- Published
- Updated
- 9 min read

TL;DR: Chrome is testing WebMCP, a way for a page to hand AI agents a list of actions. Before that matters, a machine has to read your site at all. For a small business that means honest structured data, labeled forms and plain text pages. Google says AI search needs no special files, only good pages.
Correction, October 2026: an earlier version of this note quoted traffic growth figures from a security vendor, scanner counts, an agency benchmark, an unnamed retailer's before and after numbers, and forecasts we could not tie to a primary source. Those are removed. It also said WebMCP was already in an origin trial when Google I/O happened; Google opened the Chrome 149 origin trial on June 9, 2026. The rest is rewritten against Google's and Adobe's own publications.
Your website now has two kinds of visitors. People, who look at it. And software, which reads it on a person's behalf: search crawlers, AI assistants and, more and more, AI agents that try to fill in a form or book a slot. Structured data is the part of your site written for the second group, and most small business sites still have little or none of it.
Google made this concrete at I/O 2026 with WebMCP, a proposal that lets a page describe its own actions to an agent. My wife Amelia, Kief Studio's co-founder and CEO, covered WebMCP itself in detail in what an agent-ready website needs now. This note is the layer underneath it: what a machine needs to read your site before any agent can act on it.
What did Google actually announce for Chrome?
Google's Chrome team wrote: "At Google I/O 2026, we shared how you can prepare your websites to help agents perform tasks efficiently." It then opened testing: "In Chrome 149, you can sign up for the WebMCP origin trial." That post is dated June 9, 2026. (Chrome for Developers (opens in a new tab))
WebMCP exists to replace guessing. Chrome's documentation describes the old way as actuation, "the act of an agent simulating manual mouse clicks and text input." With WebMCP, "the website declares the element's purpose," which Google calls "more reliable than actuation." (Chrome for Developers (opens in a new tab)) The name comes from the Model Context Protocol (MCP) that AI tools already use to call outside services.
Is WebMCP finished?
No. Chrome says WebMCP "is under active discussion and subject to change in the future." (Chrome for Developers (opens in a new tab)) The form-based shortcut that got passed around, two attributes named toolname and tooldescription on an existing form, comes from a separate draft (WebMCP declarative API explainer (opens in a new tab)). The main project page says that approach "is being explored separately" and is "not part of the specification at this time." (WebMCP explainer (opens in a new tab))
So it is worth learning and not worth rebuilding your site for this month. I also wrote up the proposal from the engineering side in Google just proposed turning every website into an MCP server (opens in a new tab).
Is AI traffic big enough to care about yet?
For retail, it is growing fast and it converts. Adobe reported that AI traffic to U.S. retail sites in the first quarter of 2026 "grew 393% year over year," and that by March 2026 it "was converting 42% better" than non-AI traffic. A year earlier, in March 2025, "AI traffic converted 38% worse than standard non-AI sources like paid search and email." Adobe bases this on tracking "over one trillion visits to U.S. retail sites." (Decrypt via Yahoo Finance, reporting Adobe data (opens in a new tab))
Two cautions. That is growth from a small base, and Adobe did not publish what share of all visits it is. And it is retail data, not a plumber or a dental office. But the direction matters for everyone: the people who arrive through an AI assistant arrive ready to act.
Can AI systems actually read most sites?
Not fully. The same report found "a significant portion of U.S. retail websites aren't fully readable by the models generating that traffic," with homepages averaging 75% and product pages 66% on Adobe's AI Content Visibility Checker. (Decrypt via Yahoo Finance (opens in a new tab)) The pages closest to a purchase were the hardest to read. That is the gap structured data, clean HTML and real text are meant to close.
What is structured data, and why does it come first?
Google defines structured data as "a standardized format for providing information about a page and classifying the page content." It supports JSON-LD, Microdata and RDFa, and "In general, Google recommends using JSON-LD for structured data if your site's setup allows it." (Google Search Central (opens in a new tab))
In plain terms, structured data is a short block of labeled facts hidden in the page code: this is a business, here is its name, here is its address, these are its hours, this page is a service, this one is an article with an author. A person never sees it. A machine reads it in a fraction of a second and no longer has to guess.
That is why it comes before WebMCP. An agent that cannot tell what your business is, where it is, or what a page offers will not get far enough to call a tool. Structured data answers "what is this?" WebMCP answers "what can I do here?" You need the first answer before the second one helps.
Which structured data should a local business start with?
Start with your business itself. Google says: "With Local Business structured data, you can tell Google about business hours," plus "different departments within a business, reviews (if your site captures reviews about other businesses), and more." The required properties are the name and the physical address of the business. (Google Search Central (opens in a new tab))
From there, a practical order for most service businesses:
- Organization or LocalBusiness on the home page: name, address, phone, hours, logo, the area you serve.
