The short version: ChatGPT Wallet, launched by OpenAI on November 13, lets its dots agents pay for things on a shopper’s behalf, starting with US ChatGPT Pro subscribers. The same family of AI engines that will steer those agents showed shoppers a fake or dead store site in 8 percent of ChatGPT’s where-to-buy answers in a September study, and smaller brands got roughly double the exposure of big ones. The agent can now pay. The question for a small online seller is whether it can find the real you, and OpenAI’s own business guide says the fix starts with you claiming your listing.
That is the part of today’s launch written for business owners, and most coverage stopped before it. Below: what ChatGPT Wallet actually does, what the study found about where AI answers send shoppers, the line in OpenAI’s merchant FAQ about who handles a bad order, and four checks a small shop can make this month, most of them free.
What is ChatGPT Wallet?
ChatGPT Wallet is a payment method for AI agents. According to OpenAI’s announcement, read November 13, 2026, a shopper adds a card once. Before each purchase they review the items, the total and the selected card, then approve with a security check such as Face ID or a fingerprint. After approval, the card network issues a payment token, a stand-in card number with its own security code and expiration date, so neither the agent nor the store ever sees the real card details.
It was built with American Express, Mastercard and Visa. Shoppers keep their card’s existing rewards and their protections for eligible disputes. It works only inside dots, OpenAI’s always-on agents, and only for ChatGPT Pro subscribers aged 18 and older with eligible US-issued cards. CNBC reports that OpenAI is taking no commission on these purchases during the rollout.
So on day one the audience is small: the top-priced ChatGPT plan, in one country, inside one product. That is good news for a small business. It means there is time to get ready before this reaches ordinary shoppers, and OpenAI says plainly that it intends to expand.
What OpenAI is asking businesses to do
Alongside the wallet, OpenAI published a Commerce Partner Playbook aimed at the stores on the other end of the purchase. Its advice comes in four parts: make your product information available, make sure agents can reach your site, work with your payment and fraud providers so legitimate agent orders are not declined, and, optionally, build a plugin so agents can check stock or manage a cart directly.
The first step points to Businesses in ChatGPT, launched in beta three days earlier on November 10. Through Business Hub, an eligible business can claim its listing, update its details once verified, and upload a product feed with current prices and stock. Shopify merchants can connect from the Shopify admin, and OpenAI says eligible Shopify products already flow into ChatGPT through Shopify Catalog. Business Hub also reports how often your business is mentioned and clicked in ChatGPT. Neither announcement mentions a fee for claiming a listing.
Read on their own, those are marketing steps. Read next to a study published the day before the wallet, they look more like a lock on the front door.
How often do AI answers send shoppers to fake stores?
The study comes from Podqi, a brand protection company, using data from the AI search tracking firm Gauge; The Verge reported it first on November 12. Podqi asked ChatGPT, Gemini and Perplexity the same five where-to-buy questions about 151 brands over two waves in September 2026, collected 6,795 answers and reviewed every website those answers named. Its research page, read November 13, gives these results:
- About 9 percent of all answers put a risky site in front of the shopper. Podqi counts a site as risky if it impersonates a brand, sells suspected counterfeits, or is dead.
- Perplexity showed a risky site in 15 percent of its answers, ChatGPT in 8 percent and Gemini in 4 percent.
- When a risky site appeared, the engine recommended it as a place to buy 57 percent of the time and warned the shopper away only 9 percent of the time. ChatGPT warned in 24 percent of its risky answers, the best rate of the three.
- Of the 718 risky domains, 351 were fakes and 367 were dead.
- The riskiest question was the most innocent one: does this brand ship internationally, and what are its regional websites? Sixteen percent of those answers showed a risky site.
The finding that matters most for this audience comes from the coverage rather than the summary page: according to Startup Fortune’s account of the report, questions about smaller brands surfaced risky sites at roughly twice the rate of questions about large ones. A big brand has years of links and press telling an AI engine which site is real. A small one often does not.
Two fair cautions. Podqi sells brand protection, so it has a commercial interest in the problem looking large. And it tested 151 brands in one month, so the percentages are a snapshot, not a standing rate. The study also found improvement between its two waves, with risk falling from 12 percent to 7 percent, and Podqi’s own conclusion is that the problem is fixable.
Why the wallet changes the stakes
When a chatbot names a bad store, a person still has to click the link, look at the page and type a card number. Each step is a chance to notice that something is off. An agent with a wallet removes most of those steps, which is the whole point of it.
OpenAI has kept one: every purchase needs the shopper’s approval, and the shopper sees the items, the total and the card. What the announcement does not say is whether that approval screen names the store’s web address. We checked the announcement and the merchant guide for that detail on November 13 and found nothing either way. If the store is not shown, the human check confirms what is being bought and for how much, not who is selling it.
