Returns and Refunds in Agentic Commerce: Who Handles What When an Agent Buys (2026)

When an AI agent buys from your store, you stay the merchant of record, which means returns, refunds, and support stay your process and your policy. Under the agentic-commerce flows shipping in 2026 (OpenAI's Instant Checkout and the Agentic Commerce Protocol it is built on), the agent is a checkout surface, not a reseller. The order is yours. The customer is yours. The refund, when one is needed, is issued by you under the same policy you already run. What changes is not who owns the post-purchase relationship, it is how the buyer reaches you, what they expect, and how clean your order and contact data has to be for the whole thing to work.

That distinction is the reassuring part, and it is worth stating plainly before the caveats: agentic commerce does not hand your customer service to a third party. But because the buyer may have discovered, compared, and paid entirely inside a chat, the post-purchase experience has new failure modes. This guide covers who handles what, what actually changes, and a readiness checklist, with the honest caveat that this space is early and the specifics keep moving. Verify anything program-specific against the current official docs rather than treating any rule here as fixed.

Who is the merchant of record when an agent buys?

In the ACP-style flows live as of mid-2026, you are. The protocol is designed so that the agent (ChatGPT, or another assistant built on the same rails) presents your products and collects payment, but the sale completes into your store as a normal order. Practically, that means:

• The order lands in your Shopify admin like any other, with line items, a customer record, and a payment you can refund through your normal tools.
• Your return and refund policy governs it. There is no separate "agent returns policy" you have to write. The terms a shopper agreed to are yours.
• Your fulfillment, your shipping, your support handle the order end to end. The agent does not warehouse, ship, or adjudicate anything.

This is the single fact that reassures most merchants: agentic commerce is a new front door, not a new middleman between you and your customer. The economics, the liability, and the relationship stay where they were. Treat the exact division of responsibility as something to confirm against OpenAI's and Shopify's current merchant documentation, because the programs are young and the fine print evolves, but the merchant-of-record principle has been the consistent design intent.

What actually changes after the sale

The ownership does not move. The buyer's experience does. Four things are genuinely different when a sale arrives through an agent:

1. The buyer may never have seen your site. They found the product, compared it, and paid inside a chat interface. They may not know your brand name cold, may not have an account, and may not think to go to yourwebsite.com/returns when something is wrong. Their mental model of "where I bought this" might be "ChatGPT," not you.
2. They may expect to initiate support through the agent. A shopper who did everything in-chat may naturally return to the chat to say "this didn't fit." Whether and how that routes back to you depends on the program's current capabilities, which are still maturing. Do not assume the agent will hand you a support ticket cleanly, and do not assume it will not.
3. Contact and order data has to be clean. Your ability to reach the real customer (order confirmation, shipping updates, return instructions) depends on the email and contact details flowing through correctly from the agent handoff. A malformed or agent-proxied email address is the difference between a smooth return and a customer who cannot find you.
4. Less site familiarity means more first-contact confusion. A returning direct shopper knows your returns portal. An agent-sourced first-time buyer does not. Your post-purchase emails carry more of the load, because they may be the buyer's first and clearest link back to you.

None of this breaks the model. It just means the post-purchase touchpoints you may have under-invested in (the confirmation email, the returns instructions, the support contact path) matter more when a share of buyers arrive with no site relationship.

Making your return and refund policy machine-readable

Here is the part that folds post-purchase back into acquisition: your return policy is not just a support document, it is a signal an agent reads when deciding whether to recommend you. Assistants building a shortlist weigh risk to the shopper, and a clear, generous, easy-to-parse return policy lowers that risk.

• State the policy in plain, structured text on a real page, not locked inside an image or a script-loaded widget a crawler never executes. Return window, condition requirements, who pays return shipping, refund timeline: each stated directly.
• Keep it consistent everywhere. The policy on your site, in your Shopify settings, and in any feed or program data should agree. Contradictions make you a fuzzier, riskier entity to recommend.
• Lead with the reassuring facts. "30-day free returns, refund within 5 business days" is exactly the kind of concrete, quotable line an assistant lifts into an answer to reassure a hesitant buyer.

