Google Just Let Gemini Buy From Flipkart Directly. Your Product Catalog Isn't Ready.

On 28 September 2026, Google quietly added a "Buy" button to a handful of Flipkart listings inside Gemini and AI Mode. Weeks earlier, Reuters reported that India's National Payments Corporation is building a framework — the Unified Agent Protocol — to let AI agents make small UPI payments without a human approving every transaction. Neither development is live at scale; both are early enough to watch rather than panic about. But the direction is unambiguous: a meaningful slice of future purchases in India will not start with a customer typing into Google and clicking through to your website. It will start with a customer asking an AI agent to find and buy something — and the agent deciding, on its own, whether your brand is even eligible to be bought from.

What Actually Happened in September

According to MediaNama's reporting on 28 September 2026, Google is running a limited test that lets select Indian users complete a Flipkart purchase directly inside Gemini and AI Mode, via a "Buy" button on certain listings. Amazon listings appear alongside Flipkart's but currently lack the direct-purchase option. MediaNama says the feature could expand around Big Billion Days on 9 October — and flags that Google hasn't disclosed the payment mechanism, commercial terms with Flipkart, or how returns and disputes get resolved. That list of unknowns matters as much as the headline.

Around the same time, Reuters reported on 1 September 2026 that India's National Payments Corporation (NPCI) is preparing a "Unified Agent Protocol" letting AI agents execute small, low-value UPI payments — groceries first — without transaction-by-transaction approval, leaning on existing UPI Circle and Reserve Pay features with rule-based spending limits. Deccan Chronicle confirms Mastercard and Visa are building comparable capabilities separately. UPI is no small testbed: it processed roughly 24.51 billion transactions worth about $314 billion (~₹26 lakh crore at ₹83/$) in August 2026 alone.

Both moves sit inside a bigger, slower shift. Google's global Agent Payments Protocol (AP2), launched September 2025 with over 60 merchants and banks, is the likely standard behind this Flipkart test — yet AP2's own documentation lists UPI only as a future expansion, not a current one. In the US, Amazon's $50 billion investment in OpenAI (~₹4.1 lakh crore), announced February 2026, came with OpenAI pulling Instant Checkout from ChatGPT in favour of Amazon — proof that even the biggest platforms are still fighting over who owns checkout, not just the recommendation.

Why This Should Matter Beyond Flipkart

The mechanics will keep changing. The structural point will not: once an agent can check stock, compare price, and complete a purchase without a human scrolling your website, your product data — not your homepage — becomes the storefront. A brand with great photography but no machine-readable price, stock or spec data is invisible to an agent the same way a brand with no AI-citable content is invisible to a GEO recommendation. Two sides of one requirement, which is why we treat agent-readiness as an extension of AI search visibility work, not a separate project.

Hynova's Read: Get Agent-Ready, Don't Rush to Build Agent Checkout

This is not a call to bolt an AI checkout widget onto your site this quarter. The commercial terms, fees and liability rules for agent-led purchases in India aren't settled — Google hasn't published them for the Flipkart test, and NPCI's protocol is still a proposal. Building custom agent-checkout infrastructure against rules that don't exist yet is the kind of premature build we'd tell any D2C or retail client to avoid. The realistic move: make sure the data an agent would need to recommend and transact with you is already correct, structured and current, so you're ready once the rails stabilise — and better placed in GEO results meanwhile, regardless of how checkout evolves.

What to Build Now vs. What to Avoid

Must-have, starting now

• Accurate product schema (Product, Offer, AggregateRating) on every live SKU — price, availability and specs an agent can parse without guessing.
• Real-time stock status on the page itself, not just in a backend system; an agent recommending an out-of-stock item damages trust immediately.
• One canonical product page per SKU, so there's a single source an agent or AI answer can cite.

Should-have, once the above is solid

• Automated catalog-to-schema sync, so pricing and stock propagate without manual edits — a workflow we scope under AI automation, not one-off dev work.
• A ratings pipeline that keeps AggregateRating data current, since AI engines weight recency.

Could-have, once terms are public

• Direct integration with a named agent-checkout protocol (AP2, or NPCI's Unified Agent Protocol), once fee structures, liability rules and dispute processes are actually published.

Avoid for now

• Building custom agent-checkout infrastructure against protocols that are still pilots or proposals. There is nothing stable to build against yet, and MediaNama's own reporting lists the payment mechanism and commercial terms as unresolved even for the Flipkart test.

The One Live Example Right Now

The clearest case study today is also the most incomplete one: Google's Flipkart test itself. It proves agent-led checkout is technically possible at a major India retailer, timed for peak festive demand. It also proves the gap between "technically live" and "commercially ready" — no public fee structure, no stated return policy for agent-purchased items, no sign of when it opens beyond Flipkart. Treat it as a leading indicator, not a template. The practical first step for most brands is simpler: find out whether AI engines already see your catalog correctly, with a free AI Growth Scorecard, before worrying whether they can buy from you.

The Takeaway

Agentic commerce in India is a real, fast-moving signal and worth tracking closely through this festive quarter. It is not yet a reason to rebuild your checkout, and it is not yet proven to move revenue for most categories beyond low-value, frequent D2C and retail purchases. What it is a reason to do, immediately and cheaply, is get your product data agent-ready: clean schema, live stock, one canonical page per SKU. That work pays off in GEO visibility today and keeps you eligible for whatever agent-checkout standard actually stabilises in India over the next 12–18 months.

Frequently Asked Questions

Can AI agents already buy products on my customers' behalf in India?

Only in a narrow pilot so far. Google began testing a direct "Buy" button for select Flipkart listings inside Gemini and AI Mode in late September 2026, timed close to Flipkart's Big Billion Days sale on 9 October. Google has not disclosed payment mechanics, commercial terms with Flipkart, or how returns and disputes will be handled, and the test does not yet extend to other retailers. This is a limited experiment, not a live, generally available feature.

Do I need to rebuild my website before AI shopping agents arrive?

No. The more urgent, lower-cost step is making your existing product data machine-readable: accurate Product, Offer and AggregateRating schema, live inventory status, and clean canonical product pages. This is largely the same structured-data groundwork that improves whether ChatGPT, Gemini and Perplexity cite you at all, so it pays off in AI search visibility now, whether or not agent checkout becomes common in your category.

Is this relevant for high-ticket B2B brands, or only D2C and retail?

Right now it is mostly a D2C and retail story. India's proposed Unified Agent Protocol is explicitly aimed at low-value, frequent purchases like groceries routed through UPI, not high-ticket B2B or real estate decisions. High-ticket buyers will keep relying on GEO-style AI recommendations rather than agent-executed checkout for the foreseeable future. But every brand with a product catalog should start the structured-data work now, because it is the same standard that determines whether AI engines can see you at all.