Visa Agentic Commerce: What Store Owners Control
Search for Visa and AI shopping agents and you get a pile of pages written for bankers and fintech watchers. They tell you what Visa announced, which partners showed up on stage, and what the press release called it. They say almost nothing about what you, the person running a store, should open and fix on Monday morning.
That gap is the whole reason I am writing this. You do not sign up to a Visa program to become shoppable by an AI agent. Your store gets there, or does not, through three things you already own: product data an agent can read, return and shipping policies it can parse, and a checkout an automated buyer can finish without a human stepping in.
So the payment-network news is background, not your to-do list. Think about it this way - the best card tokens in the world do nothing if the agent cannot tell which size is in stock, or hits a login wall at step three. Puffff, that is where orders die, long before any network gets involved.
Below I explain what Visa's agentic push is in plain terms, then go through those three things one by one, with a numbered audit you can run on your own store this week. Read the announcements later, after your own house is in order. Miser pay twice))
If you read nothing else
You do not join a Visa program to sell to AI agents. You get there by fixing your product data, your policies and your checkout. The payment network is your processor's job, not yours.
What Visa Intelligent Commerce is, in plain terms for merchants
Visa Intelligent Commerce is Visa's program for letting AI agents pay on a cardholder's behalf.
“AI agents are becoming autonomous participants in the payment process, capable of initiating transactions on behalf of shoppers.”
The agent gets a secure, tokenized credential instead of a raw card number, and the cardholder sets limits on what it can spend.
“Soon people will have AI agents browse, select, purchase and manage on their behalf,” said Visa’s Chief Product and Strategy Officer Jack Forestell. “These agents will need to be trusted with payments, not only by users, but by banks and sellers as well.”
The pieces you see named in the press, like Visa Intelligent Commerce Connect and the Visa Intelligent Commerce API, are the plumbing that banks, agent builders and payment partners plug into.
That plumbing sits between the cardholder's bank and the agent. Your store is not a party to it in the way a bank is. If you already take Visa cards through Stripe, Adyen, Shopify Payments or whatever processor you use, the network side is mostly your processor's conversation to have.
Visa introduced Intelligent Commerce in 2025, and the merchant-facing piece, Intelligent Commerce Connect, is still in a limited pilot with payment partners.
“As consumers increasingly rely on AI agents to make purchases, businesses – whether they are building agents, selling to them, or processing transactions – need a simple way to get started. Intelligent Commerce Connect, part of the Visa Intelligent Commerce portfolio, meets that need.”
So most stores get access through their existing processor, not a direct Visa sign-up.
“As consumers begin to delegate tasks to AI agents, maintaining control and security in payments is critical. Visa Intelligent Commerce is designed to support these experiences by enabling solutions that help consumers authorize agent-driven payments with confidence and clear boundaries.”
The network side also handles the parts that sound scary: authenticating that the agent is allowed to act for the shopper, verifying it is a trusted agent and not a bot, and confirming the shopper's intent to buy. That is Visa's job and your processor's, not yours. The benefit to you is simple: an agent that can read your store and pay through it finishes the order; one that cannot moves on.
| Visa and your processor handle | You handle |
|---|---|
| Tokenizing the card so the agent never sees the raw number | Product data clean enough for the agent to pick the right item |
| Authenticating and verifying the agent is allowed to act | Return and shipping policies a parser can read |
| Confirming the shopper's intent and spending limits | A checkout an automated buyer can finish |
| Fraud checks and authorization on the network | Showing the real total before the last click |
Mastercard has its own version of this, so Mastercard agentic commerce is the same story with a different logo. Both networks are solving "how does an agent pay safely". Neither one solves "can the agent find the right product and finish the order on your site".
Why Visa agentic commerce is mostly your store's job
Ask yourself where an agent run actually fails. From what I've seen, it rarely fails at the card. It fails earlier, when the agent cannot read your data or cannot get through your pages.