A vague or punitive return policy does not just cost you a return, it can cost you the recommendation, because the agent has less evidence that choosing you is safe for the shopper.

The trust angle: returns as a recommendation signal

Generous, clearly stated return terms function the way deep review volume does: as trust evidence. When two products are close, the one whose return policy visibly de-risks the purchase is the easier, safer answer for an assistant to give. This is not a trick to game, it is the ordinary logic of recommendation. The systems are trying to send shoppers somewhere they will not regret, and a merchant who stands behind the sale is, by construction, the lower-regret choice.

There is a conversion side to this that matters because agent traffic behaves differently from a cold click. A shopper the agent sends arrives pre-qualified and high-intent: they were compared, chosen, and often already sold on the need. The product page's remaining job is closing, and which reviews, UGC, and trust sections you surface (return guarantees, ratings, real customer proof), and in what order, decides how well it closes. This is what Eevy does: it continuously optimizes which social proof each shopper sees on your product pages with a genetic algorithm, evolving toward the combinations that actually convert, and stores running it lift conversion by about 18% on average. The same optimized proof renders as real on-page HTML, so it doubles as the trust evidence AI crawlers read. There is a permanent free plan up to 25,000 monthly visitors, then plans from $99/mo. Tool or no tool, the principle holds: the on-page evidence that closes the human is the same evidence that reassures the agent.

What does not change (and where merchants overthink it)

A few worries come up that the current design does not warrant:

• You do not need a separate returns team for agent orders. They are normal orders in your admin. Your existing returns flow handles them.
• You are not liable for the agent's conversation. You are responsible for the product and the fulfillment, as always. What the agent said in chat before the handoff is the agent's surface, not your policy document.
• You do not rewrite your refund terms. One clear policy serves human shoppers and agent-sourced shoppers alike. The work is making it clear and reachable, not making it different.

The mistake in the other direction is assuming nothing changed at all. It did: the buyer's path back to you got longer and less familiar, so the post-purchase communications have to be sharper.

A practical readiness checklist

Concrete steps, roughly in priority order:

1. Verify order and contact data flows cleanly from any agent-checkout handoff into your store. Send yourself a test order if the program allows it, and confirm the confirmation email reaches a real, reachable address.
2. Make your return and refund policy a clean, crawlable page with return window, conditions, shipping responsibility, and refund timeline stated in plain text.
3. Audit your order confirmation email. Assume it may be the buyer's clearest link back to you. It should name your brand, restate the return window, and give an obvious support and returns path.
4. Make your support contact reachable from the confirmation and the site, not buried. An agent-sourced buyer who cannot find how to reach you is a chargeback or a lost customer.
5. Keep policy consistent across your site, Shopify settings, and any program data, so nothing contradicts.
6. Do not over-engineer. No separate agent-returns policy, no new team. Clean data, clear policy, reachable support.
7. Re-check the official docs periodically. Instant Checkout and ACP capabilities (including how support requests may route back to merchants) are still evolving in 2026.

The honest summary: agentic commerce keeps you as the merchant of record, so returns, refunds, and support stay your process under your policy. What it asks of you is unglamorous and entirely within your control: clean order data, a clear and generous return policy stated where both shoppers and crawlers can read it, and a confirmation email that reaches the real customer. Do that, and the same clarity that makes a return painless also makes an agent more willing to recommend you in the first place.

Related Reading

• Agentic Commerce Protocol on Shopify: how the protocol behind agent checkout works and what it means for merchants.
• OpenAI Instant Checkout for Shopify: the checkout surface where these agent purchases actually complete.
• AI Agent Trust and Fraud in Ecommerce: the wider trust, verification, and risk picture around agent-driven purchases.