Google's Two Shopping Surfaces Agreed on 1.28% of Products. Both of Them Run on Your Dealership's Feed.

Somebody put Google's two shopping surfaces side by side, ran the same query through both on the same day, and counted how often they agreed.

They agreed 1.28% of the time.

That number comes from a data study Productrise published on September 1. Over 23 days, August 9 through August 31, 2026, the company tracked more than 2 million product listings across more than 100,000 search results pages and AI Mode responses, running the same product query on both sides on the same calendar day, in the US and the UK. On the classic side it looked only at the Popular products carousel. On the AI side it took the product cards inside the answer. It matched the two using Google's stable product identifier.

Only 1.28% of the products that ranked in the carousel also showed up in AI Mode for that same search on that same day. When a product did appear on both, the merchant listed first was a different company 49.6% of the time. Prices disagreed on 38.1% of matched products, and when they disagreed, AI Mode carried the higher one 68.4% of the time.

Google's response, given to Futurism on September 2, is the part dealers should sit with: "While we haven't verified the accuracy of the claims in this report, all shopping results on Google Search, including AI Mode and the search results page, are powered by the same data source: our Shopping Graph."

Read that as a confession rather than a rebuttal. Google is saying both surfaces draw from the same well. The measurement says the same well produced a different answer 98.7% of the time.

What the study actually measured, and what it does not

The full method is published, which is more than most vendor research gives you. Over the 23 days the company took the lead offer on each side, the one in the lowest listing position, and compared seller and price. For the overlap figure it calculated a match rate each day (matched products divided by every traditional-search listing carrying a product identifier) and then averaged those 23 daily rates.

Now the caveats, because they matter and because the counter-case is real.

Productrise sells rank tracking, feed audits and feed optimization for Google Shopping. Every finding here points back at its own product. That does not make the data wrong. It does mean you read it with your hand on your wallet.

The dataset captures what was displayed on a screen. Search Engine Journal's write-up makes the limit explicit: no clicks, no purchases, so it cannot tell you what anybody actually paid. SEJ also notes the queries are primarily products the company already monitors for clients, and the study never claims to be a random sample of Google searches.

Brodie Clark, quoted inside the study itself, supplies the strongest deflator. The higher price may be a top-level grid display artifact, and in his experience "the retailer with the lowest price gets the click in the end." Click any product on either surface and a panel opens listing other sellers and their prices. The first number is one of several.

And the headline average is soft. Across matched products AI Mode ran 21.6% higher on average, but where AI Mode was the expensive side the median gap was 22.2% while the average was 88.5%, which the study says is dragged up by extreme cases such as a used item in classic search matched against a new one in AI Mode. Use the medians. When AI Mode was the cheaper side, the median gap was only 7.8%.

Strip all of that away and one finding survives every objection, because it does not depend on price at all: the two surfaces pick different products. An average of 27.8 products in the carousel. An average of 3.9 in AI Mode. An average of 0.94 products in common. AI Mode accounts for 12.3% of all products shown in the dataset. The company's July study found the same shape at a broader level, with AI Mode ranking roughly 5% of what ranks in classic search.

Same warehouse, different salesperson

Twenty years of search marketing rested on a quiet assumption: one index, one ranking, one place to be found. Position was a proxy for visibility. You were third, you wanted to be first, everyone in the tower understood the argument.

The Shopping Graph is the warehouse. AI Mode and the classic carousel are two different salespeople walking a customer through it. Same shelves. Different picks, different order, and about half the time a different vendor's name on the tag.

That is why "powered by the same data source" is technically true and operationally useless. Nobody shops the warehouse. They take whichever tour Google gives them, and Google has spent the last year making the AI tour the default one. The study points out the entry points: the AI Mode button on the search box, the AI Mode button in Chrome's address bar, and the AI Mode tab above the results. Each carries the shopper's exact query straight over. From the customer's side of the counter, nothing appears to have changed.

Where this lands on a franchised dealership

Start with the boundary, because getting this wrong is how a good number turns into a bad decision.

This study measured Shopping Graph product listings. It did not measure vehicle listings. Vehicle ads run through a separate Merchant Center add-on with its own data source, tied to your rooftops through the store_code attribute and a linked Google Business Profile. The 1.28% figure says nothing about how your VDPs behave in AI Mode. Anybody who tells you otherwise is extending a number past its evidence.

Here is what it does touch.

Your parts, accessories and tire business is literally in this dataset's world. If your store runs a parts e-commerce site with a Merchant Center feed, you are competing on exactly the two surfaces that were measured. The carousel gave you 27.8 slots to fight over. AI Mode gives you 3.9. Scarcity that severe makes visibility close to winner-take-most, and the study found that carousel position carried over into AI Mode almost never.

Price may be a weaker lever there than you are used to. Hugo Huijer, who ran the study, reads the pattern as evidence that "the cheapest price is less of a factor in AI Mode." Treat that as his interpretation and not as a measured finding, because the data shows correlation on display prices and never tests why. If it holds, it is genuinely good news for a franchise parts counter that cannot beat a marketplace on price and never could.

A different seller on your own product is the finding to worry about. Callum Lockwood of Re:signal, also quoted in the study, reads the 49.6% seller swap as AI Mode "picking a different one on its own criteria" rather than re-ranking the same offer. For a dealer selling OEM parts against aftermarket sellers and marketplace resellers, the criteria question is the whole game.

Google's own report will not show you the problem. Merchant Center now has an AI performance insights report covering conversational shopping queries on AI Mode and AI Overviews, live for English queries in the US. Read its metric list closely: your share of voice, your competitors' average share, frequency, and products showing. There is no seller metric and no price metric anywhere in it. So if AI Mode is listing a competitor first on your part number at a higher price, the report Google hands you will register an impression and stay silent about the two facts that decide the sale.

And it settles an argument about vendor reporting. We have written before that a single AI visibility reading is mostly noise. This adds a cleaner version of the same point. If a vendor shows you a classic-surface ranking and calls it AI visibility, the overlap between those two things was measured at 1.28%. Ask which surface they sampled, and on what day.

What to do about it this week

1. Pull your 20 highest-margin parts and accessories SKUs. Run each one as a shopper would in classic search and in AI Mode, same day, same location. Write down who is listed first and at what price on each side. This takes an afternoon and it is the only reading of your own store that nobody can sell you.
2. On every one of those, click into the product panel. Confirm you are in the seller list at all, and at what price. Being absent from the panel is a different problem from being second in it.
3. Open Merchant Center, go to Analytics, then Products, then the AI performance tab. Note what it tells you and what it leaves out. Share of voice is calculated against the competitor set you define in Merchant Center, so a lazy competitor set produces a flattering number.
4. Before you cut price to chase AI Mode placement, finish the feed work. Richer attributes cost you no margin, and the price theory here is one founder's interpretation of his own data.
5. Ask your feed vendor which surface their reporting samples. If the answer is the carousel, you now know what that is worth as an AI Mode signal.
6. Test vehicle listings separately. Different feed, different add-on, different behavior, and no study has measured it the way this one measured products.

Google spent this year moving shoppers onto a surface that shows fewer options and picks them differently. The plumbing underneath is the same plumbing your store already feeds. What changed is that the number of slots collapsed and the selection logic went private, and Google's own reporting will not tell you who is standing in your slot.

If you want a read on how your dealership's inventory and content actually surface across AI answers, book a walkthrough and we will run it with you.