In Brief
- Core Answer: Agentic commerce is buying and selling in which an AI agent — not a person clicking through a website — searches, compares, chooses and completes the transaction on the user's behalf, within limits the user sets. ChatGPT Instant Checkout, the Stripe–OpenAI Agentic Commerce Protocol, and the Mastercard, Visa and PayPal agent-payment programs are the infrastructure that made it real in 2025–2026.
- Why It Matters: When the buyer is an agent, the product page, the checkout flow and the ad are no longer the moment of decision. The agent decides from structured data and from what answer engines say about you. Brands that are not machine-readable and not recommended do not get bought.
- Best For: Ecommerce, marketplace and B2B leaders who need to understand what agentic commerce is, how the protocols work, and what to change so agents can find and transact with them.
Agentic commerce is the form of digital commerce in which an AI agent completes some or all of the purchase journey for a person: understanding what they want, searching the market, comparing options, and executing the transaction with pre-authorized payment credentials. The user states a goal and constraints — a budget, a delivery date, a brand preference — and the agent does the shopping. Mastercard calls it the agent "closing the loop"; Stripe and OpenAI built an open protocol for it; and since OpenAI launched Instant Checkout in ChatGPT in September 2025, hundreds of millions of people have had a shopping agent inside the app they already use every day.
What Agentic Commerce Actually Is
Three things have to be true for a transaction to count as agentic commerce rather than AI-assisted shopping.
- The agent acts, not just advises. A chatbot that recommends three laptops is assistance. An agent that adds one to a cart and pays is commerce.
- The user delegates within bounds. Constraints are set up front — spend limits, merchants allowed, approval thresholds — and the agent operates inside them without asking at every step.
- The merchant is reachable by machine. The agent needs structured product data, real-time inventory and price, and a programmatic checkout. If it has to scrape a page and guess, the merchant is at a disadvantage against one that exposes clean data.
Wikipedia's definition captures the spectrum: semi-autonomous agents that prepare a purchase for human approval at one end, fully autonomous agents that complete it at the other. Most 2026 deployments sit in the middle — the agent does everything except the final confirmation.
How Agentic Commerce Works: The Stack
The conversation layer
The buyer's intent is expressed in natural language inside ChatGPT, Gemini, Perplexity, Claude, Copilot or a retailer's own assistant. This is also where recommendation happens — the agent narrows the market to a shortlist before any transaction. Profound's research has called this "the shortlist is the new shelf": being in the two or three options the agent presents is now the equivalent of eye-level placement.
The discovery layer
Agents find products through search indexes, product feeds and structured data — schema.org Product and Offer markup, merchant feeds, and increasingly the same catalog feeds OpenAI now requires for ChatGPT shopping ads. Schema markup for AI search and crawler access are the technical preconditions; without them the agent cannot read price, availability or variants reliably.
The protocol layer
The Agentic Commerce Protocol (ACP), developed by Stripe and OpenAI and published as an open standard, defines how an agent and a merchant exchange product information, create a checkout, and complete payment programmatically. Google, Shopify, PayPal, Mastercard, Visa and American Express have each announced agent-commerce programs or partnerships; Mastercard and PayPal's joint agentic-payments partnership and Stripe's Agentic Commerce Suite are the two most searched. The payment networks' contribution is tokenized, agent-scoped credentials so a merchant can verify that an agent is authorized to spend on a specific person's behalf.
The transaction layer
Instant Checkout in ChatGPT is the reference implementation: a user sees a product in a conversation and buys it without leaving. Shopify merchants were auto-enrolled, Etsy enrolled all US sellers, and Salesforce Commerce Cloud brought in brands including L'Oréal, Pandora and Saks. The merchant fulfils the order and pays a fee; the conversation never leaves ChatGPT.
Agentic Commerce Examples in 2026
- Consumer: a user tells ChatGPT "find me a mid-weight running jacket under $150 that ships by Friday," receives a shortlist, and buys through Instant Checkout.
- Replenishment: an agent reorders consumables when inventory or usage data crosses a threshold, choosing the supplier by price and lead time each time.
