Google Gives Gemini a Buy Button, Just in Time for Nobody to Know Who Handles Returns

For most of the generative AI era, asking a chatbot what phone to buy has ended the same way: a confident paragraph, three bullet points, a link, and a human being who still has to go do the actual buying. Google has decided that last step is the problem. As of this week, some shoppers in India are seeing a “Buy” button on select Flipkart product listings inside Gemini and Google’s AI Mode. Tap it, and a Flipkart-branded checkout flow opens without you ever leaving the AI interface. The chatbot no longer merely recommends the phone. It sells you the phone.

The test, first reported by TechCrunch on September 26, is small by design: limited users, a narrow slice of products — smartphones, electronics, mobile accessories — and everyone else still seeing ordinary Flipkart listings with no button at all. Google’s official comment was that it is “always testing new features and experiences to help people discover and connect with businesses more easily,” which is the corporate equivalent of a shrug delivered while quietly building a cash register. Flipkart declined to comment. The broader rollout, per the reporting, is planned for later in October, ahead of India’s festive shopping season — the single most lucrative stretch of the year for Indian e-commerce, and therefore the obvious time to debut a checkout system whose fine print hasn’t been written in public yet.

Here is what we know about how it works: there is a button, and it leads to a checkout. Here is what we don’t know, per a helpfully itemized MediaNama rundown of the open questions: the commission structure, whether Google takes a cut, whether the checkout happens entirely inside Gemini or hands off to Flipkart, which payment methods are supported, who handles returns and refunds, who resolves disputes, and whether Google’s own Universal Commerce Protocol — the open standard it announced earlier this year specifically so AI agents could complete purchases — is even what’s powering this. That is a fairly complete list of the things a shopper might want to know before buying a phone, and it is, at the moment, entirely unanswered.

MediaNama also raises the most delightfully practical problem with letting an AI handle Indian e-commerce in particular. The price displayed on a listing often isn’t the price anyone actually pays, thanks to the layered thicket of bank-card offers, conditional coupons, and payment-method-specific discounts that define the festive sales. A human shopper spends twenty minutes working out that the phone is cheapest with one particular bank’s credit card on EMI. An AI agent optimizing for “buy the phone” may simply buy the phone. Convenience, in this case, has a very specific and possibly measurable cost.

Then there’s the question of who gets a button. Google invested roughly $350 million in Flipkart in 2024 for a minority stake. Flipkart is the retailer currently being tested with one-tap checkout inside Google’s AI products. Amazon listings, meanwhile, continue to appear in those same AI results — minus any Buy button. There may be a perfectly neutral technical explanation for this, and Google has not said otherwise. But when the first merchant to receive the frictionless checkout treatment inside the world’s largest search company’s AI happens to be one that the search company partly owns, it’s the kind of detail that competition regulators tend to circle in red pen before anyone else has finished reading the press coverage. MediaNama notes the fairness questions explicitly, pointing to existing competition disputes in India over how AI platforms treat marketplaces.

It’s also worth remembering that Google is not the first company to try turning a chatbot into a storefront. According to MediaNama, OpenAI charged merchants fees of up to 4 percent for its own in-chat checkout — up to 7 percent with payment fees — before scaling the feature back in March of this year, citing low merchant adoption and minimal actual purchasing through ChatGPT. The lesson there was not subtle: people were perfectly happy to ask a chatbot what to buy, and considerably less eager to let it do the buying. Google appears to have read that result and concluded that the problem was the timing, the retailer, and the country, rather than the premise.

None of this makes the idea bad. If the entire value of an AI assistant is saving you steps, the purchase is the most obvious step left to save, and India’s festive season is the most rational possible venue to test whether people will take the shortcut at scale. But the order of operations is revealing. The button exists. The rollout date exists. The festive shopping calendar certainly exists. What doesn’t yet exist, publicly, is the answer to who pays whom, who’s on the hook when the phone arrives broken, and why one retailer got the button first. For a product whose entire pitch is that it can do the research for you, it has left a remarkable amount of research for everyone else.

Sources: