Google tests Flipkart checkout inside Gemini and AI Mode
The limited pilot turns AI recommendations into a purchase path, but production teams should test the channel without surrendering checkout infrastructure to one platform.
What changed
Google is testing purchasing for a subset of users in India. Selected Flipkart listings in Gemini and AI Mode now show a Buy button that opens a Flipkart-branded checkout flow inside the AI interface.
The test covers a limited catalogue including smartphones, electronics and mobile accessories. Google had already named Flipkart as a partner for agentic shopping built around its Universal Commerce Protocol.
The new distribution layer
The assistant is no longer merely a recommendation channel. It is becoming an intermediary between user intent and a completed transaction, influencing both product visibility and which merchant receives the shortest path to purchase.
For production teams, that creates a new dependency. The platform can change its interface, attribution rules or integration requirements while the merchant still owns pricing, inventory, payment failures, returns and customer support.
What we would do
Virtual Arc would build one stable commerce API and keep the Gemini integration behind a replaceable adapter. Order creation should be idempotent, price and availability should be revalidated, and every step should expose enough telemetry to measure completed purchases and operational cost.
We would not wait for agentic commerce to become mature, but we would not build a roadmap around this pilot either. We would deepen the integration only if it lowers cost per completed purchase without increasing failures, disputes or support work.
The important change is not that Gemini can recommend a phone; it is that the assistant is beginning to control the route from intent to payment. Once an AI interface decides which products appear and which merchants receive a direct purchase path, distribution becomes as consequential as model quality. Virtual Arc would not rush to build a bespoke Gemini storefront or bind core commerce logic to a protocol that is still settling. We would expose a vendor-neutral layer for catalogue data, live availability, pricing, authorisation and idempotent order creation, then place a thin adapter in front of this channel. We would judge the pilot by completed purchases, retries, abandonment, support load and reconciliation failures—not conversational engagement. This is worth testing now, but it is not yet worth surrendering ownership of checkout, attribution and customer access to a platform whose interface and commercial terms can change independently of our release cycle.