Agentic Commerce Protocol: ChatGPT's Game-Changer for E‑Commerce Marketers & Developers

ACP is an open standard co-developed by OpenAI and Stripe that lets AI agents, buyers, and merchants interact through structured messages to discover products, build carts, and complete purchases. It was open-sourced in February 2026 and is governed by the Agentic AI Foundation (co-founded by OpenAI, Anthropic, and Block).

Merchants stay the merchant of record, set their own prices and taxes, and handle payments and fulfillment through their existing systems. ACP defines how to create, update, and confirm a checkout — powering "Instant Checkout" in ChatGPT and third-party apps.

Agentic commerce refers to AI systems that act on behalf of users to discover products, compare deals, negotiate prices, and complete purchases — often within chatbot or conversational interfaces. Instead of browsing websites manually, a consumer describes what they want and an AI agent handles the research, selection, and transaction.

Major payment executives have called it more transformative than the original rise of e-commerce. As of early 2026, 38% of consumers use AI when shopping, 41% use dedicated AI platforms for product discovery, and ChatGPT alone processes over 50 million shopping queries per day.

A shopper opens ChatGPT and types: "Find me a quiet cordless stick vacuum for a small apartment under $300." The AI agent researches products across multiple retailers, compares reviews and prices, presents a personalized buyer's guide, and — through apps like Instacart, Target, or Expedia — lets the user complete the purchase without leaving the conversation.

Other live examples include: Walmart's Sparky agent (35% higher average order values), Google's AI Mode in Search with direct checkout via Etsy and Wayfair, and Instacart's native ChatGPT grocery checkout covering 98%+ of U.S. households.

Agentic commerce operates through a chain of standardized protocols:

1. Discovery: The AI agent uses product feeds, merchant catalogs, and shopping graphs to find relevant items based on the user's intent.
2. Selection: The agent compares options using structured product data (price, availability, reviews, shipping) and presents curated recommendations.
3. Checkout: Using protocols like ACP or UCP, the agent creates a secure checkout session with the merchant, passing order details via signed JSON messages.
4. Payment: Delegated payment tokens (e.g., Stripe SPT, Visa Trusted Agent, Mastercard Agent Pay) authorize the transaction without exposing raw card data.
5. Post-purchase: Order tracking, returns, and customer support flow back through webhooks between the merchant and the AI platform.

Instead of redirecting users to a webpage, ACP uses signed JSON messages to create and confirm orders directly in conversational interfaces. The merchant backend still controls pricing, inventory, and risk checks — but the buyer interacts through a chat window or app.

As of March 2026, OpenAI has shifted its strategy: ChatGPT now focuses on product discovery and research, with Instant Checkout transitioning to retailer apps (Instacart, Target, Expedia, Booking.com). ACP still serves as the infrastructure connecting users to merchants, but the final transaction increasingly completes in the merchant's own environment rather than natively inside the chatbot.

ACP and UCP are complementary standards that serve different parts of the agentic commerce lifecycle:

Industry consensus (including data from Checkout.com) shows that dual ACP+UCP implementation captures 40% more agentic traffic than single-protocol approaches. Retailers like Walmart and Shopify are adopting both.

MCP (Model Context Protocol) is an interoperability standard developed by Anthropic that allows AI agents and systems to share context, intent, and data. It defines how an AI model connects to external tools, databases, and APIs — acting as the "plumbing" for agent capabilities.

UCP (Universal Commerce Protocol) is a commerce-specific standard developed by Google and Shopify. It defines how AI agents discover products, negotiate checkout, link identities, and manage post-purchase workflows across retail surfaces.

In practice, UCP is compatible with MCP — retailers can expose their commerce capabilities via MCP servers while using UCP to structure the shopping journey. Worldline, Criteo, and Microsoft Dynamics 365 all deploy MCP servers for payment and catalog access within agentic commerce flows.

These protocols are complementary, not competing. A single transaction may use MCP for tool connectivity, UCP for product discovery, AP2 for payment authorization, and ACP for checkout completion.

Yes, but the model has evolved. As of March 2026, OpenAI has shifted ChatGPT's commerce strategy to focus on product discovery and research rather than completing transactions directly inside the chatbot.

