70% of Top Retailers Just Failed the AI Shopping Agent Test. Your Dealership Runs on the Exact Same Plumbing.

Yesterday, Search Engine Journal published an audit that should make every retailer nervous — and every dealer curious.

Reza Moaiandin, co-founder of SALT.agency, pulled 207 top-traffic product pages from 29 of the highest-traffic retailers on the web and scored them against a 10-point rubric built from Google's own Universal Commerce Protocol documentation. Fifty-two of those pages couldn't even be read — they returned 403 or 404 errors. Fourteen more were misidentified. That left 141 real product pages to grade.

The results are worse than they sound in the headline. Every single one of the 141 pages had Product schema installed. Ninety-nine percent published a price and availability. Ninety-nine percent had a manufacturer part number or SKU. Ninety-six percent had a brand. By the standards of 2019 SEO, this is a room full of A students.

Then he checked the fields that actually matter to a shopping agent:

• 18% carried priceValidUntil — how long the price is good for.
• 13% carried shippingDetails.deliveryTime — when the customer gets it.
• 11% carried hasMerchantReturnPolicy.merchantReturnDays — what happens if they don't want it.

70% were missing all three. And 15% of the URLs — 32 pages, including ones belonging to household-name apparel and footwear brands — returned a flat 403 Forbidden to the crawler, which is the digital equivalent of locking the showroom door and going home.

Moaiandin's read on what that means is blunt: "If any of those fields are absent or incomplete, Gemini won't simply rank the product lower... Your product won't be included. Period."

Not ranked lower. Not included.

Why This Standard Suddenly Has Teeth

Six months ago this would have been an academic exercise. It isn't anymore.

On January 11, 2026, Sundar Pichai stood up at the National Retail Federation's annual show in New York and launched the Universal Commerce Protocol — in Google's words, "a new open standard for agentic commerce that works across the entire shopping journey — from discovery and buying to post-purchase support." Google didn't build it alone. UCP was co-developed with Shopify, Etsy, Wayfair, Target and Walmart, and endorsed by more than twenty other companies including Visa, Mastercard, American Express, Best Buy and The Home Depot.

Read that list again. That is not a standards committee. That is most of American retail and the entire payments rail agreeing, in public, on how machines will be allowed to buy things.

UCP now powers checkout on eligible product listings inside AI Mode in Google Search and the Gemini app for eligible U.S. retailers. And here's the part dealers should sit up for: retailers don't activate it by rebuilding their websites. They activate it through Merchant Center, using — per Google's own developer docs — "your existing Merchant Center account shopping feeds." Google also announced "dozens of new data attributes in Merchant Center designed for easy discovery in the conversational commerce era."

The feed became the storefront.

Meanwhile OpenAI moved in the same direction from the other end. In March, it stepped back from its own Instant Checkout feature inside ChatGPT, telling Digital Commerce 360: "We're prioritizing making ChatGPT search and product discovery great, with ACP serving as the infrastructure that connects users to merchants across the full shopping journey."

Both of the companies your shoppers actually talk to have landed on the same conclusion. The fight isn't over the checkout button. It's over whether the machine knows your item exists and trusts the data enough to put it in the answer.

You Have Already Lived This Once

Here's why I don't think this is hype, and why I don't think dealers get to sit it out: you already run this exact system. You've run it for years.

Every franchised dealer advertising inventory on Google is pushing a vehicle feed into Google Merchant Center. Google's own documentation for vehicle ads says the quiet part plainly — you upload your vehicle data sources, you "fill all necessary attributes for your vehicle inventory," and "Google uses these details to match a customer's search to the most relevant cars."

Now go read the rule Google publishes in its vehicle listings feed documentation. You must "submit the feed at least once every day." Google recommends every four hours. And then the sentence almost nobody at a dealership has read:

"Your vehicles stop showing to users after three days from the time Google last received and successfully processed the feed."

Three days of a broken feed and your entire inventory ceases to exist on Google. Not ranked lower. Gone. That rule has been sitting there the whole time, quietly enforcing the same principle UCP just made explicit for all of retail: in machine-mediated shopping, stale data is indistinguishable from no data.

The retailers in that audit didn't fail because they were lazy about SEO. They failed because they treated the feed as plumbing — an IT deliverable, set up once, owned by nobody, reviewed never. That is precisely how most dealerships treat theirs.

