On November 13, 2026, OpenAI published "Introducing ChatGPT Wallet," with a subtitle worth reading twice: "Building trusted agentic payments with networks, banks, and businesses." The trade named in the title — Agents Shop, You Approve Every Purchase — is not a slogan but the literal mechanics of the release: the payment network issues a payment token instead of exposing a card number, and the approval step sits on the user's own fingerprint or face. AI buying things on a user's behalf stops being a question of "can it happen" and becomes a question of "how does it get approved." The official positioning sentence deserves to be read in full: "It's the first wallet that lets agents pay without access to users' actual card details and that requires a security check for every purchase, while also preserving existing card rewards and protections for eligible disputes." Two qualifiers ride inside it — the agent never touches real card details, and every purchase requires a security check — and neither can be dropped when summarizing. This article checks only what the official announcement page, the dots page, and the help center support verbatim.
The launch scope: five qualifiers, and none of them is optional
The cooperation and availability sentence reads, verbatim: "Built in partnership with American Express, Mastercard, and Visa, ChatGPT Wallet will initially be available through dots for ChatGPT Pro subscribers in the US." A second sentence adds the age and card conditions: "aged 18 and older in the US with eligible US-issued cards." Stitched together, the initial availability is fenced by five conditions at once: the channel is dots, the tier is ChatGPT Pro, the region is the US, the age is 18 and older, and the cards are eligible US-issued cards. Drop any one of them in a retelling and the scope warps. OpenAI adds its own reason for the narrow start: "We're starting with limited availability to ChatGPT Pro users, so we can learn from real-world use and work with businesses, banks, and payment networks to improve compatibility and reliability." That reason is quotable and does not need to be invented on the company's behalf.
What dots is needs spelling out, because dots is the launch channel itself. The official dots announcement page dates the product September 29, 2026, and describes it verbatim as "Dots are remarkably capable, always-on agents built to handle everything," powered by GPT-6 Astra. The cost clause is explicit: "Your first dot is included in your Pro or Business Premium plan at no extra cost." An official update on November 3, 2026, adds that the family of frontier models powering dots is being expanded to include GPT-6.1 Terra. The dot that completes purchases did not appear alongside Wallet as something new; it is an always-on agent published at the end of September, and Wallet plugs payment capability into it.
The three card networks each left a quote on the announcement page. For American Express, Luke Gebb, EVP and Head of Global Innovation, said verbatim: "ChatGPT Wallet marks our first American Express Agentic Commerce Experiences integration, enabling eligible Card Members to use their Amex Card while reviewing and approving each purchase." For Mastercard, Sherri Haymond, EVP, Global Head of Digital Commercialization, said verbatim: "An agent making a purchase should strengthen, not bypass, the trust built into payments." For Visa, Rubail Birwadker, Head of Global Growth, said verbatim: "We're partnering with OpenAI to bring Visa's trusted infrastructure to the experience, helping connect user-approved purchases with the systems businesses already use." The common denominator across the three quotes is the vocabulary of the approval chain in the next section: reviewing and approving each purchase, strengthening rather than bypassing, user-approved — the approval decision stays with the person.
The scale number appears once, and it carries a label: the official page self-reports that "1.2B people use ChatGPT each week, and many of them turn to it to research products, compare options, and decide what to buy." This is OpenAI's own self-reported figure, with no third-party verification on the page; it is written here as 1.2B, verbatim, neither inflated nor discounted.
The subscription context comes from the current wording of the official help center article 9793128: ChatGPT Pro has three monthly tiers, Pro 100 at $100, Pro 200 at $200, and Pro 500 at $500. There is a trap to avoid: an older announcement described "ChatGPT Pro at $200 USD/month," which predates the tier split and must not be reused as the current price. When a price is needed, it is written the way the help center writes it — three tiers, Pro 100, Pro 200, Pro 500. How the tiers differ belongs to pricing-selection coverage.
One approval, end to end: select the card, run the check, issue the token
The official description of the purchase flow breaks into four steps, each with text to point at.
Step one, add a card once. Verbatim: "Users add a card once, then approve each purchase without re-entering their payment details." But "no re-entry" does not mean "no approval." The remaining three steps put the limit back into that sentence.
Step two, approve each purchase separately. Verbatim: "Before each purchase, the user reviews the items, total, and selected card, then approves it with a security check such as Face ID or a fingerprint scan." Two details matter. First, "such as" is exemplary language: Face ID and a fingerprint scan are the two kinds of check OpenAI names as samples, not an exhaustive list, and writing that these are the only two methods would overstate the text. Second, the sentence continues: "Each approval creates a clear record of the purchase the user authorized the agent to make." Every approval leaves a clear record of which purchase the user authorized. That is the official source of the phrase Approve Every Purchase: approval is per purchase, with no opening for one approval to cover the rest.
Step three, the payment network issues the payment token. Verbatim: "After the user approves, the payment network provides a payment token — a substitute for the user's actual card details — with a card number, security code, and expiration date. The payment is charged to their selected card without ChatGPT Wallet sharing their actual card details with the business or agent." Across the whole chain, the card number exists only inside the network-side stand-in; what the person and the machine each hold is a credential for one approved charge, not the card itself. Step four: "Users keep their card's existing rewards and protections for eligible disputes with businesses, for example when an item is defective." This step is the one most often dropped in retellings, yet it says the mechanism does not move the user out of the existing card system; it only changes how the payment credential is submitted.