Merchants can sell through ChatGPT in two ways:

• ChatGPT Apps: Retailers like Instacart, Target, Expedia, and Booking.com have dedicated ChatGPT apps where users discover products via the AI and complete purchases within the retailer's app environment.
• Shopping Research: ChatGPT's shopping research feature (powered by GPT-5 mini) creates personalized buyer's guides, and users click through to merchant sites to complete purchases.

ACP remains the underlying protocol connecting merchants to ChatGPT's discovery engine, but Instant Checkout inside the chat window itself is being phased to third-party apps.

1. Generate a product feed (CSV, TSV, JSON, or XML) containing: ID, title, description, price, inventory, shipping options, and flags such as enable_search and enable_checkout. Feeds should update every 15 minutes.
2. Implement ACP endpoints: create, update, complete, cancel — secured with HMAC + HTTPS.
3. Support delegated payment tokens (Stripe SPT or compatible).
4. Apply via the ChatGPT merchant onboarding portal at chatgpt.com/merchants.

Alternatively, merchants using platforms like Shopify, BigCommerce, Wix, or commercetools can use pre-built integrations via Stripe's Agentic Commerce Suite for streamlined onboarding.

ChatGPT commerce refers to the shopping capabilities built into OpenAI's ChatGPT platform. With over 800–900 million weekly active users and 50 million shopping queries per day, ChatGPT has become one of the largest product discovery surfaces on the internet.

Key features include:

• Shopping Research: A deep-research experience that builds personalized buyer's guides using GPT-5 mini with reinforcement learning for shopping tasks.
• ChatGPT Apps: Dedicated retailer integrations (Instacart, Target, Expedia, Booking.com) for end-to-end shopping within familiar brand environments.
• Agentic Commerce Protocol: The open infrastructure connecting merchants, AI agents, and payment providers across the full shopping journey.

AI Platforms

• ChatGPT (OpenAI) – Shopping research + retailer apps (Instacart, Target, Expedia, Booking.com)
• Microsoft Copilot – Copilot Checkout powered by Stripe (in rollout)
• Salesforce Agentforce – Instant Checkout integration with Stripe
• Perplexity – Shopping experience with PayPal checkout

E-commerce Platforms

• Shopify – ACP + UCP dual-protocol support (1M+ merchants)
• BigCommerce – Stripe ACS integration live
• Wix – ACP via Stripe and PayPal integrations
• commercetools – AI Hub + Agentic Jumpstart (JD Sports first enterprise adopter)
• Mirakl – Nexus platform connecting 100k+ merchants to agentic channels via Stripe

Retailers

• Etsy – Live on both ChatGPT and Google AI Mode
• Walmart – Dual ACP + UCP; Sparky agentic agent live
• Target – ChatGPT app live
• Wayfair – Live on Google AI Mode via UCP
• The Home Depot – Magic Apron AI assistant expanding to Google AI Mode

Shopify co-developed Google's Universal Commerce Protocol (UCP) and supports ACP via Stripe's Agentic Commerce Suite, making it one of the first platforms with dual-protocol support.

For Shopify merchants, this means:

• Products automatically discoverable by AI agents across ChatGPT, Google AI Mode, Gemini, and other surfaces.
• Support for discount codes, loyalty credentials, subscriptions, and complex checkout flows across any payment processor.
• Over 1 million Shopify merchants (including Glossier, SKIMS, Spanx, Vuori) coming online for ChatGPT commerce.
• Merchants remain the seller of record with full control over pricing, fulfillment, and customer relationships.

ACP is payment-processor agnostic. The ecosystem has expanded significantly:

Core Infrastructure

• Stripe – Shared Payment Token API; machine payments via stablecoin micropayments; $1.9T total payment volume in 2025
• PayPal – ACP server connecting tens of millions of small businesses; agentic commerce services live
• Apple Pay, Google Pay, Link by Stripe – Digital wallet support

Card Networks

• Visa – Trusted Agent Protocol (live with Fiserv)
• Mastercard – Agent Pay with Agentic Tokens (pilots with Microsoft Copilot, Google UCP)

