A shopper tells ChatGPT to buy the jacket, taps "buy for me" in Google's AI Mode, lets Amazon's Buy for Me place an order on a brand's own site, or checks out with Perplexity's Buy with Pro. Two weeks later the jacket comes back. The question a brand then has to answer, "how do you handle returns and refunds for purchases made by AI agents", has a short answer that every one of those platforms has published: the brand does, as the merchant of record, under the same rights and the same policy as any other order. What changes is who the customer is talking to and what the return flow has to expose so an agent can run it. This page sets that out from the platforms' own published pages, fetched on 2026-10-22, and from the live Pango guides on return rights; nothing here is a prediction.
Short answer: On every published agentic checkout, the brand stays the seller of record and handles the return. OpenAI's Agentic Commerce Protocol documentation states that "OpenAI is not the merchant of record" and that the merchant "is responsible for handling refunds and chargebacks, as you accepted the payment directly from the customer as the merchant of record", with an order-update webhook to keep ChatGPT in sync; Amazon's Buy for Me page states that "delivery, returns and exchanges, and customer service are managed by the brand store"; Google writes that "no matter which way you buy, the brand stays the merchant of record" and that for a return "you and the merchant are looking at the same record". So the customer of record is the shopper, the return rights are the ones that already apply to a distance sale (the EU's 14-day withdrawal right in the EU, the FTC's rule on shipping time in the US), and the work is to make eligibility, the exchange offer, the label, the status and the refund timing available to the agent as well as to the person.
Free for DTC brands: Pango runs the shipping and returns audit on your own order and carrier data as part of a demo: fifteen numbers across checkout, carriers, warehouse handover, tracking and returns, the three that matter most for your operation, and the same orders shown running on one system. Book the free audit.
Who the customer of record is, platform by platform
| Platform | What its own pages say about the order | What that means for returns |
|---|---|---|
| ChatGPT Instant Checkout (OpenAI, Agentic Commerce Protocol) | "OpenAI is not the merchant of record in the Agentic Commerce Protocol." The merchant "performs validation, determines fulfillment options, calculates and charges sales tax, analyzes payment and risk signals on their own stack, and charges the payment method with their existing payment processor." Refunds and chargebacks: "Your platform is responsible for handling refunds and chargebacks, as you accepted the payment directly from the customer as the merchant of record." | The shopper is your customer and paid you. You run the return under your policy and tell ChatGPT the outcome: "Use the ORDER_UPDATE webhook to notify ChatGPT (or any integrated partner) when a refund or chargeback status changes so order state stays synchronized." |
| Amazon Buy for Me | Amazon "makes the purchase by securely providing the customer's encrypted name, address, and payment details to complete the checkout process on the brand's website." "Delivery, returns and exchanges, and customer service are managed by the brand store, and customers can contact the brand store with any questions regarding their order." The customer "will receive an order confirmation email from the brand store" and can track it in the Amazon app's Buy for Me Orders tab. Its Terms of Use make the purchase subject to the merchant's terms and policies, "including, but not limited to, those relating to shipping, returns". | The order is a normal order on your site with the shopper's details. Your return policy applies, not Amazon's, and the shopper contacts you. |
| Google agentic checkout (Universal Commerce Protocol, AI Mode, Gemini, Universal Cart) | "Retailers remain the seller of record." "No matter which way you buy, the brand stays the merchant of record." "If you ever need to make a return, you and the merchant are looking at the same record." | The shopper and the brand share one order record; the return runs on it under your policy. |
| Perplexity Buy with Pro | Pro users in the US check out on Perplexity's website or app "for select products from select merchants"; if Buy with Pro is unavailable, "we'll redirect you to the merchant's website to complete your purchase." Perplexity's page does not describe returns or refunds. | The order is placed with the merchant; the return follows the merchant's policy, as on any order placed from a referral. |
The pattern is the same on all four: the platform places or hands off the order, the brand sells and is paid, and the brand owns fulfilment, support and returns. None of the four publishes a return policy of its own for these orders.
How return rights apply
The rights do not change because an agent pressed the button; a distance sale to a consumer is a distance sale. Two live guides set out the rules the brand already ships under:
- In the EU, the right of withdrawal lets a shopper cancel most online orders within 14 days, for any reason, and get their money back; it is a statutory right under the Consumer Rights Directive, not a store policy you set, and the refund timing, shipping-cost rules and local-language messaging differ by country. The rule and its country variants are in EU right of withdrawal.
