Last month I needed a specific vacuum filter. I typed the model number into Google, and the first thing on the page was a product card with a price. I almost bought it. Then I did what I always do out of habit from years of buying tech gear in Dhaka, where the price you see and the price you pay are rarely the same thing: I opened a second tab and searched again the old way. The same filter was about 20 percent cheaper on the classic results page.
I assumed I had caught a glitch. It turns out I had caught a pattern.
A research team ran a controlled version of my two-tab experiment at scale, for 23 days, across more than 2 million product listings. Their finding: when the exact same product appears in both Google AI Mode and traditional Google Search on the same day, the price shown in AI Mode is 21.6 percent higher on average.
That number deserves scrutiny, because it is easy to read it wrong. The headline is not "Google secretly marks up prices." Google does not set the prices. The real finding is more interesting and, for anyone who shops online, more actionable: the AI layer of search has a different idea of what you should see first, and that idea is expensive.
The study that measured both tabs
The data comes from Productrise, a company that tracks how products appear organically inside Google. Its founder, Hugo Huijer, published the study on September 1, 2026. The setup is simple to describe and hard to argue with.
- The team ran identical shopping queries through Google AI Mode and through traditional Google Search on the same calendar day, at the same moment.
- The window ran 23 days, from August 9 to August 31, 2026, across the United States and the United Kingdom.
- The dataset covers more than 2 million product listings across more than 100,000 results pages and AI Mode responses.
- Products were matched across the two surfaces using Google's own stable product identifier, so the comparison is same item to same item.
The study compared the "lead offer," the first price attached to a product on each surface. That distinction matters and I will come back to it, because it is the strongest defense Google could make and the study still stands without it.
Six findings, and the one that should worry you most
Finding 1: same product, higher price. For products that appeared on both surfaces for the same query on the same day, the lead price in AI Mode averaged 21.6 percent higher than in traditional search.
Finding 2: the whole shelf is pricier. Look beyond the matched products at everything each surface shows, and the median AI Mode product costs $149 against a median of $100 in traditional search. That is roughly 49 percent higher. AI Mode is not subtly nudging you. Its entire product mix sits in a more expensive tier.
Finding 3: the two surfaces barely overlap. Only 1.28 percent of products ranking in traditional search also appeared in AI Mode for the same search on the same day. This is the finding I keep thinking about. You are not looking at two views of one catalog. You are looking at two different catalogs, and the one you see depends entirely on which tab you are standing in.
Finding 4: when prices disagree, AI Mode is pricier two-thirds of the time. For matched products, the two surfaces showed the same price about 62 percent of the time. When they disagreed, AI Mode was the more expensive side in 68.4 percent of cases, with a median gap of 22.2 percent. When AI Mode happened to be cheaper, the gap was much smaller, a median of 7.8 percent. The asymmetry is the tell: the expensive direction is bigger than the cheap direction.
Finding 5: the seller changes half the time. On 49.6 percent of matched products, the main seller differed between the two surfaces. Same product, same query, same day, but a different merchant behind the lead price.
Finding 6: AI Mode shows far fewer options. An average of 3.9 products per AI Mode response against 27.8 in traditional search. Fewer choices, drawn from a pricier shelf.
Individually, each finding could have an innocent explanation. Together they describe something structural.
What is actually happening
The obvious accusation is that Google is deliberately hiding cheap options to make more money. The evidence does not support that, and making it would weaken the argument. What the data supports is stranger and, frankly, harder to fix.
Traditional product search is built on a straightforward ranking logic. Among listings with comparable quality signals, the cheapest offer wins the best placement. Google's own shopping results reward the low price with prominence. SEO consultant Brodie Clark, commenting on the study, noted the contrast: it is "often the cheapest product getting the most attention" in classic listings, and seeing AI Mode run 21.6 percent higher is "a stark contrast to the way we've seen Google rank products within product listings."
AI Mode does not appear to inherit that logic. A generative system assembles its answer from sources it selects, and price is evidently not the dominant factor in what it picks. The study's own observation is that AI Mode seems more likely to link to a brand's own website than to a third-party marketplace. Brand sites charge full price. Marketplaces compete on price. If your selection leans toward the former, your prices will look like the latter never existed.
