricing. FAQPage schema that maps to the actual questions your customers ask AI systems - not the questions you wish they asked. LocalBusiness schema with accurate, granular details for every location. HowTo and Service schema for anything that involves a process or outcome a user might ask an agent to help them complete.
Beyond markup, there is an accuracy maintenance question. Structured data that goes stale - prices that no longer reflect reality, availability signals that are wrong, contact details that are out of date - actively damages your AI visibility. An agent that cites your data and gets it wrong will not cite you again. Build processes to keep structured data in sync with your actual business state. This is operational discipline, not a one-off project.
The Broader Signal: Infrastructure Is Shifting
NLWeb is part of a broader pattern. The web's infrastructure is being re-engineered for machine-to-machine communication. The ASK protocol sits alongside other emerging standards and protocols designed to make websites legible to AI systems operating on behalf of users. Whether NLWeb becomes the dominant mechanism or one of several competing approaches is less important than the direction of travel.
Brands that invest now in structured data quality, content accuracy, and machine-readable site architecture are building capability that pays off regardless of which specific protocol wins. The brands that treat this as a wait-and-see situation are the ones that will be scrambling to retrofit when agent-mediated search becomes a meaningful share of their traffic - or their absence from that traffic becomes measurable.
The question is not whether your site will be queried by AI agents. It is whether your site will be able to answer when it is. That preparation starts with your structured data, and it starts now.