A token is useless to a buyer that never reaches the pay button. So the work that decides whether you get the sale is work on your side of the fence, and none of it needs a Visa contract.
Two kinds of AI shopping agent, and what each one reads
One distinction changes what you fix, so get it straight before the checklist. Some agents drive your site like a person - a browser or computer-use agent that loads your pages and clicks through your real checkout. Others buy through a protocol, like Visa Intelligent Commerce, the Agentic Commerce Protocol behind ChatGPT's checkout, or Shopify's agentic checkout - they place the order through an API and never see your form.
Clean product data and readable policies matter to both. Your checkout pages only matter to the first kind. You do not get to pick which one shows up, so the safe play is to get all three right - but if an agent never touches your form, the fix for it is your feed and your processor's protocol integration, not your form fields.
What the network cannot fix for you
1) Product data. The network does not know your sizes, stock or prices. If your feed says "one size" and your page says four, the agent picks one and guesses.
2) Policies. A return window buried in a PDF or an image is invisible to a parser. Put it in plain text on a page with a stable URL.
3) Checkout. Forced account creation, CAPTCHA at the last step and pop-ups that block the form stop an automated buyer cold.
I have no hard number on how often each of these breaks a run, so I will not invent one. But all three are things you can open and look at in an afternoon.
Where the announcements fit
Read the network news to know what is coming. Do not wait on it to start. A store with clean data and a plain checkout is ready for whichever rail the agent pays through, and a messy store is not saved by any of them.
Clean product data: what an AI agent reads before it buys
An agent shopping for a customer reads your structured data or feed first, and your page a distant second if at all. If the two disagree, it has to pick one, and you do not get to choose which.
Open one product page and check these, in this order:
1) Price and currency. Match the Product and Offer schema to the visible price. Test it in Google's Rich Results Test (search.google.com/test/rich-results) and fix any mismatch.
2) Availability. Mark each variant in stock or out of stock in the markup, not just the parent product. "One size" in the feed and four sizes on the page is the classic break.
3) Identifiers. Add GTIN, MPN or SKU where you have them. An agent comparing across stores needs something to match on, and a product title alone is weak.
4) Variants as text. Put size, color and material in plain text on the page. A size chart saved as an image is invisible to a parser.
5) One feed, one truth. If you push a Google Merchant Center feed, run it against the live page. Merchant Center's diagnostics tab lists the mismatches for you.
Your feed and your page tell the same story? Most stores I look at, they do not, and the drift is usually stock or price after a sale.
| What the agent reads | When it is clean | When it is missing or wrong |
|---|---|---|
| Availability per variant | Offers a size you actually have | Guesses, and may offer a sold-out size |
| Price in the schema | Quotes your real price | May quote a stale sale price |
| GTIN, MPN or SKU | Matches your product across stores | Cannot match, so you drop out of the comparison |
| Return window in the markup | Compares your terms fairly | Treats returns as unknown and loses the tie |
This is where visa agentic commerce stops being a payments story. The token only matters after the agent has decided your product is the right one, and it decides that from your data.
Return and shipping policies an AI agent can parse
An agent comparing two stores with the same product often breaks the tie on delivery date and return terms. If it cannot read yours, it treats them as unknown, and unknown loses to a store that states them.
What to fix
1) Write the policy as plain HTML text. Do not ship it as a PDF or a screenshot. A parser reads a heading and a paragraph, not a picture of one.
2) Give it a stable URL, like /returns and /shipping, and link both from the footer and the product page.
3) State numbers. "30 days from delivery", "free over $50", "ships in 2 business days". "Fast shipping" and "easy returns" tell an agent nothing.
4) Add the markup. Use MerchantReturnPolicy and OfferShippingDetails in schema.org so the terms sit next to the Offer. Validate with the Schema Markup Validator (validator.schema.org).
5) List exceptions in the same place. Final-sale items, regions you do not ship to, restocking fees. A rule hidden on page three of a PDF is a rule the agent never sees.
