A customer may soon start a purchase in an AI conversation and finish it in a retailer’s checkout without ever visiting the retailer’s homepage. TechCrunch reported on September 27 that Google is testing a Buy button on selected Flipkart listings inside Gemini and AI Mode in India. It opens a Flipkart-branded checkout within the AI interface. Before building an AI shopping agent, retailers have a more immediate question: will their product information, price, delivery terms, and checkout still agree when a buyer crosses that boundary?
TL;DR
Start with one product category. Compare what an AI shopping surface says about those products with what your own checkout actually offers. Correct mismatches in price, availability, variants, shipping, returns, and product descriptions. Measure where buyers abandon the handoff before paying for a new AI commerce integration.
This is a test, not a new universal checkout
The scope matters. TechCrunch saw the Buy button on a small selection of products, including phones and accessories, for some users in India. Other shoppers still saw ordinary Flipkart listings. Listings from competing retailers appeared in the same AI experience without the direct purchase option. Google confirmed it tests new shopping experiences but did not provide details of this specific rollout. A reported plan to expand later in October is not a general availability announcement.
TechCrunch could not establish what technology powers this particular test. Google has separately described a broader purchase system for retailers, but the Flipkart screen cannot tell you how your own store would integrate or what your fees and obligations would be.
What the test does show is a change in the shopper’s path. Search used to send someone to a product page where the retailer controlled the sequence: description, comparison, shipping promise, cart, payment. In this test, discovery and the decision to press Buy happen on Google’s surface. The retailer’s checkout remains, but it arrives later in the conversation. By then, the customer has formed an expectation. An incomplete listing is harder to correct.
A checkout cannot repair a recommendation that described the wrong product.
The unglamorous work comes first
Imagine a customer asking for a phone that arrives before a trip. The AI shows a listing with the right storage size and a plausible price. The buyer presses Buy, then discovers that the selected color is back-ordered, the delivery date depends on their postal code, or the discount applies only to a loyalty member. That is an illustrative scenario, not a reported Flipkart failure. It names the ordinary retail facts that must survive the move from answer to order.
Google’s September 16 retailer guidance puts product-feed accuracy ahead of the fancier shopping features. Google says merchants adopting its core Merchant Center feed practices see, on average, a 5% increase in conversions the following month. That is Google’s reported average for merchants using those practices, not a prediction for any individual store or a result from the Flipkart pilot.
Google also reports that product details supplied by lululemon as conversational attributes were incorporated in 50% of relevant AI Mode recommendations during its test. That number does not tell us whether the recommendations converted. It does tell us why a retailer should care about the information supplied upstream. If material, fit, compatibility, or intended use is absent or wrong, the AI has less reliable detail to use when a buyer asks a specific question.
Start smaller than an AI-channel strategy. Take twenty products in a category where shoppers ask detailed questions. Compare the information in your product feed, the product page, and the final checkout. Can a buyer determine the correct variant, total price, fulfillment window, return terms, and stock status at each step? Record the discrepancies. Assign someone to correct the underlying source, not just the wording on a campaign page.
Own the point where expectations become obligations
A Buy button makes a promise feel immediate. Someone in the business needs to own the seam between what a third-party AI surface presents and what the merchant can fulfill. For a small retailer, that may be the person already responsible for catalog accuracy and order operations. For a larger one, commerce, merchandising, and customer support will need a shared exception process. The owner matters more than the org chart.
Set up a weekly review of a few real purchase journeys. Ask where shoppers first see an item, what terms they are shown, which checkout they enter, and what changes before payment. Save screenshots and order details when something diverges. Track abandoned checkouts and support contacts that mention a different price, product, or delivery promise. Those signals will be more useful than counting how often the company is mentioned in an AI answer.
Google says its broader retail purchase integration already supports direct checkout for hundreds of thousands of brands and retailers and is adding cart transfer to merchant sites and more checkout testing. The new capabilities are rolling out gradually in the United States, with Australia and Canada to follow early next year, according to its September guidance. That is a separate program with its own eligibility and geography, not evidence that the Flipkart test is open to every retailer.
There is a temptation to buy infrastructure because the Buy button moved. Resist it until you know which handoff is broken. If the catalog is accurate but buyers leave at payment, work on checkout. If checkout works but AI surfaces describe the wrong product, fix the source information. If the terms differ between channels, decide which system is authoritative and who updates the rest.
The retailer may never control the AI conversation. But the answer that brought the buyer to the button still needs to match the order the retailer can fulfill.
Research and structure: Mai. Direction and voice: John Lipe.