Agentic Commerce in India: How SMBs Get Their Products Sold by AI Shopping Agents in 2026

A new kind of shopper arrived in India in 2026: the AI agent. NPCI and Razorpay partnered with OpenAI so ChatGPT can complete UPI purchases, with Tata's BigBasket the first Indian merchant to enable conversational checkout (BW Disrupt, 2025). When an assistant does the buying, your marketing copy stops mattering and your product data starts mattering — agents read feeds, not adjectives.

This is still a small channel, but it is the fastest-growing one in commerce. Here is how big it is, how agents actually choose products, and the seven steps an Indian SMB should take this quarter to be sellable by machines.

What exactly is agentic commerce, and why did it arrive in India in 2026?

Agentic commerce is when an AI assistant — ChatGPT, Gemini and the like — does the shopping for a customer, from finding a product to placing the order. India got there fast because of UPI: NPCI and Razorpay enabled ChatGPT to complete UPI payments in 2025, letting an agent pay on a shopper's behalf on a public payment network (BW Disrupt, 2025). Most markets are still stitching together checkout; India's rails were already there.

How big is the AI shopping channel right now, and how fast is it growing?

Small base, steep curve. AI referral traffic to retail sites grew 393% year-on-year in Q1 2026, and AI-referred shoppers now convert roughly 42% better than non-AI traffic — a reversal from a year earlier when they converted worse (Elogic Commerce, 2026). Looking out, McKinsey estimates agentic commerce could orchestrate up to $1 trillion in US retail revenue by 2030 and Gartner projects 20% of digital commerce transactions running through AI agents by then (commercetools, 2026).

How do AI agents actually find and choose which products to buy?

Agents read structured data — feeds, schema, prices, stock status — not marketing prose. Research shows 71% of pages ChatGPT cites include structured data (Elogic Commerce, 2026). So the question shifts from "is my copy persuasive?" to "can a machine parse my catalogue, trust the price, and confirm it's in stock?" Clean product data is now the entry ticket.

What does a machine-readable product feed need to include?

At minimum: a unique product ID, precise title, current price and currency, live availability, GTIN/brand where relevant, and Product schema marked up on your product pages. Vague titles, stale prices and missing stock flags are what get a product skipped. If your site was built for humans only, this is the gap to close first — our e-commerce build service ships product schema and clean feeds as standard.

How do UPI, ONDC, and Razorpay make agentic checkout work in India?

India's edge is native payment. Using UPI Circle and UPI Reserve Pay, agents can pay directly on a public network, and by early 2026 Razorpay, Cashfree and PayU had all shipped LLM-native UPI payments (Stellagent, 2026). That closes the checkout gap that slowed agentic commerce elsewhere — the agent doesn't need a saved card on a walled platform, it uses the same UPI rails your customers already trust.

Should you sell inside ChatGPT or redirect shoppers to your own site?

The 2026 consensus: AI handles discovery, merchants keep checkout. OpenAI itself pivoted toward merchant-hosted checkout in early 2026 after Walmart saw roughly 3x lower conversion for in-chat purchases versus redirecting to its own site (commercetools, 2026). Let agents surface your product, but bring the shopper home to complete the purchase where you control the experience and the data.

What are the 7 practical steps an Indian SMB should take this quarter?

One, add Product schema to every product page. Two, publish a clean, current product feed. Three, keep prices and stock status live, not stale. Four, enable a UPI-native checkout via Razorpay, Cashfree or PayU. Five, list where your category buyers already are, including ONDC. Six, keep checkout on your own storefront. Seven, measure AI-referred traffic separately so you can see the channel grow. Once your data is machine-readable, layer in demand capture with SEO and performance marketing so human and agent shoppers both find you. For the search-answer side of this shift, see our answer-engine optimization playbook.