Google has been trying to crack shopping for 20 years. Is this time different?

Google’s shopping history is a 20-year loop of the same experiment with different names.

Google’s commerce graveyard, 2002-2026

Sources: Google, TechCrunch, Digital Transactions, Marketplace Pulse, Chrome Unboxed. Compiled by Lengow.

Notice the pattern. Google never lost on technology. Checkout processed payments fine, Express delivered packages. Buy on Google had a working cart and, in its final form, charged merchants nothing at all. What Google could never do was change where people shop. Consumers had Amazon habits, saved cards, Prime subscriptions. Merchants had no reason to invest in a channel their customers weren’t using. Shopping passed through Google; it never happened on Google.

There’s a second pattern, less discussed. Every one of these products competed with Google’s own ads business for internal attention, and ads always won. Buy on Google’s zero-commission model wasn’t generosity, it was an admission that commerce would never be a profit center. The money was in the click, not the transaction. When the marketplace experiment threatened neither Amazon nor Google’s ad revenue, it was quietly retired.

The 2025–2026 stack is different in scope. Not one product, but an entire layer.

Agentic checkout started rolling out in November 2025 with Wayfair, Chewy, Quince and select Shopify merchants: track a price, and when it drops, Google buys the item for you on the merchant’s site with Google Pay. UCP followed at NRF in January, an open standard co-developed with Shopify, Etsy, Target, Wayfair and Walmart. Business Agent went live the same week with Lowe’s, Michaels, Poshmark and Reebok, putting a branded sales assistant directly inside Search results. At I/O in May, Google added Universal Cart, a cross-surface basket that began its US rollout on May 19, with YouTube and Gmail to follow. And in July, AI Mode started linking directly into apps like Instacart, moving from answering questions to completing tasks.

Underneath all of it sits the one asset none of Google’s previous attempts had at this scale: distribution at the moment of intent. AI Mode passed one billion monthly active users within a year of launch, with queries more than doubling every quarter. The Shopping Graph now holds over 60 billion product listings, two billion of them refreshed every hour. When someone types “lightweight carry-on under 2 kg for a weekend in Barcelona” into a conversation instead of a search bar, Google captures the full intent, not a keyword fragment.

Then came the political victory. On April 24, 2026, Amazon, Meta, Microsoft, Salesforce and Stripe joined the UCP Tech Council. Read that again: Amazon, which blocked AI shopping bots, sued Perplexity over its Comet agent and won in March 2026, sat down at Google’s table a month later. Meanwhile OpenAI, the loudest challenger, walked back native checkout in ChatGPT that same month after discovering that users browse but don’t buy. The protocol war looks, for now, decided.

Three things have changed since Buy on Google.

The incentive finally points the right way. Every previous attempt asked Google to build a commerce business against its ads business. This time, agentic commerce defends the ads business. If shopping conversations migrate to ChatGPT or Perplexity, Google loses its position at the top of the commercial funnel, and with it the most valuable ad inventory on earth. UCP isn’t a side bet anymore; it’s survival infrastructure. Google companies don’t kill survival infrastructure.

Google isn’t asking anyone to change where they shop. Checkout, Express and Buy on Google all demanded a new habit: come shop at Google. Agentic checkout does the opposite. The purchase completes on the merchant’s own site, with the merchant as seller of record, triggered from a surface a billion people already use. Google learned the lesson OpenAI just relearned the hard way: you can’t outbuild a habit, but you can automate around it.

The behavior is finally moving. Adobe measured 693% year-on-year growth in retail traffic from generative AI over the 2025 holiday season. That’s discovery, not yet transaction, but it’s the first time in twenty years the consumer side of the equation has moved toward Google’s thesis rather than away from it.

Now the uncomfortable part.

Nobody has shown that people complete purchases through AI at scale. OpenAI had 700 million weekly users and roughly a dozen active Shopify merchants when it pulled the plug. Traffic from AI sources still converts 40–60% worse than classic search traffic. Google has announced protocols, councils, carts and agents. It has not announced a single transaction figure. In e-commerce, that silence usually means something.

Merchants have long memories. The sellers Google is now courting for UCP are the same ones who onboarded to Shopping Actions, absorbed the integration cost, watched commissions get cut to zero in a last-ditch rescue, and then received a shutdown notice. Google’s credibility problem in commerce isn’t technical, it’s contractual: why invest in a Google channel when Google has closed four of them?

The regulators arrived before the customers did. UCP’s terms drew immediate antitrust criticism for price-parity language that resembles the most-favored-nation conduct at issue in the cases against Amazon, and this from a company operating under active search and ad-tech remedies. Senator Warren has publicly questioned the data implications of Google combining search intent, conversation context, offers and payments in one closed loop. Google’s biggest commerce push in history is happening at the moment its freedom to operate is most constrained.

And Amazon’s council seat is not Amazon’s catalog. Joining the Tech Council gives Amazon a vote on the rules, not an obligation to play by them. Its own agent, Rufus, remains the only one welcome inside its marketplace. The company that represents roughly 40% of US e-commerce is inside the room precisely so the protocol doesn’t get built against it.

Split the work in two, because Google has.

The discovery half is available now and gated by nobody. Google added six product attributes built for conversational surfaces: question_and_answer, related_product, document_link, item_group_title, variant_option and popularity_rank. They’re optional, which is precisely why most catalogues don’t carry them yet, and precisely why they’re worth doing. They give Gemini material to reason with instead of keywords to match against. Lengow supports all six natively in its Google feed structures, for both API and URL feeds, so mapping them in NetAmplify is a configuration change rather than a project. No approval queue, no waiting on Google.

The checkout half is gated, and heavily. Google Native Checkout is in early access: prepare your Merchant Center account to spec, submit through the UCP Integration Interest Form, complete the technical implementation, then wait for Google to approve you. What you get in return is the detail that separates 2026 from 2018. You remain Merchant of Record. Fulfilment is yours, payment configuration is yours, the customer relationship is yours. Buy on Google made Google the seller and merchants the supplier. UCP does the opposite, and that reversal is the single best argument that Google actually learned something from the graveyard.

Then the caveat nobody in Europe should skip. Universal Cart is in beta with Nike, Sephora, Target, Walmart and Wayfair, in the US. Canada and Australia follow, the UK after that, and Google has published no timeline at all for continental Europe. For a French, German or Spanish retailer, the checkout race hasn’t started yet. The discovery race has, and it’s running on the same product data you already send to Merchant Center every day.

Two things to do while you wait. Test your best sellers in AI Mode with the way a real shopper talks, not with keywords, and see whether you appear at all. And decide in advance how much of the customer relationship you’re prepared to concede: when the conversation, the recommendation, the offer and the payment all happen on Google’s surface, Merchant of Record status protects the transaction. It doesn’t protect the intimacy you used to get from having them on your site.

This is the best-positioned commerce attempt Google has ever made: right incentives, right distribution, weakened rivals, an industry coalition that even Amazon couldn’t afford to skip. If any version of Google shopping was ever going to work, it’s this one.

But nothing is won. The graveyard exists because Google repeatedly mistook infrastructure for adoption, and in 2026 it’s once again shipping infrastructure at a spectacular pace while the adoption question sits unanswered. The one metric that killed Checkout, Express and Buy on Google, and that just humbled OpenAI, is whether ordinary people actually complete purchases this way. Until Google publishes that number, treat every UCP milestone as what it is: a very impressive foundation for a building nobody has moved into yet. Get your product data ready for the tenants. Just don’t sign the lease on their behalf.