- Travel: an agent books flights and hotels against a policy — budget, loyalty program, times — and files the receipts.
- B2B procurement: an agent requests quotes, compares against a rate card, and raises a purchase order for approval. This is where the largest dollar volume will sit, and where the buyer-side agent is most likely to be built in-house.
What Changes for Brands
The visible change is that the transaction moves into the conversation. The important change is that the decision moves there too, and the decider is a model.
Old lever | Agentic-commerce equivalent |
|---|---|
Product page persuasion | Structured product data the agent can parse |
Paid search on the query | Being recommended in the answer, or a sponsored card below it |
Checkout optimization | Protocol support (ACP, agent payments) so the agent can transact |
Brand awareness | What answer engines say about you when asked, unprompted |
Reviews on your site | Third-party corroboration the model trusts |
Conversion rate | Recommendation rate: how often the agent shortlists you |
Two consequences follow. First, AI visibility stops being a marketing vanity metric and becomes a distribution metric: if the agent does not recommend you, the transaction cannot happen. Second, revenue attribution has to be rebuilt, because the click that used to mark the start of a purchase no longer exists. Tracking revenue impact from AI chat — through checkout-protocol data, order-source fields and referral tagging — is the only way to know what agentic channels are worth.
What to Do Now
- Make the catalog machine-readable. Complete Product and Offer schema, a validated product feed, and real-time price and inventory endpoints.
- Get into the shortlist. Run an AI visibility audit to see how often ChatGPT, Gemini and Perplexity recommend you for your category's buying questions, and fix the gaps with corroborating third-party coverage rather than more on-site copy.
- Support at least one checkout protocol. If you sell on Shopify, Etsy or Salesforce Commerce Cloud you may already be enrolled; confirm it and read the order data it produces.
- Decide on paid coverage. ChatGPT ads can put a sponsored card beneath a conversation you are not winning organically. Use it to cover gaps, not to replace recommendation.
- Instrument attribution before volume arrives. Tag agent-originated orders now so the first meaningful quarter of agentic revenue is measurable.
The payment layer beneath all of this has its own rails and rules — see agentic payments for how Mastercard Agent Pay, Visa Intelligent Commerce, Stripe and the x402 standard let agents transact — and the shopping and payment agents are two of the twenty AI agent examples now running in production.
Frequently Asked Questions
What is agentic commerce in simple terms?
Shopping where an AI agent does the searching, comparing and buying for you, within limits you set — instead of you browsing sites and filling in checkout forms.
What is the difference between agentic AI and agentic commerce?
Agentic AI is the general capability of software to pursue goals autonomously. Agentic commerce is that capability applied to buying and selling. Every agentic-commerce system is agentic AI; most agentic AI has nothing to do with commerce.
Is agentic commerce real yet?
Yes, at consumer scale. ChatGPT Instant Checkout has been live since September 2025 with Shopify, Etsy and Salesforce Commerce Cloud merchants enrolled, and the Agentic Commerce Protocol is published as an open standard. B2B and procurement use cases are earlier and mostly custom-built.
What is the Agentic Commerce Protocol?
An open standard developed by Stripe and OpenAI that defines how AI agents and merchants exchange product data, create checkouts and complete payments programmatically. It is the interface behind Instant Checkout in ChatGPT.
How do brands get chosen by shopping agents?
By being machine-readable and recommended: complete structured product data the agent can parse, and enough trusted third-party corroboration that the answer engine shortlists the brand when asked. Paid placement can supplement this but does not replace it.
References
- https://www.agenticcommerce.dev/
- https://stripe.com/guides/agentic-commerce
- https://www.ibm.com/think/topics/agentic-commerce
- https://www.mastercard.com/us/en/news-and-trends/stories/2025/agentic-commerce-explainer.html
- https://en.wikipedia.org/wiki/Agentic_commerce
- https://searchengineland.com/agentic-commerce-chatgpt-ads-482010
- https://cloud.google.com/transform/a-new-era-agentic-commerce-retail-ai
- https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-agentic-commerce-opportunity-how-ai-agents-are-ushering-in-a-new-era-for-consumers-and-merchants