Your website has a new audience, and it's not human.
On May 19, Google announced WebMCP at I/O 2026. It's a proposed open web standard, co-authored by Google and Microsoft engineers, that lets websites expose structured actions directly to AI agents running in the browser. Instead of an agent scraping your page and guessing which button does what, your site tells the agent exactly what's available and how to call it.
It's in origin trial in Chrome 149 right now. Only Gemini in Chrome can consume it today. But the trajectory is obvious, and the implementation cost is close to zero.
Here's why you should care even if you don't care about Google.
AI Traffic Isn't Hypothetical Anymore
Adobe's Q1 2026 data tells the story pretty clearly. AI-driven traffic to U.S. retail sites grew 393% year over year. That traffic converts 42% better than non-AI traffic. Revenue per visit is 37% higher. AI-referred visitors spend 48% more time on site and browse 13% more pages.
A year ago, AI traffic converted 38% worse than human traffic. That flipped completely in twelve months.
HUMAN Security's 2026 report found that automated internet traffic grew 8x faster than human traffic in 2025. AI agent traffic specifically grew 7,851% year over year. Read that number again.
Traffic is showing up. It's spending money. And most small business websites can't even be read by the systems sending it.
The Invisibility Problem
Adobe built an AI Content Visibility Checker. Retail homepages average 75% readability for LLMs, which sounds decent until you see that product pages average 66%. A third of purchase-decision content is invisible to AI systems.
The top-performing retailers (measured by AI visit share) score 62% higher on homepage readability than the bottom group. That's not a rounding error. That's a structural gap.
Cloudflare launched an Agent Readiness Scanner in April. They scanned 200,000 of the top websites globally. MCP Server Cards and API Catalogs appeared on fewer than 15 sites. Fifteen. Out of two hundred thousand.
StudioMeyer benchmarked 21 web agencies against AI assistants. The companies that build websites for a living. Twenty out of twenty-one scored invisible. Average score: 11 out of 100.
The firms building your website haven't figured this out for their own sites yet.
What WebMCP Actually Does
WebMCP has two APIs. The Imperative API uses JavaScript to register complex tools via navigator.modelContext. That's for apps with dynamic workflows.
The Declarative API is the one that matters for small businesses. You add two HTML attributes to an existing form:
<form toolname="contact_us" tooldescription="Send a message to our team"
action="/contact" method="POST">
<input name="name" type="text" placeholder="Your name">
<input name="email" type="email" placeholder="Your email">
<textarea name="message" placeholder="How can we help?"></textarea>
<button type="submit">Send</button>
</form>
That's it. Two attributes. Your contact form just became AI-callable. Browsers that don't support WebMCP ignore the attributes entirely. Nothing breaks. It's progressive enhancement, the same principle the web has used for decades.
Your booking form, your quote request, your newsletter signup -- each one is two attributes away from being a structured tool that an AI agent can find and use on behalf of a user.
The Bigger Picture Is Bigger Than One Standard
WebMCP is one piece. The "agent-ready web" stack is forming fast:
llms.txt -- a plain-text file at your domain root that explains your site to language models. Think robots.txt, but for AI.
agents.json -- a machine-readable manifest for tool and agent discovery.
Schema.org / JSON-LD -- still the foundation for structured data, and still missing from most small business sites.
Cloudflare's isitagentready.com -- free diagnostic scanner, scores 0-100 across five dimensions. You can check your site right now.
Dan Petrovic, a technical SEO researcher, called WebMCP "the biggest shift in technical SEO since structured data." That tracks. Structured data told search engines what your content means. WebMCP tells AI agents what your site can do.
The Platform Gap Is Already Here
Major e-commerce platforms ship with AI agent endpoints enabled by default. Some have had live MCP integrations since mid-2025. Their merchants are agent-accessible without lifting a finger. AI-attributed orders on these platforms are up 11x since January 2025.
If you're running a custom-built site, a static site, or anything on a platform that hasn't added agent support, you're already behind the merchants who got it for free.
That's not a criticism of your choices. It's a build decision you didn't know you needed to make. Now you do.
This Is a Build Decision, Not an SEO Tweak
I keep seeing people frame this as a marketing optimization. Add some tags, tweak some meta descriptions, move on. That misses the point.
Your website now has two audiences. Humans who browse it. And AI agents that need to read it, understand it, and act on it -- on behalf of those same humans.
Building for both audiences is an architecture decision. It affects how you structure your HTML, how you expose your forms and actions, how you describe your services in machine-readable formats, and whether your site even shows up when someone asks an AI assistant to "find me a [your service] near me."
A mid-market e-commerce brand started with a Cloudflare Agent Readiness score of 42. After 100 days of structural changes (structured data, llms.txt, content reformatting, no new content created), they hit 78. AI-driven traffic increased 920%.
No new content. No ad spend. Just making the existing site readable to machines.
What You Should Actually Do
I'm not going to give you a step-by-step playbook. But I'll tell you where the bar is.
Run your site through Cloudflare's agent readiness scanner. Look at your score. If it's below 50, your site is functionally invisible to AI traffic.
Check whether your forms have semantic markup. Check whether you have structured data (JSON-LD) on your key pages. Check whether you have an llms.txt file. Check whether your HTML is clean enough that a machine can parse your service offerings without guessing.
These aren't exotic technologies. They're the basics of a machine-readable web, and most small business sites skip all of them.
The entry point is low-cost. Two HTML attributes on a form. A text file at your domain root. Structured data on your service pages. None of this requires rebuilding your site. But it does require someone who understands what AI agents are looking for and how to give it to them.
The Window
Gartner projects 40% of enterprise apps will embed task-specific AI agents by end of 2026, up from under 5% in 2025. McKinsey projects agentic commerce could drive $1 trillion in U.S. retail revenue by 2030.
Those are enterprise numbers, but the traffic flows downhill. When a user asks an AI assistant to book a service, buy a product, or find a contractor, the agent is going to pick from the sites it can actually read. Right now, that list is very, very short.
The businesses that get readable first will absorb that traffic. The rest won't know what they missed.
We build sites that speak both languages -- human and machine. We handle WebMCP implementation, structured data, llms.txt, and agent readiness audits as part of how we build. First conversation is free. No commitment. kief.studio/contact