Ask ChatGPT to find a supplier for packaging tape with same-day shipping and, increasingly, it doesn't just recommend a vendor — it can complete the purchase. That shift isn't a chat interface trick. It's the product of two new technical standards, the Agentic Commerce Protocol (ACP) and the Agent Payments Protocol (AP2), that define how AI agents discover products, prove they're authorized to buy, and pay merchants without a human clicking "buy." For B2B sellers, understanding what these protocols actually standardize — not just that "AI shopping is coming" — is what separates being agent-ready from being invisible to the agents doing the buying.
We've already covered why your product data needs to be agent-ready for procurement AI. This post goes one layer deeper: the transaction infrastructure connecting buyer agents to seller systems, who built it, and what it requires operationally.
Before either protocol existed, every AI agent needed a custom, one-off integration with every merchant it wanted to transact with. Neither ACP nor AP2 is a product you buy — both are open specifications that define a shared "language" for agent-to-merchant transactions, so a business can build the integration once instead of once per agent.
The Agentic Commerce Protocol was released as an open standard on September 29, 2025, codeveloped by Stripe and OpenAI under an Apache 2.0 license. It powers Instant Checkout inside ChatGPT, which launched the same day and initially let US users buy from Etsy sellers, with a rollout to more than a million Shopify merchants — including Glossier, Vuori, Spanx, and SKIMS — following shortly after. ACP standardizes three things: machine-readable product feeds so agents can discover what a merchant sells, a conversational checkout flow that supports back-and-forth rather than a single "buy now" click, and a Shared Payment Token that lets ChatGPT initiate payment without ever seeing the buyer's card details. Merchants remain the merchant of record — they keep control of pricing, fulfillment, and returns. Because it's published under Apache 2.0, any business can adopt ACP with any payment processor, not only Stripe.
The Agent Payments Protocol, announced by Google on September 16, 2025, tackles a narrower but harder problem: proving an AI agent actually has authority to spend a specific person's or company's money. AP2 does this through three cryptographically signed "mandates" — an Intent Mandate (what the user authorized the agent to do), a Cart Mandate (the exact items and price the user approved), and a Payment Mandate (the credential the payment network sees). Together they create an audit trail answering who authorized what, and under what limits. AP2 extends Google's Agent2Agent (A2A) protocol and works alongside the Model Context Protocol (MCP). More than 60 organizations — including Mastercard, American Express, PayPal, Salesforce, and ServiceNow — are collaborating on the standard.
Put simply: ACP is closer to a discovery-and-checkout spec built around one dominant surface (ChatGPT) with a payments company (Stripe) behind it. AP2 is a broader trust-and-authorization framework meant to work across many agents, merchants, and payment rails — and Google has explicitly named B2B scenarios like autonomous procurement and software license scaling as target use cases.
| Protocol | Built By | What It Standardizes | Status (July 2026) |
|---|---|---|---|
| ACP Agentic Commerce Protocol | OpenAI + Stripe | Product discovery feeds, conversational checkout, Shared Payment Token for delegated payment | Live — powers ChatGPT Instant Checkout since Sept 2025 (Etsy, expanding to 1M+ Shopify merchants) |
| AP2 Agent Payments Protocol | Google + 60+ partners (Mastercard, Amex, PayPal, Salesforce, ServiceNow, others) | Cryptographically signed Intent, Cart, and Payment mandates proving authorization, authenticity, accountability | Open protocol extending A2A/MCP; partner ecosystem forming, not tied to a single consumer checkout surface |
ACP's first public use case is consumer shopping inside ChatGPT, which makes it tempting for B2B sellers to file this under "not my problem yet." Forrester's research disagrees.
Forrester's 2026 predictions state that at least one in five B2B sellers will have to respond to AI-powered buyer agents with dynamically generated counteroffers from their own seller-side agents — a prediction about behavior expected within the current calendar year, not a distant hypothetical tied to full protocol maturity.
