Agentic Shopping in Gemini and What It Means for Your Store

Google has begun testing a way for shoppers in India to buy Flipkart products without leaving Gemini or AI Mode, as reported by TechCrunch. A Buy button appears on selected listings and opens a Flipkart checkout flow inside the AI interface. For anyone running an online store, this is the moment where AI stops recommending products and starts handling the transaction.

What actually changed

The test is small. It covers a limited set of users and a narrow product selection — smartphones, electronics and mobile accessories. Other users still see ordinary Flipkart listings in Gemini and AI Mode with no direct purchase option. People familiar with the matter told TechCrunch that a wider rollout is planned for later in October, before India's festive shopping season.

A Google spokesperson said only that the company constantly tests new features to help people discover and connect with businesses, and had nothing further to share.

Two details are worth holding onto. First, the checkout in this test is Flipkart-branded, which appears different from the Google-hosted checkout the company demonstrated earlier this year with its Universal Commerce Protocol — an open standard meant to let AI agents interact with retailers across the shopping journey, including checkout. Google has since extended UCP so shoppers can also be handed over to a retailer's own site to finish buying. It is not clear which technology powers the Flipkart test.

Second, this is not open to everyone. In the experience TechCrunch saw, Amazon listings appeared next to Flipkart products but carried no direct buy option. Google also holds a minority stake in Flipkart after investing around $350 million in 2024. Early access here looks like a partnership, not a public feature.

Why it matters for a small business

You are not Flipkart, and you will not get a bespoke integration. But the direction is the thing to read, not the pilot.

When an AI assistant answers a shopping question, it picks a handful of products and presents them. If it can also complete the purchase, the assistant becomes a sales channel rather than a referral source. The stores that show up in that answer are the ones whose product data is machine-readable, current and unambiguous. The stores whose prices live only inside a rendered page, or whose stock counts are three days stale, are invisible to a system that has to decide what to recommend and whether it can be bought right now.

There is also a standards signal. Google framed UCP as open. Open standards eventually reach smaller merchants, usually through platforms and integrations rather than direct deals. The preparation work is the same either way, and it is work that pays off in your own store regardless of what Google ships next.

What to do about it

• Audit your product feed. Every item should carry a stable identifier, a clear title, brand, price with currency, availability and shipping terms. If your feed is generated by hand or exported once a month, that is the first thing to fix.
• Make stock and price live. An agent that sends a buyer to a sold-out product wastes the buyer's time and your ranking. Connect your catalog to whatever actually holds the truth — warehouse system, accounting, or the single spreadsheet everyone edits — and let it update automatically.
• Use structured data on product pages. Marking up price, availability and reviews in machine-readable form costs little and is what discovery systems read first.
• Test checkout from a cold start. Assume a buyer lands directly on a product page from an AI answer, with no browsing history and no cart. Can they pay in a few taps, on a phone, without creating an account? That is the handover path Google's newer UCP capability describes.
• Track where orders come from. If AI-sourced traffic starts arriving, you want to see it in your analytics and CRM rather than guess. Tag entry points and check monthly.

The calm read

One pilot, one retailer, one market, one category cluster, with an investor relationship in the background. Nobody should rebuild their business around it. But the underlying requirement — clean, current, structured product data that a machine can act on — is not speculative. It already determines how you appear in search, in marketplaces, in price comparison and in messaging channels. AI checkout just raises the cost of getting it wrong. Spend a week on your catalog data this month, and you will be ready for whatever version of this reaches your market.