Google is testing a way to buy things from Flipkart inside Gemini. Google is also handing you off to Flipkart’s checkout to actually pay. Both of those things are true at the same time, and the gap between them is the whole story.
The facts, as confirmed: Google has begun a limited test in India that lets people shop Flipkart products directly from Gemini and from AI Mode in Search. The test covers select products and select users. When it comes time to pay, you land on Flipkart’s checkout, not a Google-hosted one. A wider rollout is planned for later in October.
That’s it. That’s the feature. And I’ve reviewed enough AI tools to tell you that the checkout detail is not a footnote — it’s the review.
Why the checkout choice matters more than the demo
Every agentic shopping demo I’ve tested follows the same arc. The AI understands you beautifully. It suggests three products that are roughly right. Then you hit the part where money moves, and the whole thing either falls apart or quietly punts you back to a normal website.
Google is punting on purpose here, and I’d argue that’s the smart call. Hosting checkout means owning payment failures, refund disputes, fraud rules, and India’s payment stack in all its specificity. Flipkart already does all of that. Routing to Flipkart’s checkout means the hard, boring, regulated part stays with the company that has spent years building it.
But it also caps what this feature can be. If you’re imagining a fully agentic flow where you say “get me a charger under 800 rupees” and it just happens, that’s not what’s being tested. What’s being tested is discovery and intent inside Gemini, with the transaction handled elsewhere. That’s a shorter path to the same store, not a new way to buy.
What I’d actually be watching for
Since this is a limited test, most of what people want to know isn’t public yet. Here’s what would tell me whether this is useful or just a nicer link:
- Does the handoff preserve context? If Gemini recommends a specific variant — a color, a size, a storage tier — and the Flipkart checkout loads with the right item already in the cart, that’s real value. If it dumps you on a search results page, it’s a bookmark with extra steps.
- How honest is the product selection? An AI that recommends products is making editorial choices. Whether those choices reflect what’s actually good or what’s positioned well inside a catalog is the question nobody answers in a launch post.
- What happens when it’s wrong? Recommendation mistakes are cheap in a chat window and expensive when they turn into a delivered box. Who owns that error when the recommendation came from Google and the sale came from Flipkart?
- Does it survive a wider audience? Curated tests with select products and select users look great. October’s broader rollout is the actual test.
The standards angle
There’s context worth holding onto here. Google has talked about an open standard for letting AI agents interact with retailers across the shopping journey, checkout included. The Flipkart test reads like an early, cautious instance of that idea — agent-assisted discovery on Google’s side, transaction infrastructure on the retailer’s side.
If that pattern holds, the future of AI shopping isn’t one assistant that owns everything. It’s a set of handoffs between an assistant and merchants who keep their own payment rails. Less dramatic than the demos, far more likely to work.
My read as a reviewer
I’m mildly positive, with the enthusiasm dial set low. A test in one market, with a subset of products and users, where the money still moves through the retailer’s existing system, is a sensible experiment rather than a new category. Google gets to learn how people shop inside a chat interface without taking on payment liability. Flipkart gets traffic with intent attached. Walmart gets a look at what agentic retail actually does to conversion.
The thing I’ll push back on is the framing that keeps showing up around features like this — the suggestion that shopping has been reinvented. It hasn’t. A search box got conversational and learned to pass you along more precisely. That’s an improvement. It’s not a different world.
The version worth reviewing properly arrives with October’s rollout, when the guardrails come off and ordinary users start asking Gemini for things that don’t map neatly to a product page. Until then, treat the current test as a signal about architecture, not a verdict on the experience. Google picked the cautious integration. For a first attempt at putting a store inside an assistant, cautious is the right answer.
🕒 Published: