Google Just Gave AI Agents a Way to Use Your Website, Not Just Read It. Most Dealer Sites Will Fail the Test.

For two years, the entire conversation about AI and your dealership website has been about one thing: being read. Can ChatGPT parse your inventory? Does Gemini understand your service department? Are you structured cleanly enough to get cited in an AI answer? Legitimate questions, and we've written about all of them.

A development that landed quietly this spring changes the question. It's no longer just can an AI read your site. It's can an AI use it.

This spring, Google and Microsoft introduced WebMCP — a proposed open web standard that lets a website expose "structured tools like JavaScript functions and HTML forms to browser-based agents." In plain English: instead of an AI squinting at your page and trying to click the right button like a confused human, your site can hand the AI a clean menu of things it's allowed to do — search inventory, get a price, book a test drive, schedule service — and the AI just calls them. Google says the experimental WebMCP origin trial starts in Chrome 149, that "Gemini in Chrome will soon support WebMCP APIs," and that Expedia, Booking.com, Shopify, Credit Karma, TurboTax, Redfin, Etsy, Instacart, and Target are already experimenting with it.

Read that partner list again. Those are transaction businesses — booking, buying, applying. The companies whose entire value is "a customer completes an action on our site" are the ones racing to let an AI complete that action for them. Your dealership is one of those businesses too.

Two Bets: The Brochure and the Cash Register

The sharpest framing I've seen on this comes from a Search Engine Journal piece that argues the agentic web is splitting into two separate bets, and most website owners have only noticed one of them.

The first bet is identity — telling an AI who you are. This is the world of llms.txt, the markdown file Jeremy Howard proposed in September 2024 to give language models a clean index of your content. It's also the world of schema, structured data, and everything we've been preaching about making your dealership legible to a machine. Identity is a brochure. It answers, what is this business?

The second bet is capability — telling an AI what it can do here. That's WebMCP. Capability is the cash register. It answers, can the customer's agent actually get the thing done?

Here's why that distinction matters now and not in some hazy future. In June, Cloudflare reported that automated traffic overtook human traffic on the web for the first time — bots and agents now make up roughly 57% of requests to web pages, against about 43% from people, a crossover that arrived earlier than Cloudflare itself expected, driven by exactly the kind of agentic AI that WebMCP is built to serve. The visitor mix on the open web has already flipped. A growing share of the "traffic" hitting your VDPs isn't a person scrolling — it's an assistant acting on a person's behalf.

We've lived through a version of this before. When smartphones arrived, "mobile-friendly" went from a nice-to-have to a ranking requirement, and the sites that treated it as optional spent years catching up. When Schema.org launched, structured data quietly separated the dealers who won rich results from the ones who stayed a plain blue link. Each time, a new kind of visitor showed up, the platforms defined what "ready" meant, a window opened for early movers, and then readiness became table stakes. Agentic capability is the next turn of that same wheel. The new visitor doesn't scroll and it doesn't click a phone number — it wants a tool it can call.

What This Means for Your Dealership

Picture the shopper you actually want. She tells the Gemini assistant in her browser: "Find me a certified CR-V under $30k within 40 miles, get the out-the-door price, and book a test drive Saturday morning."

On a site that has placed the capability bet, the agent finds a search_inventory tool, a get_pricing tool, and a book_appointment tool, calls them in order, and comes back with a confirmed Saturday slot. The dealership never lifted a finger and just won the appointment.

On your site — if it's like most dealer sites — the agent tries to do the same job by looking at the page like a person would. It hits an inventory grid that loads through JavaScript it may never fully execute. It finds a price hidden behind an "unlock savings" gate or a "call for price" button. It reaches a test-drive form built for a human's fingers, mis-maps a field, and stalls. The agent doesn't email you to say it got stuck. It just moves on to the dealer down the road whose site let it finish.

This is the uncomfortable upgrade to everything we've said about AI readability. Being readable got you into the AI's consideration set — it's how you get found. Being usable is how you get chosen, because the agent will favor the store where it can actually complete the shopper's task. A dealer can be perfectly legible, perfectly cited, and still lose the sale at the last step because the agent couldn't book the appointment.

And notice which parts of your business are in the blast radius. WebMCP's own example use cases are booking, e-commerce, customer support, and form handling — a near-exact description of the high-value actions on a dealership site: inventory search, pricing and quotes, credit pre-qualification, trade-in appraisals, test-drive and service scheduling. These aren't top-of-funnel blog reads. They're the bottom-of-funnel conversions your website exists to capture.

What To Actually Do About It

You don't need to become a WebMCP engineer this quarter. You need to do a few specific things, in order.

• Ask your website vendor one direct question: "What's your plan for WebMCP and the Chrome agentic origin trial?" Most dealer sites run on a vendor platform, so you can't add these tools yourself — but you can find out whether the company you pay is watching this or asleep on it. The answer tells you a lot about who you're partnered with.
• Fix the actions before the standard, because the standard only exposes actions that already work. WebMCP can't book a test drive if your booking flow is broken, and it can't return a price if you hide your prices. Go use your own site the way an agent would: can you get a real out-the-door number, schedule service, and start a trade appraisal without hitting a dead form or a "call us" wall? If you can't finish, no agent will either.
• Get your inventory and pricing into clean, machine-reachable data — real structured vehicle records with year, trim, mileage, drivetrain, and price, not just images and JavaScript. This is the same durable homework that earns AI citations today; capability is built on top of it.
• Don't buy the snake oil yet. WebMCP is an early origin trial, Gemini's support is "soon," not shipped, and it's incubating in a community group, not a finished standard. Within months a vendor will email you promising to make your site "WebMCP-ready" or "agent-ready" for a fee. The standard is real; the urgency in that pitch is manufactured. Do the durable work, and treat any "secret agent format" upsell the way you'd treat every "secret Google formula" scam before it.
• Start watching your agent traffic. If more than half of web requests are already automated, you want to know what share of your traffic is agents versus people — because that ratio is where the puck is going.

Where This Goes

The last two years taught dealers to ask, "Can AI find me?" The next two will be about a harder question: "When a shopper's AI shows up to do something, can it finish?" Identity got you into the conversation. Capability is how you close it. The good news is that the work is the same work we've been pointing at all along — clean data, real prices, forms and flows that actually function — just with higher stakes and a new kind of visitor keeping score.

If you're not sure whether your site would pass that test, that's exactly the thing worth finding out before your competitor does. See how your dealership stacks up for the AI-driven web — while the window for early movers is still open.