PSPs & Orchestration

• Checkout.com – ACP adopted for enterprise merchants
• Mollie – Delegated Payments API for European merchants
• Worldpay – Open-source MCP server for agentic payments
• Worldline – MCP servers on Global Collect; supports UCP and AP2
• FIS – Industry-first bank-led agentic commerce offering with Visa and Mastercard
• Spreedly – Agentic commerce live as a channel; UCP and ACP support coming Q1 2026 (Priceline already testing)
• Nexi Group – Google Cloud partnership for UCP/AP2 across Europe (announced March 2026)

BNPL & Installments

• Klarna – UCP infrastructure partner; flexible payment options for AI agents
• Splitit – Card-linked installments endorsed by Google UCP (announced March 2026)

Stablecoin / Crypto Rails

• Circle (USDC) – Claims ~99% of measured agentic payments in USDC; x402 protocol for machine-to-machine transactions
• Stripe (Tempo) – Purpose-built payments blockchain for agent-to-agent settlement

No. Stripe built the first reference implementation, but ACP works with any provider that supports delegated payment tokens. Merchants can continue using existing processors or PSPs by adopting the Delegated Payments Spec.

As of March 2026, PayPal, Checkout.com, Mollie, Worldpay, Worldline, Spreedly, FIS, and Nexi all support or are actively building ACP and/or UCP compatibility. Spreedly specifically enables merchants to process agent-initiated transactions over their existing PSP relationships without switching providers.

A programmable token issued by a payment provider (like Stripe) representing a limited authorization to charge up to a specific amount. It replaces raw card data, reducing PCI scope and improving security.

SPTs are scoped to a specific merchant + amount, encrypted, and verified by HMAC-signed requests over HTTPS. They can be used even by businesses that don't process payments with Stripe. Related primitives include Visa's Trusted Agent tokens and Mastercard's Agentic Tokens, which use similar tokenisation approaches.

All major cards and digital wallets (Apple Pay, Google Pay, Link by Stripe). The ecosystem is rapidly expanding to include:

• Buy Now, Pay Later: Klarna (via UCP), Splitit card-linked installments (via UCP)
• Stablecoin payments: USDC micropayments via x402 protocol (Stripe + Coinbase on Base)
• Regional methods: Expanding as PSPs like Mollie (Europe), Worldline (global), and Nexi (Europe) adopt ACP/UCP
• PayPal: Available via ACP server and Perplexity integration

Yes, when using established protocols with proper trust layers. ACP and UCP both enforce multiple security mechanisms:

• Delegated tokens: Encrypted, scoped to specific merchants and amounts — no raw card data is shared with AI agents.
• HMAC-signed requests: All communications verified over HTTPS.
• User confirmation: Buyers explicitly confirm each step before any transaction executes.
• Cryptographic mandates: AP2 (used with UCP) uses cryptographically signed authorization mandates.

ACI Worldwide's Chief Strategy Officer outlined three requirements for trusted agentic commerce: permission that can be verified, identity that persists across the entire transaction, and evidence that ensures fair outcomes when something goes wrong.

However, the industry is still evolving: only 4% of consumers currently allow AI to complete purchases autonomously, and new frameworks for disputes ("I didn't buy that, my AI did") are being developed by firms like Chargebacks911.

Key risks being actively addressed by the industry include:

• Liability ambiguity: When an AI agent makes a purchase the consumer didn't intend, existing chargeback frameworks may not clearly assign responsibility. New "Know Your Human" (KYH) verification processes are emerging alongside traditional KYC.
• Data privacy: AI agents require access to shopping preferences, payment credentials, and personal data. Regulations like the EU AI Act classify many agentic systems as high-risk.
• Fraud and impersonation: Visa warns of a "material increase in AI-powered identity attacks" in 2026. Identity verification and "Know Your Agent" frameworks are critical.
• Merchant visibility: Brands not optimized for AI agent discovery risk becoming invisible. AI agents evaluate structured data, not emotional branding.
• Tax and compliance: OpenAI's checkout pivot was partly driven by not having built state sales tax collection infrastructure — highlighting the regulatory complexity of AI-mediated transactions.

Always the merchant selling the goods — not OpenAI, Google, Stripe, or any AI platform. Your business name appears on the buyer's statement, and you handle refunds, chargebacks, returns, taxes, and fulfillment.