What This Means for Your Dealership

Let me be straight about the boundary, because there's a lot of breathless vendor email going around right now. Nobody is buying an F-150 inside Gemini this quarter. UCP checkout launched for retail products, not vehicles, and a car transaction involves trade equity, F&I, lender approval, titling and a test drive. That is not getting compressed into a chat window on Google's timeline.

But discovery is not waiting for checkout. Discovery is live right now, and it is being fed by the same structured product data. When a shopper asks an assistant "what certified RAV4s under $32,000 are available near Tampa, and which dealer can get me one this week," the assistant is not browsing your website like a person. It is looking for machine-readable answers to four questions. Map them to what the audit found missing:

1. What is this, exactly? Retailers failed on GTIN — 65% didn't publish one, which is what lets an agent compare the identical item across sellers. You have something better than a GTIN. You have a VIN, the most precise product identifier in any industry on earth. If your VDPs and your feed don't expose VIN, trim, drivetrain, options and mileage as clean data — not as text baked into a photo carousel or a JavaScript payload a crawler never executes — you've thrown away your single biggest structural advantage over every other retail category.

2. Is this price real, and for how long? This is the priceValidUntil failure, and it is the one that should worry dealers most. Your advertised price is a stack: MSRP, dealer discount, conditional rebates, finance-dependent incentives, expiration dates. On most dealer sites that stack lives in a disclaimer paragraph written for a human and a compliance auditor. An agent reading that page cannot tell whether $31,995 is good today, good through month-end, or contingent on the shopper being a recent college graduate financing through captive. When a machine can't resolve the price, the safe move is to leave you out of the comparison.

3. Can the customer actually get it? For retailers that's shipping time. For you it's availability truth — whether the unit is physically on your lot, in transit, sold-pending, or a phantom that's been live for eleven days because nobody reconciled the DMS. Every sold-but-still-listed unit trains an assistant that your data is unreliable, and it costs you the walk-in that shows up for a car you don't have.

4. What happens after? Retailers failed on return policy. Your version is delivery radius, exchange window, and CPO warranty terms — the trust signals a shopper asks an assistant about before they ever ask about price.

And then the ugliest finding of all: 15% of those pages simply blocked the crawler. We've written before about why whitelisting AI bots one at a time is a losing strategy. A WAF rule or a bot-mitigation product that your vendor turned on for good reasons two years ago can be silently returning 403s to the exact agents deciding whether your inventory exists. Nobody gets an alert for this. There's no line item on a report called "invisible."

What To Do About It

None of this requires a website rebuild. It requires somebody owning the data.

1. Check your feed's actual health this week. Log into Merchant Center, look at the vehicle data source, and confirm the last successful processing timestamp and the item disapproval count. Not the vendor's dashboard — Google's. If it's been more than a day, you're on a three-day clock.
2. Fetch your own VDP the way a machine does. Have someone request one of your live vehicle pages with a non-browser user agent and see what comes back. A 200 with full content is the goal. A 403, a challenge page, or an empty shell that only fills in via JavaScript is a finding.
3. Give your price an expiration date in the data, not just the disclaimer. Every advertised price should carry a machine-readable valid-through date and a clean separation between the unconditional price and conditional incentives.
4. Reconcile inventory against the DMS on a schedule. Sold units come down same-day. Set a standing number — how many live VDPs, how many physical units — and make somebody responsible for the gap.
5. Publish delivery, exchange and CPO warranty terms as structured facts, on their own pages, in plain language, not buried in a footer PDF.
6. Assign an owner. Not "the website vendor." A person at your store whose job includes the sentence "our inventory data is accurate and machine-readable." The 70% failure rate in that audit is what happens when the answer is nobody.

The Window

Every one of these platform shifts has run the same way. A standard gets published. For a while, adopting it is a competitive advantage. Then it becomes table stakes, and the businesses that waited spend two years and real money catching up to where the early movers started.

We're in the advantage window right now. Seventy percent of the biggest retailers on the internet — companies with data teams, budgets, and a direct line to Google — just failed a test built from Google's own published documentation. Your competitor down the road has not read that developer doc. Neither has their vendor.

The good news is that this is not a creative challenge. It's a discipline challenge. Accurate data, published in a form a machine can read, refreshed faster than Google's three-day patience runs out. That's it.

If you want to know how your store actually looks to the machines making these decisions, book a walkthrough and we'll pull your feed and your VDPs apart with you. It takes about twenty minutes to find out whether you're in the 30% or the 70%.