Two wording boundaries need holding. First, the official page says "approve each purchase," "requires a security check for every purchase," and "user-approved purchases"; the phrase "explicit approval" does not appear verbatim in it, so paraphrases should not be quoted as official text. Second, whether the payment token is single-transaction specific is not stated verbatim on the official page, which lists only the three contents — card number, security code, expiration date. The finer "scoped to a given transaction" phrasing exists only in a media interview retelling, and this article does not cite it as an official term.
The future direction is written in the expectant mood rather than the present tense: "Over time, we expect users to be able to delegate purchases entirely to their agent by setting clear parameters — such as spending limits, purchase categories, or time periods." Fully delegating purchases to an agent, pre-approved through such parameters with what the official text calls enhanced safeguards, sits entirely inside "we expect." It is not a shipped feature, and retellings must not turn "will do" into "already has."
Read together, the four steps form a deliberately narrow authorization list: the card number does not leave, approvals are logged per purchase, the charge still lands on the selected card, and card benefits stay intact. Every link has official verbatim text behind it.
Why merchants have to move too: CDN, risk, and PSP layers
The user-side chain is half the engineering. The other half is on the merchant side, and the official background sentence is candid about it: "Agentic commerce is still in its early days. As agents begin to take on more shopping and purchasing tasks, many websites and payment systems are still adapting to this modality. Some sites may unintentionally block purchases through ChatGPT Wallet or other agentic payment products, and we are focused on helping businesses make that transition." The overall requirement comes in two clauses: "To accept payment from ChatGPT Wallet, websites need to ensure agents are not inadvertently blocked, and payment and risk systems need to recognize legitimate purchases made by agents." Neither clause is optional.
The specific guidance comes in four items, each pointing at a retrofit point in an existing system. Product information: claim your business in ChatGPT and, once verified, update business details and upload a product feed with current pricing and availability. Agent access: "Configure your CDN or firewall to recognize signed requests using Web Bot Auth and confirm your general site security measures are not inadvertently blocking ChatGPT agents. Browser access does not require a plugin or product feed." Payment and risk: "Wallet's payment tokens work with existing checkout flows, but risk systems may still flag legitimate purchases made by agents, leading to declines or cancellations. You can work with your providers to understand how they recognize these transactions." Extensions: a plugin can be built to let agents check availability, manage carts, or take other actions through existing systems, and websites can also add WebMCP tools.
The partner list is given verbatim in one sentence: "We're also working with fraud prevention partners including Forter, Riskified, Sardine and Signifyd and payment service providers including Adyen, Checkout.com, and Stripe to help their systems recognize legitimate purchases made by agents." On the fraud-prevention side the names are Forter, Riskified, Sardine, and Signifyd; on the payment service provider side they are Adyen, Checkout.com, and Stripe, with one stated purpose: helping those systems recognize legitimate agent purchases. Vendors absent from the list are not added here, and none on it are removed. Alongside the announcement, OpenAI published a Commerce Partner Playbook, described on the page as covering making product information available, ensuring agent access, and working with payment and risk providers to reduce unnecessary declines; the playbook's full text was not verified verbatim for this article, so nothing beyond that summary is elaborated.
The division of labor across the ecosystem is stated too: "Banks and networks bring their expertise to evaluating transactions. Users transact directly with businesses, which keep their customer relationships and control over pricing, fulfillment, returns, and customer service." OpenAI places itself at the rails-and-verification layer, not at the layer of running the merchant's business. And none of the four guidance items asks a merchant to rewrite its checkout system: "Wallet's payment tokens work with existing checkout flows" is the official sentence; the retrofit concentrates on recognition and allow-listing, not on tearing out the register.
What the official page does not say: fees, rates, and timelines stay blank
As of November 16, 2026, the official announcement page carries no statement on any of the following, and this article writes each one as "not mentioned" rather than supplying a number. First, fees: whether Wallet charges users anything extra, what merchants pay per transaction, whether an integration fee exists — the page contains no fee clause of any kind. A media interview retelling has circulated a claim that OpenAI is not taking a cut of transactions to start; that is an interview account, not a clause on the official page, and it is not cited here as fact. Second, timelines: the number of users in the initial rollout, a timetable for widening availability, and plans for markets outside the US are all unmentioned; the official text offers only the status word "limited availability." Third, support documentation: as of the retrieval point, no dedicated ChatGPT Wallet help-center article could be found, and a guessed article ID returned 404; refund terms, limits, and card-unlinking procedures are unpublished. Fourth, the card list: beyond American Express, Mastercard, and Visa, the specific set of supported cards is not mentioned.
Listing the silences separately is not fault-finding; it is the actual boundary of this announcement. What can be checked is the authorization mechanism and the ecosystem roster; what cannot be checked is the money and the calendar.
The division of labor against existing in-site write-ups is worth stating once. The DeepSeek full-suite App APIs article covers the developer-side ledger: capabilities attached to an API, tool usage fees billed on top of tokens. This article covers the payment-side ledger: how the approval chain runs and where the card number gets substituted. The preview-state products article is about reading "limited availability" style product states in general; this article answers only for one announcement. The pricing-selection article handles model unit prices and plan choices; this article touches the Pro tier prices only while stating the subscription background. The free-tier comparison article keeps the API-side quota ledger, a different layer from a payment authorization chain. As of November 16, 2026, what can be checked is what is written above; the rest is left blank rather than filled in.
Related reading