There is also a coverage problem. AI Mode shows 3.9 products where classic search shows 27.8. A generative answer has to commit to a handful of options, and whichever handful it commits to, most of the market is invisible. If a $100 option makes the cut instead of a $79 option, nobody sees the $79. In a list of 28 products, the cheap one is right there on the page. In an AI answer of 4, it has to be one of the chosen.
And the routing is the multiplier. Google now puts AI Mode one tap away everywhere: a button in the search box, a tab above the results, even an icon in Chrome's address bar. The query carries over automatically, so from the shopper's point of view nothing changes when they cross into it. They do not know the catalog changed. They do not know the ranking logic changed. They just see a confident answer with a price on it, and the price is 21.6 percent higher than the one one tab away.
The strongest counterargument, and why it only half-works
Here is the fairest pushback. The 21.6 percent figure measures the lead price, the first offer displayed, not the cheapest price available for that product. Click any product on either surface and a panel opens listing other sellers, some cheaper. So a defender can say: the cheap options were always there, one click away.
That is true, and it does not hold up. The study's point is precisely about what happens before the click, which is most shoppers. The click-through to the seller panel is the exact behavior AI Mode exists to eliminate. If the answer to "AI Mode shows you a pricier first price" is "you should click through and check the panel anyway," then the convenience argument for AI Mode collapses into the five-tabs problem it was supposed to solve.
There is a second, subtler defense worth naming because an HN commenter documented it: classic search's rock-bottom prices are not always real. Some retailers show Googlebot a lower price than they charge humans, or advertise a low price and recover it at checkout with fees and failed coupon codes. If AI Mode is weighing "the price you will actually pay," higher displayed prices could partly reflect honest pricing. It is a plausible partial explanation for some gap. It cannot explain a systematic 21.6 percent average across 2 million listings, and it definitely cannot explain why AI Mode shows 3.9 products where classic search shows 27.8. Fewer options is not honesty. It is a narrower catalog.
Why this is different from an AI being wrong
When an AI chatbot gives you a wrong answer, you lose time. When a shopping AI shows you a systematically pricier catalog, you lose money, quietly, on every purchase, and you never see the invoice.
No single AI Mode response is wrong. Every price it shows you is real. The failure mode is compositional: a narrower catalog, selected by a logic that does not prioritize price, delivered through an interface that feels like an upgrade. This is why it will not generate an outrage cycle the way a hallucinated answer would. There is nothing to screenshot. There is no wrong answer, just an absence, and absences do not go viral.
But over a year of shopping, the difference between a $149 median and a $100 median is not a rounding error. It is the whole margin on most purchases.
What I do now, and what you can steal
My two-tab habit turned out to be accidentally correct, so it stays. Here is the practical version.
- Treat AI Mode as a research surface, not a checkout surface. Use it to understand what to buy, which model, which features matter. Do not buy from the first price it shows you.
- For anything over $50, run the same query on classic Google Shopping before paying. The study says 1.28 percent of classic results even appear in AI Mode. You are not double-checking a price. You are checking a catalog AI Mode never showed you.
- Check the seller panel, but do not trust the lead price. On half of matched products, the two surfaces picked a different lead seller. The first price is a recommendation, not a market.
- Watch condition and bundle details. The study caught outliers where classic search matched a used price against a new one in AI Mode. A dramatic gap usually means the two listings are not truly identical.
- For big purchases, add one non-Google price check. A marketplace search or a price tracker. If the AI layer is becoming the default front door, the escape hatch is a bookmark that does not pass through it.
The one-line rule: AI answers the question "what should I buy." Classic search answers the question "where is it cheapest." Keep using the first to decide and the second to pay.
The uncomfortable takeaway
Google will likely respond that AI Mode is early, that the Shopping Graph integration is improving, that prices are served live by merchants. All fair. None of it changes the mechanism: the moment a company controls both the question you ask and the catalog that answers it, the definition of "best result" quietly becomes negotiable. For twenty years the negotiation ran toward the cheapest option, because that was the user's interest encoded in the ranking. The AI layer does not inherit that encoding. Whatever values it does inherit, nobody voted for them and nobody published them.
The next time an AI search shows you a product and a price, remember the study's quietest number. The two surfaces shared only 1.28 percent of their products. The cheap one was probably on the other tab.
I write about AI, developer tools, and the systems shaping how we use technology every week. Subscribe, it is free.
Have you compared AI Mode prices against classic search yourself? Did you find a gap? Tell me in the comments.
Sources:
Top comments (0)