AI Protocol Adoption: Where the Web Stands in March 2026

Three protocols are emerging as the infrastructure layer between websites and AI agents. llms.txt tells AI systems what your site is about and which pages to prioritize. NLWeb lets AI agents query your site's content using natural language. MCP (Model Context Protocol) creates a structured interface for AI agents to interact with your site's data and services.

When we launched AgentReady™, I was curious to see how quickly these protocols were being adopted. We checked all 5,000 sites in our dataset for each protocol. The numbers are small, but the trajectory matters more than the absolute values.

Think of where HTTPS adoption was in 2014 -- around 30% of the web, growing slowly, with most sites seeing it as optional. Within four years it was essentially mandatory. AI protocols are earlier in that curve, but the adoption speed is faster because the stakes are clearer and the implementation effort is lower.

12% of sites in our dataset have an llms.txt file. This is up from approximately 6% when we ran preliminary scans in December 2025. The doubling suggests strong growth momentum, though from a very low base.

Adoption varies enormously by industry. Tech/SaaS leads at 24%, which makes sense given developer familiarity with protocol files. Media & Publishing follows at 18%, driven by publishers who want to ensure their content is discoverable by AI systems. Education sits at 14%.

At the bottom, e-commerce has only 5% adoption, healthcare 4%, and real estate 3%. These are the industries where AI protocol awareness has barely penetrated.

Quality varies too. Of the 600 sites with llms.txt, only 41% have well-structured files with proper descriptions, categorized links, and accurate content summaries. The rest have minimal files, sometimes just a URL list with no context. Our llms.txt creation guide covers what a well-structured file looks like.

The score impact is significant: sites with a well-structured llms.txt file score +8 points on average versus sites with no llms.txt.

NLWeb adoption sits at 3%, roughly a quarter of llms.txt adoption. The lower number reflects NLWeb's higher implementation complexity. While llms.txt is a static text file, NLWeb requires a server-side endpoint that can process natural language queries and return structured responses.

The sites implementing NLWeb are overwhelmingly Tech/SaaS companies (62% of NLWeb adopters) and large publishers with engineering resources. We found zero NLWeb implementations among small business sites, Wix sites, or basic Shopify stores.

However, NLWeb has the highest score impact per protocol: sites with NLWeb endpoints score +14 points on average. This is because NLWeb directly enables AI agents to query content, which lifts both the AI Protocols subscore and the Bot Access subscore (since the NLWeb endpoint itself functions as an AI-accessible content interface).

The NLWeb implementation guide covers who should implement it and who should wait.

MCP server adoption is at 1.5%, making it the least adopted of the three protocols. MCP's implementation requires creating a server that exposes structured tools and data to AI agents, which is significantly more complex than either llms.txt or NLWeb.

The 75 sites in our dataset with MCP implementations are almost exclusively enterprise SaaS platforms, API-first businesses, and developer tool companies. These are organizations building products that AI agents will interact with as tools, not just information sources.

For most websites, MCP is not yet relevant. It becomes important when your site provides a service (booking, purchasing, data lookup) that AI agents might execute on behalf of users, rather than just information they might reference. Our MCP guide for website owners explains the distinction.

The score impact of MCP is +6 points for sites where it's applicable, but the Protocol Trifecta (llms.txt + NLWeb + MCP) yields a combined +22 points, suggesting synergy between the protocols.

Just 40 sites out of 5,000 have implemented all three protocols. We call this the Protocol Trifecta, and these sites score an average of 89 out of 100 on AI readiness.

The Trifecta sites share a profile: they're well-funded tech companies with dedicated developer relations or technical SEO teams. Companies like Stripe, Vercel, and Anthropic themselves have led the way, which makes sense given their proximity to the AI ecosystem.

But I don't think the Trifecta will stay exclusive for long. SaaS tools and WordPress plugins are starting to automate llms.txt generation. NLWeb-as-a-service platforms are emerging. And MCP hosting platforms are simplifying server deployment. By our next quarterly scan, I expect the Trifecta adoption rate to cross 2%.

For the full strategic breakdown of all three protocols, see our comprehensive protocol guide.

Given the data, the priority order is clear.

Start with llms.txt. It takes 10-30 minutes, works on any platform, and delivers +8 points. There's no reason not to have one. It's the AI equivalent of having a sitemap.xml -- a baseline signal that costs nothing and helps AI systems understand your site.

Add NLWeb if you have engineering resources. The +14 point impact is compelling, but the implementation requires a server-side component. If you're on Next.js, Nuxt, or any framework with API routes, it's feasible. If you're on Wix or basic Shopify, wait for platform-level support.

Consider MCP only if you're a service provider. If your business provides a service that AI agents should be able to invoke (search, booking, calculations, data lookups), MCP is worth the investment. If your site is primarily informational, MCP won't add meaningful value yet.

The adoption data will evolve quickly. We'll update these numbers in our June 2026 report.