Google Native Checkout Turned On Without You

Most infrastructure decisions arrive as a project. Somebody writes a brief, somebody argues about it, somebody signs off, and six weeks later the thing exists. The decision about whether your customers can buy from you without ever loading your site arrived as an email telling a lot of merchants it had already happened.

On September 22, 2026, Barry Schwartz at Search Engine Roundtable reported that Google had started sending Merchant Center notifications with the subject line that your products are now eligible for native checkout. The body explained that a matched Shopify store had enabled native checkout on Google AI Mode and Gemini. Then the line that matters: eligible products are automatically included, and there is nothing for you to set up.

Read that as a product announcement and it is unremarkable. Read it as a governance event and it is the most consequential thing that happened to ecommerce merchandising this month. A decision about where the transaction completes, which payment rails carry it, and what data comes back to you was made by a catalog match between two systems you already use.

Those four numbers are the whole argument in miniature. A large default, no friction, a narrow geography, and a recent precedent that did not work. None of them tells you to opt out. Together they tell you this is a unit economics question with a deadline attached, and the deadline is the fourth quarter.

What Google native checkout actually changed this week

Google native checkout lets a shopper complete a purchase on a Google surface, specifically AI Mode in Search and Gemini, while the merchant remains the seller of record. Payment runs through Google Pay using cards and addresses already saved in Google Wallet.

It sits on the Universal Commerce Protocol, an open standard Shopify and Google built together for agent-mediated buying. The backer list is the part people underrate: Amazon, American Express, Etsy, Mastercard, Meta, Microsoft, Salesforce, Stripe, Target, Visa and Walmart. Protocols with that much of the payments industry attached do not usually disappear quietly, whatever happens to any single implementation of them.

Look down the middle column and the right column together. Four rows are unchanged, two rows are narrowed, and two rows are genuinely new. That is a much smaller change than the discourse suggests and a much larger one than the notification email suggests, which is the usual position for anything Google ships into a default.

The default flipped at the platform layer, not in your settings

Two accounts of how a merchant gets into native checkout are both true at once, and the gap between them explains why so many teams were surprised by an email about a feature they had never evaluated.

Google's public Merchant Center documentation describes a deliberate path: meet the requirements, submit an interest form, complete a technical implementation, and tag the individual product listings with the native commerce checkout eligibility attribute that makes the buy button appear. That reads like an opt-in, because it is one.

The Shopify path is different. Shopify built the protocol alongside Google and ships an agentic sales channel in the admin, so when a Shopify catalog is matched to a Merchant Center account, the plumbing is already there. The merchant did not fill in a form. The platform did the integration once, on behalf of everybody on it.

None of this is a scandal. Shopify shipped a capability its merchants broadly want, and Google documented the manual route for everyone not on a platform that did the work for them. The failure mode is quieter than a scandal: a material change to the buying path that never gets discussed, because at no point did anybody have to say yes to it.

The last agentic checkout cycle already failed once

In-chat checkout has a track record, and it is twelve months old. OpenAI launched Instant Checkout in ChatGPT in September 2025 with Etsy sellers first and a stated ambition to reach over a million Shopify merchants. It was retired in March 2026 and folded into an apps-based approach.

Forrester principal analyst Emily Pfeiffer put the real adoption number at roughly thirty live Shopify merchants as of February 2026, which she characterised as a rounding error against what had been promised. The reported reasons are mundane rather than philosophical: onboarding merchants was harder than expected, product data was often wrong, multi-item carts and loyalty memberships were not solved, and as of February the system still had no mechanism for collecting and remitting United States state sales taxes.

“The first agentic checkout did not lose an argument about consumer preference. It lost to catalog accuracy, tax remittance and cart mechanics, which is a much more solvable set of problems.”

That last point is why the failure should not be read as a verdict. Every obstacle on that list is the kind of thing a protocol with Stripe, Visa, Mastercard and the major retailers attached is specifically built to solve. UCP exists because the first attempt demonstrated exactly which plumbing was missing. Predicting the second attempt from the first one's results is the same error as writing off mobile commerce in 2010.

The honest position is that the demand side is unproven and the supply side is now serious. A Semrush survey of 1,030 United States shoppers in December 2025 found 22% had ever completed a purchase directly inside an AI tool, while 50% had bought something after using AI to research it. Both numbers are self-reported and a year old at this point. Read together, they describe a population that trusts AI to shortlist far more readily than it trusts AI to transact.

The conversion penalty and the acquisition premium point opposite ways

Walmart published the most useful operator data on in-chat buying so far, and it cuts in two directions at once, which is why partisans on both sides of the argument tend to quote half of it.

A conversion rate one third of your baseline is terrible. A new-customer rate twice what search delivers is excellent. Both were measured by the same retailer on the same surface, and any recommendation that only mentions one of them is selling you something.

The reconciliation is not complicated. Agentic surfaces are behaving like a top-of-funnel acquisition channel that happens to have a buy button on it, not like a bottom-of-funnel replacement for your own checkout. If you price an AI-mediated sale using your site conversion rate you will conclude the channel is broken. If you price it using blended customer acquisition cost against first-order margin on genuinely incremental customers, you get a different answer, possibly a much better one.

Which answer applies to you depends almost entirely on repeat purchase economics. A brand whose lifetime value is concentrated in orders two through six cannot afford a channel that acquires a customer without reliably handing over an email address. A brand selling a considered one-time purchase can. That is a merchandising and finance conversation, and it is being resolved in most companies right now by nobody having it.

How to decide whether to leave Google native checkout on

The decision is not binary and it is not urgent in the way the notification implies, but it does have a natural deadline. Peak trading season is the wrong time to discover that a channel you never evaluated is now a meaningful share of orders.

Notice what is not on that list. There is no recommendation to opt out on principle, and no recommendation to lean in because the future is agentic. Both of those are postures, and postures are what people adopt when they have not done the arithmetic.

Instrument it before Q4, not after

Measurement is the part that will hurt, because an order that completes on a Google surface arrives in your systems without the session that normally explains it. Your analytics will show revenue with no journey attached, and the instinct will be to file it under direct.

That is the same class of problem as AI-referred traffic generally, where sessions arrive stripped of the context that made them, and it corrupts the bid signals and channel reports built on top. We worked through the mechanics in how AI traffic distorts attribution, and agentic orders make it sharper, because here the missing journey is not a visit you failed to attribute, it is a visit that never happened.

Three things to have in place before peak. First, a separate order tag or channel label for agentic orders, so they never silently join direct. Second, a baseline count from the four weeks before you make any change, because without one you cannot tell cannibalisation from incrementality later. Third, a repeat-rate cohort for agentic-acquired customers, tracked against your site-acquired baseline, which is the single number that will settle the strategy argument inside two quarters.

Pair that with the exposure side. The exposure data Google now reports for generative surfaces tells you how often you appear inside AI Mode, and an agentic order count tells you what that appearance converts into. Neither number means much alone. Together they are the first real read anyone has had on whether presence inside an AI surface is worth what it costs to earn.

There is also a competitive angle that gets missed. If your products are eligible and a rival's are not, you are the one with a buy button inside the answer. We saw a version of this dynamic in our enterprise print and publishing work, where the merchants who made their catalogs machine-readable early spent the following year being the default option rather than the alternative to it. Defaults compound.

Questions merchants are asking this week

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.