Agentic Commerce in India (2026): A Store-Readiness Checklist for AI Shopping Agents

Separate current availability from future readiness

Use three labels in planning. Assisted commerce means an AI helps research, compare or prepare an action while the customer approves the transaction. Agent-initiated commerce means an authorized agent can start or complete defined steps through merchant and payment interfaces. Fully autonomous commerce allows actions within pre-agreed limits and needs stronger controls. By September 2026, assisted commerce is live and several transaction protocols have production or pilot implementations, but access is not universal. Google’s Merchant Center documentation describes its UCP onboarding experience as an early-access program for products sold in the United States, while Google’s India update focuses on AI-surface performance insights and conversational product attributes. Check current scope at https://developers.google.com/merchant/ucp/guides/tools/merchant-center/overview and https://blog.google/intl/en-in/products/google-companies/google-marketing-live-2026-delivering-the-gemini-advantage-for-indian-businesses/. For ChatGPT, OpenAI’s Instant Checkout launch began with U.S. merchants and links participation to platform onboarding: https://openai.com/index/buy-it-in-chatgpt/. Treat those as dated availability facts, not a promise of immediate India checkout access.

Map discovery, decision, transaction and post-purchase protocols

Design around layers instead of betting on one acronym. Discovery needs crawlable pages, feeds, explicit attributes, availability, price and evidence. Decision needs comparison-ready facts, policies and trust signals. Transaction needs checkout state, fulfillment options, taxes, authorization, payment and order creation. Post-purchase needs status, cancellation, return, refund and support events. Google’s Universal Commerce Protocol is an open language spanning discovery through post-purchase and is compatible with AP2, A2A and MCP; its guide covers checkout, cart, catalog and identity capabilities: https://developers.google.com/merchant/ucp and https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/. The OpenAI–Stripe Agentic Commerce Protocol defines seller-operated checkout sessions and delegated payment flows while the merchant remains merchant of record: https://www.agenticcommerce.dev/docs and https://www.agenticcommerce.dev/docs/getting-started/sellers. Google’s Agent Payments Protocol addresses authorization, authenticity and accountability for agent-led payments across payment types: https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol. These standards overlap but solve different layers; implement only against documented partner requirements.

Make catalog truth and commerce APIs authoritative

Create one source of truth for product ID, title, brand, description, category, variants, dimensions, material or specification, price, tax treatment, availability, seller, image, rating provenance, delivery promise, return rule and warranty. Reconcile the site, structured data, Merchant Center, marketplaces and support content on a scheduled basis; log freshness and exceptions. An agent-ready transaction service must quote authoritative totals, reserve or verify stock, expose eligible fulfillment options, validate serviceable locations, handle promotions deterministically, return actionable errors and prevent duplicate orders with idempotency. Order and support systems must publish state changes rather than leaving an agent with a stale success message. Test create, update, complete, cancel, refund and expiry paths, plus out-of-stock, address, tax, payment-decline and partial-fulfillment cases. Product structured data improves machine readability, but it cannot repair contradictory inventory or policies. Start with a small product subset and versioned data contract. Do not publish guessed endpoints or fake protocol support; joining a platform usually also requires enrolment, validation and production approval.

Control identity, authorization, payment and agent access

Every action needs a clear principal, scope and audit trail: which person or organization authorized which agent, for which merchant, items, amount, currency, time window and permitted fallback. Minimize data sharing; separate browsing context from checkout data and never place payment credentials in logs or model prompts. Use tokenized, constrained credentials and your payment provider’s approved flow. Stripe describes Shared Payment Tokens as merchant-, time- and amount-scoped credentials that avoid exposing underlying payment details: https://stripe.com/blog/introducing-our-agentic-commerce-solutions. Visa’s Trusted Agent Protocol focuses on recognized agents and visibility into consumer intent: https://corporate.visa.com/en/sites/visa-perspectives/newsroom/visa-unveils-trusted-agent-protocol-for-ai-commerce.html. Publish an agent-access policy that distinguishes public catalog retrieval, authenticated account actions and transaction endpoints; robots.txt alone is not authorization. Add rate limits, signed requests or tokens, replay protection, key rotation, fraud controls, consent records and incident revocation. Require human intervention for ambiguous intent, unsupported customization, age or legal restrictions, high-risk payments, repeated failures, disputes and exceptions the agent cannot resolve safely.

Use a 20-point checklist and a 90-day roadmap

Score each item 0, 1 or 2: product IDs are stable; required attributes are complete; price and stock have freshness targets; variants map correctly; structured data matches visible content; delivery promises are computable; returns and cancellation rules are machine-readable; reviews have provenance; feeds reconcile across channels; checkout totals are authoritative; stock is verified before completion; create/update/complete/cancel paths are idempotent; order status is event-driven; refunds and returns are represented; user authority is scoped; payment credentials are tokenized; agents and requests can be authenticated; logs support audit without leaking secrets; human escalation is defined; platform and India availability is reviewed quarterly. A score below 25/40 means fix foundations before protocol work; 25–32 supports a limited sandbox; above 32 can justify partner-specific validation, subject to security and payment review. Days 1–30: data, policy, security and event audit. Days 31–60: build a sandbox for a narrow SKU set and failure cases. Days 61–90: validate with one eligible partner, run red-team and reconciliation tests, define kill switches and document ownership. Request a store-specific assessment at /free-plan and pair discovery work with /blog/geo-d2c-win-ai-product-recommendations-india.

Frequently asked questions

Is agentic commerce available in India today?

Yes for research and assisted shopping, and some global platforms support selected transaction flows. But direct checkout availability depends on the platform, merchant programme, product market, payments and current enrolment. Google’s published UCP Merchant Center pilot documentation currently describes U.S.-sold products, while its India announcements emphasize AI visibility and richer attributes. Indian teams should build readiness and verify eligibility rather than claiming unsupported local checkout access.

What is the difference between UCP, ACP and AP2?

UCP is Google’s broader commerce language across discovery, cart, checkout and post-purchase capabilities. ACP is the OpenAI–Stripe open specification for programmatic seller checkout and delegated payment. AP2 is Google’s payment-focused protocol for proving authorization, authenticity and accountability. They can coexist; choose based on the agent platform, payment provider and merchant programme you actually support.

Does robots.txt make a store agent-ready?

No. Robots.txt can express crawl preferences, but it does not authenticate an agent, grant checkout authority, prove user consent or secure a payment. Use separate public and authenticated interfaces, explicit capabilities, request authentication, rate limits, scoped authorization and auditable transaction controls. Keep human browser journeys working as the fallback.

What should an Indian merchant do first?

First reconcile product IDs, attributes, price, stock, delivery and return truth across the site and feeds. Then map checkout and order events, test failure and cancellation paths, document user authority and payment controls, and create a human escalation route. Only after that should the team evaluate a UCP, ACP or other partner-specific sandbox and confirm current India eligibility.