- Service pages that say in text what you do, for whom and where, with structured data that matches.
- Articles with a real author and a real date.
- FAQ content written as questions your customers actually ask, answered in plain sentences on the page.
What is the most common structured data mistake?
Marking up things the page does not show. Google is direct: "don't add structured data about information that is not visible to the user, even if the information is accurate." (Google Search Central (opens in a new tab)) If your structured data claims Saturday hours and your page says closed weekends, you have taught every machine that reads your site not to trust it. Keep one source of truth for your hours, address and services, and generate both the page text and the structured data from it.
Do you need llms.txt or special AI files?
Not for Google. Its guidance for AI search features says "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary," and "You don't need to create new machine readable files, AI text files, or markup to appear in these features." The same page lists "Making sure your structured data matches the visible text on the page" among its best practices. (Google Search Central (opens in a new tab))
llms.txt is a real proposal, published by Jeremy Howard on September 3, 2024, "to standardise on using an /llms.txt file to provide information to help agents use a website." (llmstxt.org (opens in a new tab)) It costs little to add, and some tools read it. Just do not mistake it for a ranking switch. Good pages and honest structured data do the heavy lifting.
What makes a form readable to a machine?
The same things that make it usable for a person with a screen reader. The W3C's accessibility tutorial says to "Provide labels to identify all form controls," that "Labels need to describe the purpose of the form control," and that "The for attribute of the label must exactly match the id of the form control." (W3C Web Accessibility Initiative (opens in a new tab))
A contact form with three unlabeled boxes and a button that says "Go" is a puzzle for an agent and for a blind visitor. A form where each field is labeled "Your name," "Email," "What do you need help with?" and the button says "Request a quote" is not. Fix that first. When WebMCP settles, a well-labeled form is the natural thing to turn into a tool, and Chrome describes its form-based approach as adding annotations to standard HTML forms (Chrome for Developers (opens in a new tab)).
What should an agent never do on your site without asking?
Buy, cancel, delete or change an account. Chrome's documentation notes developers can "request user interaction with a confirmation dialog" before sensitive actions, and that WebMCP is "primarily designed for local browser workflows with a human in the loop." (Chrome for Developers (opens in a new tab)) Write your list of confirm-first actions now, while it is a short list.
What should a small business do this quarter?
None of this needs a rebuild. Here is the order I would work in, roughly an afternoon per step for a typical small site.
- Check your current structured data. Run your home page and one service page through Google's Rich Results Test and note what it finds. If it finds nothing, that is your first job.
- Add LocalBusiness or Organization structured data with your real name, address, hours and phone, matching the page exactly.
- Rewrite one service page in plain text. What you do, for whom, where, and how to start. If the key facts only live in an image or a slider, a machine cannot read them.
- Label every form field and give each button a verb that says what happens.
- List your three to five actions: book, quote, call, buy, subscribe. Those are your future tools.
- Get a free outside check. The free website grade from LTFI (opens in a new tab) looks at security, search, DNS, email, speed and privacy basics, which is a quick way to find the boring problems before you chase the new ones.
If you want the search side of this in more depth, our guide to how answer engines choose sources explains what assistants look for when they pick a page to quote.
Frequently asked questions
What is WebMCP?
A proposed web standard that lets a page register tools, each with a name, a description and an input schema, so an AI agent can call them instead of clicking through the page. Chrome opened an origin trial in version 149, and Google says it may still change.
What is structured data in simple terms?
Labeled facts in your page code that tell machines what the page is about: your business name, address, hours, services, articles and authors. Google recommends the JSON-LD format where your site allows it.
Does structured data help with AI search?
Google says AI Overviews and AI Mode need no special files or markup beyond normal SEO practice, and it asks that structured data match the visible text. Accurate structured data helps machines read your page; it does not replace good content.
Do I need an llms.txt file?
Not to appear in Google's AI features. llms.txt is a 2024 proposal that some tools read. It is cheap to add, but it is not a substitute for clear pages and honest structured data.
Should I add WebMCP to my site now?
Usually not yet. It is in an origin trial and still changing. Label your forms, list your main actions and plan WebMCP into your next rebuild.
How do I check my site's structured data?
Run a page through Google's Rich Results Test or look at the enhancements reports in Search Console. Then compare what the structured data says with what the page shows, and fix any mismatch.
Where we fit
Kief Studio is in Shrewsbury, and this is everyday work for our SEO, AEO and GEO practice: we find the questions your buyers ask, write pages that answer them with facts you can back, and add structured data so search engines and assistants can read each page. If a number cannot be sourced, we leave it out.
When a site needs more than markup, our custom web application work maps how the work happens today before choosing a screen or a stack, and builds the forms and flows people actually use. If keeping the site current has become a second job, managed website operations covers hosting, publishing and security in one place.
First conversation is free. No commitment.




