OpenAI has published a Commerce Partner Playbook telling businesses how to accept purchases made by ChatGPT's agents, and its main ask is that their websites stop turning those agents away. The guidance launched alongside ChatGPT Wallet, which lets a dot pay on a user's behalf with a tokenised card. OpenAI gave the reason for it in the ChatGPT Wallet announcement. Some sites may unintentionally block those purchases, it said, and risk systems may flag legitimate agent purchases, leading to declines or cancellations.

The playbook builds on Businesses in ChatGPT, the profile and product-feed tools OpenAI opened in beta last week, which form its first step.

The Four Steps OpenAI Sets Out

The announcement lists four areas, in this order. A business can claim its profile in ChatGPT and, once verified, upload a product feed with current pricing and availability. It can configure its CDN or firewall to recognise ChatGPT's signed requests. It can work with its payment and risk providers so agent purchases are recognised. It can also build a plugin that lets agents check availability or manage carts, or add WebMCP tools to its site.

Only one of them stands between a store and a sale. OpenAI says browser access does not require a plugin or product feed, and its playbook overview says browser use follows a site's existing flow with little engineering work. Product feeds apply only to businesses selling physical goods. Plugins are not yet a route to the wallet either, since OpenAI says it is still working on how third-party plugins and ChatGPT Wallet interact.

Letting the Agent Through

OpenAI's agent access guide says ChatGPT's cloud browser can already use a site and does not need adding to an allowlist. The risk is a site's own security settings blocking it by accident. ChatGPT signs its requests using Web Bot Auth, a standard for cryptographically signing automated traffic, and four providers verify those signatures automatically: Akamai, Cloudflare, HUMAN and Vercel.

Each provider lists the agent under a different name, which is what a security team has to find. Sites on other CDNs are told to check the signature headers themselves against the published standard, RFC 9421, using public keys OpenAI hosts at chatgpt.com. OpenAI's help centre adds that any proxy in between has to keep those headers intact.

How four security providers identify ChatGPT’s agent
Akamai logoAkamai
Cloudflare logoCloudflare
HUMAN logoHUMAN
Vercel logoVercel
Where it appears
A custom Verified Bot named ChatGPT Agent
dots, OpenAI, under Personal Agents
ChatGPT Agent, under Known Bots and Crawlers
ChatGPT Agent, identified as chatgpt-operator
To allow it
Set that bot’s category to Allow
Allow it, or match bot tag chatgpt-agent in a WAF rule
Allow it in the traffic policy
Nothing, unless security rules block it
Last updated 16 November 2026

OpenAI says an allowlist rule for its verified visits does not change how other bots are treated. It does not yet let site owners control which actions the agent can take, though it says it requires human review before any consequential action across all its users. Support for an agent's order is handled as if the customer had placed it, since OpenAI says customers are responsible for the actions they take.

Payments and Fraud Scoring

Wallet payments run through the checkout a business already has. OpenAI's payments guide for businesses says no Wallet-specific integration is required, and that the business handles the order, fulfilment, customer support and fraud decisions. The merchant receives a network token rather than the card number and can use the Payment Account Reference to link it to the underlying account. The customer consents only to that one transaction, so OpenAI tells merchants not to reuse or save the credential for a separate purchase.

Fraud scoring is where OpenAI expects the friction. The company is rolling out a Forwarded HTTP header that relays signals about the user behind the agent, including network, country, locale, time zone and user agent. Merchants using a third-party risk provider are told to confirm it ingests those signals, and those running their own systems to start observing the header. OpenAI cautions that the signals may vary while it rolls out support across sites.

The announcement names the companies OpenAI is working with so their systems recognise legitimate agent purchases. It does not say whether any of them has shipped that recognition yet.

The partners OpenAI named for agent purchases
  • Fraud prevention
  • Forter logoForterFraud prevention
  • Riskified logoRiskifiedFraud prevention
  • Sardine logoSardineFraud prevention
  • Signifyd logoSignifydFraud prevention
  • Payment service providers
  • Adyen logoAdyenPayment service provider
  • Checkout.com logoCheckout.comPayment service provider
  • Stripe logoStripePayment service provider
Last updated 16 November 2026

A Lighter Lift Than a Protocol

Google has taken a different route to the same checkout. Its Universal Commerce Protocol, detailed in January, is an open-source standard that retailers integrate, with the retailer remaining the merchant of record. OpenAI's playbook asks for less engineering and more configuration. For most stores, the work it describes is a firewall setting and a conversation with the risk provider they already use.

HOME
Related News