How LightSite AI Makes Websites Speak With LLMs

How LightSite AI Teaches Websites to Speak with LLMs

If you're a marketer, here is the simplest way to think about the agentic web. LLMs are not just reading your site. They are trying to complete a task for a user. And when they do, they look for the fastest possible way to search, check business information, browse products, read FAQs, and pull proof points.

A "skill" is simply a machine-readable action a website makes available to an LLM. It is not a chatbot widget and not a redesign. It is a cleaner way to tell an AI system what it can do on your site instead of forcing it to guess.

What Skills Look Like (In Plain English)

When an agent lands on a website, it usually does a few predictable things. It tries to understand who the company is, what it sells, how the content is organized, and then it looks for the exact answer it needs. Skills are just the structured version of those actions.

What We Saw After Launch (7 Days Before vs 7 Days After)

On March 2, 2026, we rolled out a skills manifest across customer websites and wanted to test one thing: do AI bots actually change behavior when a website explicitly tells them what they can do?

We compared 7 days before launch versus 7 days after launch. The strongest signal was not just that some bots fetched the manifest. It was that several of them appeared to change the way they used the site once those actions were made explicit.

The other notable shift was in path diversity. ChatGPT's path diversity dropped from 51.6% to 30%. That matters because when total usage goes up while path diversity goes down, it often suggests the bot is no longer wandering around the site broadly. It has found useful endpoints and is using them repeatedly. In plain English, it starts behaving less like a crawler and more like a tool user.

How Different Platforms Behaved

ChatGPT gave the clearest before-and-after pattern, but it was not the only platform that responded to the skills layer.

That difference is important. Not every platform responds the same way. Some adopt the skills layer immediately and heavily. Others show the effect more through behavior change than raw volume.

Why Path Diversity Matters

This was probably the most interesting part of the data for us. Most people look only at traffic volume. We think behavior matters more.

If a bot is exploring lots of different paths, it often means it is still trying to figure the site out. If traffic increases while the bot starts visiting a smaller, more repeated set of endpoints, that usually means it found useful tools and is relying on them. That is exactly the shift we hoped to see.

What This Means for Marketers

For marketers, this is not just a technical curiosity. If LLMs stop guessing and start using cleaner, purpose-built endpoints, you get a better chance of shaping what they pull, how they interpret your site, and which parts of your content they rely on.

In other words, this is one of the first practical ways to see whether LLMs are simply crawling your site or actually using it more like a toolset.

A Quick Clarification on the Data

When we say "traffic" here, we mean bot traffic, not human referral traffic from LLM apps. And when we measure behavior, we do it on the links and endpoints we control inside the LightSite layer.

In simple terms, because these are links and routes we instrument, we can observe how bots move through them. We also use canary tokens in the response body to understand the bot journey, what was fetched, and how concentrated or exploratory the interaction was.

What To Do Next (If You're a Marketer)

The practical takeaway is simple. Do not think only about being crawlable. Think about being usable.

If You Want to See This on Your Own Site

If you want, reply with your domain and I'll show you what the skills layer would look like, which endpoints LLMs are most likely to use first, and what kinds of interaction patterns you should expect to measure.