Search engines read your website. AI agents have to use it.
Your customers are already asking AI who to hire. Two small files decide whether it understands your business and whether it can act on it. Most websites have neither, and every month without them is a month an assistant sends the job somewhere else.
Two files, two jobs
One is the briefing. The other is the controls.
Your website was built for people looking at a screen. AI agents don't look. They read structure, and they need to be told what they can do. If your site can't tell them, your competitor's will.
The briefing
A short, plain-language file that tells AI systems who you are, what you offer, where you work, and which pages matter. It answers the questions an assistant would otherwise have to guess at by scraping your whole site.
The controls
A new web standard backed by Google and Microsoft that lets your site hand an AI agent named actions, like "request a quote" or "check availability," so it can finish the job instead of guessing which button to click.
llms.txt
Stop making AI guess who you are.
When an assistant visits a typical business website, it gets menus, sliders, pop-ups, scripts and a footer full of links. It has to piece together what you do from all of that. Sometimes it gets it wrong. Sometimes it gives up and describes your competitor instead, because their site was easier to read.
An llms.txt file removes the guesswork. It sits at the root of your site, like robots.txt, and gives AI a clean summary in plain text:
Why it matters now
- Google now checks for it. Lighthouse 13.3 added llms.txt to the Agentic Browsing audit in PageSpeed Insights, the same report anyone can run on your site in 30 seconds.
- Your prospects can see you don't have it. So can every agency pitching your competitors. A failed check is a public signal that your site is behind.
- It puts you in control of the story. You decide which services, locations and pages AI should know about, in your own words, instead of leaving it to whatever an assistant pieces together.
WebMCP
The difference between being read and being chosen.
Assistants are moving from answering questions to doing things: booking, requesting, comparing, filling in forms. Today, an agent does that the hard way. It looks at your page, guesses which button is "Get a quote," clicks, and hopes the form doesn't change shape while it's typing.
WebMCP replaces the guessing. Your site tells the agent, in a format it understands: here are the actions you can take, here's what each one needs, and here's what you'll get back. The agent calls the action directly. No guessing, no broken forms, no abandoned requests.
What that looks like for a business
The agent passes the customer's details and need straight into your lead flow.
The agent asks when you can come out, and gets a real answer.
The agent asks whether you handle a specific job in a specific area, and gets a yes or no.
Where it stands today
- Chrome is already running WebMCP as a public trial, which means sites can switch it on for real visitors right now.
- Google's own Lighthouse audit already checks whether a site's WebMCP actions are set up correctly.
- Microsoft co-authors the standard and has it in Edge behind a setting. The two biggest browser makers are building toward the same thing.
The first business in your market that lets an agent finish the job doesn't have to be the best. It just has to be the easiest.
Same market. Same Google decline.
Two competitors. Only one is replacing the traffic.
Google is sending fewer clicks to everyone. When an AI summary appears, people click a search result about half as often. The question isn't whether your Google traffic shrinks. It's whether anything replaces it.
Competitor A
Agent-ready site: llms.txt, WebMCP, clean structureBut every month, more of those customers arrived through an assistant instead. The decline became growth.
Competitor B
Standard site: fast, but invisible to agentsFast site, good reviews, steady SEO. It still lost a third of its traffic, because the customers who switched to asking AI never found it.
Illustrative scenario, not client data. Monthly visits indexed to 100 at month 1.
If you wait
Nothing breaks tomorrow. That's the trap.
Your site won't stop working because it's missing these files. Your phone won't go silent next week. That's exactly why most businesses will ignore this until it's expensive, and why the ones that move first get the market to themselves.
We've seen this before. When Google started favoring mobile-friendly sites, the businesses that had already rebuilt didn't notice. The ones that waited spent the next year paying for rush redesigns while their competitors collected the traffic.
This time it's moving faster. AI assistants are already recommending local businesses. Google already built the audit. Chrome already lets agents take action. Every month you wait is a month your competitor gets to be the business AI finds first.
Built into every Agentic Website
Find out which competitor you are.
Every Lead Agentics website build includes a written and maintained llms.txt, WebMCP actions for your primary conversion, a clean accessibility tree and zero layout shift. We don't just recommend it. Our own site runs it: leadagentics.com/llms.txt.
Sources: BrightLocal Local Consumer Review Survey 2026; Pew Research Center (AI summaries and Google clicks, July 2025; AI chatbot use, June 2026); Google Lighthouse 13.3 release notes (May 2026); Chrome for Developers, WebMCP documentation; W3C Web Machine Learning Community Group, WebMCP draft. The competitor chart is an illustrative scenario, not client data. WebMCP is in a Chrome origin trial and may change. Agentic Browsing is marked experimental by Google and is not a confirmed ranking factor.