The demand side is moving in parallel. Deloitte's State of AI in the Enterprise research found 74% of surveyed enterprises expect to be using AI agents at least moderately by 2027, with adoption intent running well ahead of governance maturity — only 21% report a mature governance model for agentic AI in place today.
ACP and AP2 are not competing shopping apps — they're competing definitions of transaction trust and structure. A B2B seller doesn't get to pick "the winner" and ignore the other. The realistic posture is protocol-agnostic: structured, accurate, frequently updated product and pricing data that can be exposed through whichever standard a buyer's agent shows up speaking.
Whether an agent is negotiating over ACP or AP2 rails, it's pulling from the same categories of data before it commits to a purchase: current pricing, active promotions or contract terms, real-time inventory availability, and delivery or lead-time estimates. Unlike a human buyer, an agent has little tolerance for a product page that's weeks out of date — it either has a current answer or it moves to a supplier that does. We covered the data-readiness side of this problem in our earlier post on agentic commerce and B2B procurement; this post is about the transaction layer sitting on top of that data.
— Jon Sigler, EVP & GM, AI Platform, ServiceNow, on Google's Agent Payments Protocol (AP2)
Neither protocol has a mature, out-of-the-box answer for B2B-specific complexity — per-customer contract pricing, multi-stakeholder approvals, negotiated terms. That's a reason to prepare deliberately — starting with the kind of structured data and AI-visibility audit work we do for B2B sellers — not a reason to wait.
Neither protocol is finished. ACP currently centers on ChatGPT's consumer checkout; B2B-specific extensions for multi-stakeholder approval, contract pricing, and coordinated negotiation are still early. Building a deep, single-protocol integration today carries real rework risk — prioritize clean, structured data over locking into one vendor's checkout flow.
We help B2B SaaS, professional services, and B2B e-commerce teams get their product data and content structured for how AI agents actually discover and evaluate suppliers.
Book a Free AI Visibility CheckACP is an open standard for AI-agent checkout, codeveloped by Stripe and OpenAI and released under an Apache 2.0 license on September 29, 2025. It standardizes machine-readable product discovery, conversational checkout, and a Shared Payment Token that lets an agent like ChatGPT initiate a purchase without seeing the buyer's card details, while the merchant stays the merchant of record. It currently powers Instant Checkout inside ChatGPT.
AP2 is an open protocol announced by Google on September 16, 2025, built with more than 60 partner organizations including Mastercard, American Express, PayPal, Salesforce, and ServiceNow. It focuses on proving an AI agent's authority to transact using three cryptographically signed mandates — Intent, Cart, and Payment — that create an audit trail of who authorized a purchase and under what conditions. AP2 extends Google's Agent2Agent (A2A) protocol and works alongside the Model Context Protocol (MCP).
Not directly — they solve overlapping but distinct problems. ACP is primarily a checkout and product-discovery specification tied to Stripe's payment infrastructure and ChatGPT's Instant Checkout. AP2 is a broader authorization and trust framework meant to work across many agents, merchants, and payment networks. It's possible both persist as complementary layers rather than one replacing the other.
No. Neither protocol has mature, standardized support for B2B-specific complexity like per-customer contract pricing, multi-stakeholder approvals, or negotiated terms yet. The higher-priority step is making sure product, pricing, and inventory data is structured and API-accessible, since that data is what any future integration — ACP, AP2, or something else — will need to consume.
Forrester's 2026 predictions estimate that at least one in five B2B sellers will be compelled to respond to AI-powered buyer agents with dynamically generated counteroffers by the end of 2026. Consumer adoption is already live through ChatGPT's Instant Checkout, which launched in September 2025, so the infrastructure buyer agents rely on is maturing faster than most B2B sales organizations are preparing for it.
Product data readiness — clean titles, accurate specs, current pricing feeds — is the foundation. ACP and AP2 are the transaction layer sitting on top of that data: the rules for how an agent proves it's authorized to buy, how a checkout conversation is structured, and how payment is initiated. Sellers need both good data and protocol awareness, but they are different problems that require different work.