- In the US, the FTC's Mail, Internet, or Telephone Order Merchandise Rule requires a seller to have a reasonable basis for expecting it can ship within the advertised time frame or, if none is stated, within 30 days, and to seek the customer's consent to a delay; it governs the shipping promise rather than the return, and the return policy is the brand's own, stated at the point of sale. The rule and the carriers' published windows are in shipping in the US for DTC brands.
Two consequences follow for agent-placed orders. First, the policy the agent showed the shopper at checkout is the policy you are held to, so the policy text the platform pulls from your site or feed has to be the current one. Second, the withdrawal clock and the refund clock run from the same events as on any order, delivery and receipt of the return, and the shopper can exercise the right through you directly even if the agent placed the order.
What the return flow has to expose to an agent
An agent that placed the order will be asked to start the return. For that to work without a support ticket, the same six things a self-serve portal shows a person have to be available to the agent, through an API or a structured page it can read:
| What the agent needs | What it is | Where it lives today for a person |
|---|---|---|
| The order and its status | The order record with the shipment's current milestone, delivered date and the items | The tracking page and the order page |
| Eligibility per item | Window, final-sale items, condition rules, country | The portal's first step |
| The exchange offer | The right size, another product or store credit, from live stock, before a refund | The portal's exchange step |
| The label or drop-off | A prepaid label, a QR code or an in-person drop-off, booked with the carrier | The portal's label step |
| Refund timing and form | When the money goes back (request, first scan, receipt, inspection) and in what form | The policy, and the status page after the return is sent |
| The return's status | The return parcel's tracking and the refund or exchange state | The status page |
The customer-facing flow and the brand-facing rules behind it are set out in returns portal software; the decisions that can already run without a person, in the order to automate them, are in how to automate returns with AI. OpenAI's protocol already defines the order-update side of this: the merchant reports refund and chargeback status changes back so the order state in ChatGPT matches the brand's. Google describes the same outcome from the shopper's side, one record both parties look at.
Fraud and identity checks
An agent-placed order arrives with the shopper's details supplied by the platform: Amazon describes providing "the customer's encrypted name, address, and payment details"; OpenAI's merchant "analyzes payment and risk signals on their own stack". The return side needs the same discipline as any order, and two things more. First, tie the return request to the order and the paying customer, not to whoever presents the order number: the shopper's email on the order, the confirmation the brand sent, or the authenticated session the agent carries. Second, keep the usual abuse controls, returns per customer, photo evidence on faulty claims, holds on inconsistent cases, and apply them to the customer, not to the platform, because the same shopper may order through several agents. Where a platform offers a payment mandate or an authenticated order reference, use it as the identity anchor; where it does not, the order confirmation you sent is the anchor.
How to prepare
- Confirm you are the merchant of record on every agentic channel you join, and that your return policy, in the shopper's language, is what the platform shows at checkout. All four platforms above say it is yours.
- Make the return policy machine-readable: windows, exclusions, exchange rules, refund timing, per country, in one place your site, your feed and an API can serve.
- Expose the return flow as an API or a structured page: order status, eligibility, the exchange offer from live stock, label or QR booking, refund timing, return status.
- Report outcomes back: on OpenAI's protocol through the order-update webhook; on the others through the order record the platform and the shopper share.
- Anchor identity on the order and the customer, and run the same abuse controls per customer across channels.
- Run the return on the same record as the order, so the agent, the shopper and your team see one status.
How Pango does it
Pango runs returns on the same record as checkout delivery, shipping, tracking and the warehouse, with natural-language return and exchange rules compiled into per-merchant workflows, exchanges to any product in the store, and labels, cross-border documents and refund logic per merchant and per country; it exposes APIs so technical teams can build on its shipping capabilities, generate labels, retrieve real-time data and track shipments. For an agent-placed order that means the order, the shipment, the eligibility decision, the exchange offer, the label and the refund timing sit on one record that a person, the brand's team or an agent can read, and the outcome is reported from that record. The operating model behind it is in agentic post-purchase operations, and what an autonomous return does step by step is in autonomous returns.
The bottom line
Returns for orders placed by AI agents are the brand's returns: every published agentic checkout keeps the brand as the merchant of record, the shopper's rights are the ones that already apply, and the policy shown at checkout is the one you are held to. What is new is the reader: the return flow has to be exposed in a form an agent can run, and the outcome has to be reported back to the record the platform and the shopper share.