This is a fundamental design principle of both ACP and UCP. AI platforms provide the discovery and transaction channel, but the merchant retains full control and responsibility.

Merchants process refunds or chargebacks through their payment provider and send updates to ChatGPT (or other agents) via the ORDER_UPDATE webhook. Standard card network dispute processes (Visa, Mastercard) apply.

However, agentic commerce introduces new dispute categories. Chargebacks911 warns that AI-initiated transactions create ambiguity around consumer intent. The industry is developing new attribution models and evidence standards for cases where the boundary between what the consumer wanted and what the agent purchased is unclear.

1. Generate a product feed (CSV, TSV, JSON, or XML) with ID, title, description, price, inventory, shipping options, and enable_search / enable_checkout flags. Update every 15 minutes.
2. Implement ACP endpoints: create, update, complete, cancel.
3. Secure with HMAC + HTTPS.
4. Support delegated payment tokens (Stripe SPT or compatible).
5. Apply via the ChatGPT merchant onboarding portal.

For faster integration, platforms like Shopify, BigCommerce, Wix, commercetools, and Mirakl offer pre-built connectors via Stripe's Agentic Commerce Suite. Spreedly also enables existing PSP relationships to support agent-initiated transactions without infrastructure changes.

Google's Universal Commerce Protocol requires merchants to:

1. Host a UCP manifest at /.well-known/ucp (JSON format) describing your commerce capabilities.
2. Update Google Merchant Center with dozens of new data attributes for conversational commerce discovery.
3. Ensure structured product data includes schema.org attributes: name, SKU, GTIN, brand, offers, reviews, availability.
4. Support real-time data synchronization for inventory, pricing, and promotions.
5. Integrate payment options via Google Pay (primary), with PayPal, Klarna, and Splitit expanding.

Google's shopping graph indexes 50+ billion product listings with 2 billion updated hourly. Etsy and Wayfair are already live; Shopify, Target, and Walmart are rolling out. Nexi Group became the first major European payment partner in March 2026.

Listings are organic, not ads. Ranking factors include:

• Availability and price competitiveness
• Product quality and seller trust signals
• Whether Instant Checkout or app-based purchase is enabled
• Whether the seller is the primary merchant
• Structured product data quality (descriptions, specifications, reviews)

Unlike traditional SEO, AI agents evaluate structured product attributes, compatibility, reviews, and contextual fit against the shopper's specific query — relevance to the consumer's stated need matters more than keyword optimization or brand size.

The protocol itself supports multi-item carts, subscriptions, and digital goods. At launch, ACP in ChatGPT supported single-item checkouts. OpenAI's 2026 roadmap includes multi-item carts and international expansion, though the shift to app-based checkout may accelerate these capabilities through retailer-native experiences.

Currently one shipping address per checkout. Multi-shipment support is planned for a future version of the spec.

Yes — merchants pay a small per-transaction fee on completed purchases. Consumers pay no extra cost. In the early ecosystem, some partners (such as Etsy) covered OpenAI's commission fees to help jumpstart adoption.

ACP is available to U.S. ChatGPT users with expanding coverage. Instant Checkout launched to all U.S. users (Plus, Pro, and Free tiers) in February 2026 and is now transitioning to retailer apps.

UCP is live in the U.S. via Google AI Mode in Search and the Gemini app (Etsy, Wayfair). European expansion is underway with Nexi Group (March 2026), Worldline, and Mollie building UCP/AP2 infrastructure.

International rollout across both protocols is planned for 2026–2027.

Merchants do. Both ACP and UCP simply transmit the order and confirmation between the buyer and seller; fulfillment, taxes, and post-purchase support remain under the merchant's control. This was a key factor in OpenAI's March 2026 decision to move checkout to retailer apps — handling tax collection, compliance, and post-purchase support at scale proved complex for the AI platform itself.

In March 2026, OpenAI confirmed it is scaling back direct checkout inside ChatGPT. The company found that while users enthusiastically used ChatGPT for product discovery and research, very few completed purchases inside the chatbot — preferring to switch to familiar retail environments with saved accounts, payment methods, and order histories.