Two versions that disagree is worse than one
Keep the human version and the machine version saying the same thing. If the footer says 30 days and the markup says 14, the agent cannot ask which is right - it picks one, and you have created the same conflict as a wrong stock count, just on policy.
A checkout an AI agent can complete without a human
An automated buyer has no patience and no phone. Every step that assumes a person is sitting there is a step where the run ends. This section is for the agents that drive your actual site - a protocol-based agent never sees your form, so for those the checkout is your processor's or platform's integration, not your pages.
Walk through your own checkout as if you were a script, and check these:
1) Guest checkout. Turn it on. Forced account creation is the most common wall, and a script cannot verify an email it has no inbox for.
2) CAPTCHA. Do not rip it out - that is how card testers and scrapers get in. Swap a blunt "select all traffic lights" puzzle for risk-based bot management (Cloudflare Turnstile in managed mode, or your platform's bot protection) that passes real humans and verified agents quietly and only challenges traffic that looks wrong. The goal is to let the good agent through, not to drop the wall for every bot.
3) Form fields. Use standard labels and autocomplete attributes (name, address-line1, postal-code). A field called "field_7" is a guess.
4) Pop-ups. Kill newsletter overlays and cookie banners that cover the pay button or steal focus.
5) Total before the last click. Show shipping, tax and the final price before payment. A surprise fee at step four makes the agent abandon, same as a human.
6) Standard payment methods. Take cards through your normal processor, in a normal payment form. That is also where the network side lives, so it is the one part you can leave to your processor.
How to audit your store before any Visa agentic commerce rollout
Run this in one sitting, on your top-selling product, and write down every place you get stuck.
1) Product data. Open the page, then run it through Google's Rich Results Test (search.google.com/test/rich-results). Does the schema price and stock match what the page shows?
2) Feed. Open the Diagnostics tab in Google Merchant Center and clear the mismatches.
3) Policies. Open /returns and /shipping with no login. Are the terms plain text with numbers, and do they match the markup in the Schema Markup Validator (validator.schema.org)?
4) Checkout. Buy the product as a guest in a private browser window, on your phone, with no saved details. Count the steps, the pop-ups and the forms you had to fight.
5) Ask an agent. Paste your product URL into ChatGPT or Perplexity and ask it to compare your item against a competitor on price, delivery and returns. What it gets wrong, or says it cannot find, is your fix list.
Do these five and you have covered what decides whether an agent finishes an order on your store. Anything left over is for your payment processor to handle, and you can read the Visa and Mastercard announcements when it suits you.
Checkout broken today? Then no token on earth saves the sale. Fix the data, the policies and the checkout first, and the rest follows.
FAQ
Is Visa agentic commerce different from Visa Intelligent Commerce?
In practice people use the two phrases for the same thing. Visa Intelligent Commerce is the program name, and agentic commerce is the wider idea of AI agents shopping and paying for people. For a store owner the difference does not change the to-do list, which is still data, policies and checkout.
Do merchants need to sign up for a Visa program to sell to AI agents?
From what I've seen, no. The network side runs through banks, agent builders and your payment processor, so ask your processor what they support. Your own work is making the store readable and completable for an automated buyer.
What does a store actually need to be shoppable by an AI agent?
Three things: product data that matches between your schema, feed and page, return and shipping terms written as plain text with numbers, and a checkout with guest purchase and no last-step CAPTCHA. If an agent can read the product and finish the order, the store is shoppable. Payment tokens only matter after those are in place.
How do I test whether an AI agent can complete my checkout?
Buy your top product as a guest in a private browser window on your phone, and count every wall you hit. Then paste the product URL into ChatGPT or Perplexity and ask it to compare you against a competitor on price, delivery and returns. Whatever it gets wrong or cannot find goes on your fix list.
Should I wait for Visa or Mastercard announcements before I change my store?
No. A store with clean data and a plain checkout is ready for whichever payment rail the agent uses, and a messy store is not rescued by any of them. Read the announcements for context and fix your own pages first.