That is the sentence to hold on to: the agent can now pay, but the engines that will steer it still cannot always tell your store from a copy of it. For a big brand that is a nuisance. For a small brand with a lapsed old domain and no official list of its stores, it is a lost sale and a customer who blames you.
Who answers when an agent order goes wrong?
OpenAI’s merchant guide, the agent access page, has an FAQ with two lines worth reading before this reaches your customers.
On errors: “Customers are ultimately responsible for the actions they take,” and support should be handled the same way as if the customer had acted themselves. In practice, an agent order is your customer’s order. Returns, wrong sizes and complaints come to you on your normal terms.
On control: asked whether a site owner can require a human to confirm checkout, OpenAI answers that it does not yet have features that let specific site owners control what actions users can take, though it requires human review before any consequential action across all users. Today you cannot set your own rules for agents on your site; you rely on OpenAI’s.
Neither line is alarming. Both are worth knowing, because they settle who fields the phone call.
What should a small online store do now?
None of this needs a developer for the first two steps, and none of it needs you to buy anything.
1. Claim your business in ChatGPT. Go through Business Hub, verify, and upload a product feed with current prices. If you sell on Shopify, check whether your store is already connected through the Shopify admin. The point is not only visibility. It is giving the engine one authoritative answer to “which site is the real one,” which is exactly the ownership data Podqi says the engines are missing.
2. Publish an official-sites page. This is Podqi’s top recommendation and it costs an afternoon: one plain page listing your real store, any regional stores, and your authorized retailers, saying clearly which countries you do not ship to, linked from your footer. Then list every old domain you have ever used and keep paying for the ones AI answers might still recommend. Podqi’s count of 367 dead domains among the risky sites is the warning: a lapsed address can be re-registered by anyone.
3. Ask your web person one question. “Does our firewall block ChatGPT’s agent?” OpenAI says its cloud browser can already use most sites without any allowlist, and that Akamai, Cloudflare, HUMAN and Vercel verify its signed requests automatically. Only if your security settings are blocking those visits does anything need changing. On Cloudflare, OpenAI’s instructions are a custom rule that allows Bot Detection ID 129220581, the signed ChatGPT agent. If you would rather keep agents out, that is a legitimate choice, as our look at Amazon’s split rule for AI agents showed. Make it on purpose.
4. Tell your payment processor. OpenAI warns that fraud systems may decline legitimate agent purchases, and it names the fraud and payment firms it is working with, including Forter, Riskified, Sardine, Signifyd, Adyen, Checkout.com and Stripe. If you use one of them, ask how they treat agent orders. If you use something else, ask the same question.
And once a month, ask ChatGPT, Gemini and Perplexity where to buy your own products. Write down every website they name. If one is not yours, you have found the problem before a customer did. Shopify sellers should also check the AI channel settings we walked through in our piece on the Personal Agent Protocol, since those switches decide which agents can check out in your store at all.
Is this good news or bad news for small sellers?
Mostly good, if you move early. Agent shopping rewards whoever gives the AI clean, current, verifiable information, and that is one area where a small shop can act faster than a large one. Claiming a listing and publishing an official-sites page are a few hours of work. A big retailer needs a committee.
The risk is waiting. The engines fill gaps with whatever the web offers, and Podqi’s data says they currently fill a small brand’s gaps badly. ChatGPT Wallet does not create that problem, but it does make it cost money. The good part is that the fix belongs to you.
Frequently Asked Questions
Who can use ChatGPT Wallet right now?
Only ChatGPT Pro subscribers in the US who are 18 or older and have an eligible US-issued card, and only inside OpenAI’s dots agents, not ordinary ChatGPT chats. OpenAI says it plans to expand availability after learning from early use.
Does the store see the shopper’s real card number?
No. After the shopper approves a purchase with a security check such as Face ID, the card network issues a payment token with its own number, security code and expiration date, so neither the agent nor the store receives the real card details. The charge still lands on the shopper’s chosen card.
Do I have to change my website to accept ChatGPT Wallet payments?
Usually not. OpenAI says the tokens work with existing checkout flows and its cloud browser can already use most sites. You only need to act if your firewall or bot protection is blocking verified ChatGPT agent visits, or if your fraud provider is declining agent orders.
How do I stop AI chatbots sending my customers to fake versions of my store?
Claim your business in ChatGPT and upload a product feed, publish one official-sites page listing every real store and authorized retailer, keep renewing old domains you have used, and check monthly which sites ChatGPT, Gemini and Perplexity name when asked where to buy your products.
Have you asked an AI where to buy your own products yet, and did it send you to the right place?