An OpenAI spokesperson said: "We're prioritizing making ChatGPT search and product discovery great, with the Agentic Commerce Protocol serving as the infrastructure that connects users to merchants across the full shopping journey."

Instant Checkout is transitioning to retailer apps. The four partners with functioning ChatGPT apps — Instacart, Target, Expedia, and Booking.com — represent the new model: AI handles discovery and intent, and the transaction completes in the retailer's own environment.

AI Platforms

• ChatGPT (OpenAI) – Shopping research + retailer apps
• Microsoft Copilot – Copilot Checkout (Stripe-powered)
• Salesforce Agentforce
• Perplexity (PayPal checkout)

E-commerce Platforms

• Shopify (ACP + UCP)
• BigCommerce, Wix, commercetools, Mirakl, Rithum

Payment Providers

• Stripe (SPT API, machine payments, Tempo blockchain)
• PayPal (ACP server + UCP support)
• Apple Pay / Google Pay / Link
• Visa Trusted Agent Protocol (with Fiserv)
• Mastercard Agent Pay (with Microsoft, Google)
• Checkout.com, Mollie, Worldpay, Worldline, FIS
• Spreedly (agentic channel live; UCP/ACP support in development)
• Nexi Group (UCP/AP2 via Google Cloud for Europe)
• Klarna, Splitit (UCP BNPL/installments)
• Circle USDC (x402 stablecoin payments)

Infrastructure & Data

• Mirakl Nexus (agentic commerce brain – 100k+ merchants)
• Criteo (Agentic Commerce Recommendation Service via MCP)
• Ballerine (Trusted Agentic Commerce Governance Platform)
• Tredence (Agentic Commerce Solution Accelerators)
• Salsify, Syndigo, Netcore Unbxd (AI-ready product feeds)

Stablecoins are emerging as a programmable settlement layer for machine-to-machine transactions. Stripe co-founder John Collison predicts a "torrent" of agentic commerce where agents transact with stablecoins on high-throughput blockchains.

Key developments as of March 2026:

• Stripe launched machine payments using USDC on Base (Coinbase's L2) via the x402 protocol, and unveiled Tempo — its own payments blockchain with sub-second finality.
• Circle reported that ~99% of measured agentic payments use USDC; its Q4 2025 onchain volume grew 247% to $11.9 trillion.
• Stablecoin payments on Stripe doubled to ~$400 billion in 2025, with 60% estimated to be B2B.

Stablecoins are especially relevant for B2B agentic commerce, where Forrester predicts approximately one-third of payment workflows will use autonomous AI agents by end of 2026.

Key market metrics as of March 2026:

• Morgan Stanley predicts nearly half of online shoppers will use AI shopping agents by 2030, accounting for ~25% of their spending.
• Rezolve Ai projects a $1–5 trillion agentic commerce market by 2030.
• Consumer adoption: 41% of consumers use dedicated AI platforms for product discovery; 33% have fully replaced prior methods; 38% use AI when shopping.
• ChatGPT: 800–900M weekly active users; 50M shopping queries/day.
• Google AI Mode: 75M+ daily active users.
• Stripe: $1.9T total payment volume processed in 2025; valued at $159 billion.
• Trust gap: Only 4% of consumers currently allow AI to complete purchases autonomously, suggesting massive growth potential as trust frameworks mature.

Agentic commerce is the most significant shift in digital buying since the rise of e-commerce itself. ACP and UCP are the two foundational protocols — ACP for checkout and payment flows, UCP for the full commerce lifecycle — and they work together with AP2, MCP, x402, and card network frameworks (Visa Trusted Agent, Mastercard Agent Pay) to create a layered trust architecture.

As of March 2026, the ecosystem has matured significantly: OpenAI has pivoted ChatGPT to a discovery-first model, Google's UCP is live with Etsy and Wayfair and expanding into Europe via Nexi, Stripe's $159B valuation reflects its central role in the infrastructure, and new entrants like Spreedly, Splitit, Circle, and ACI Worldwide are building the payment rails for autonomous AI transactions.

For merchants, the message is clear: prepare for both protocols, optimize product data for AI agent discovery, and ensure your payment infrastructure supports delegated token-based transactions. The companies that are agent-ready now will capture disproportionate share as the ecosystem scales.