Agentic checkout is the transaction stage of an AI-assisted shopping journey. Instead of only recommending a product and sending the shopper to a website, an AI interface can help build a checkout session, exchange order details with the retailer and support completion of an authorised purchase.
That does not mean a chatbot should have unrestricted permission to spend a customer's money. The useful distinction is between an agent helping with selection and checkout preparation, and the controlled systems that validate prices, take payment and confirm an order.
For UK retailers, the opportunity is to make a good product both discoverable and purchasable through supported AI experiences. The preparation involves considerably more than adding schema markup or allowing a crawler. It requires trustworthy product information and a checkout operation that works correctly when the customer is not navigating your usual storefront.
Agentic Checkout vs Agentic Commerce
Agentic commerce describes the broader journey: researching a need, comparing options, selecting products, purchasing and potentially managing the order afterwards. Agentic checkout is the narrower step where a selection becomes a validated, authorised transaction.
A product mention in ChatGPT is therefore not an agentic checkout integration. Nor does a link from an AI answer prove that your store can accept an order inside that interface. Discovery, platform participation, checkout support and payment authorisation are separate requirements.
Our guides to agentic SEO and AI SEO for e-commerce cover the discovery side. This article focuses on what must happen after a shopper decides a product might be right for them.
How an Agentic Checkout Flow Works
The exact implementation differs by platform, but a retailer should be able to identify a clear sequence of responsibilities. An AI-generated description of an order must not replace the authoritative state held by the merchant's commerce systems.
1.Understand the request — The shopper provides constraints such as product type, size, budget and delivery location. The agent can help clarify missing details and compare suitable options.
2.Create a checkout session — A supported integration passes selected product identifiers and quantities to the merchant. The merchant validates that the items and variants exist and can be sold.
3.Calculate the actual offer — The merchant checks stock, shipping eligibility, discounts and applicable taxes, then returns an up-to-date total and delivery options.
4.Obtain appropriate authorisation — The buyer reviews the relevant order details and uses the supported approval and payment flow. Additional authentication may be needed; a product recommendation alone is not permission to charge.
5.Complete and confirm — The merchant processes the payment, accepts or declines the order and returns the confirmed state. Order updates, fulfilment and after-sales support must remain connected to that record.
Consider an illustrative request for waterproof walking shoes in a particular size, delivered in the UK for no more than £120 including delivery. A useful agent can shortlist products, but the checkout must validate the selected size, current stock and final total. If a substitution or shipping charge changes the offer, the system needs to follow its approval rules rather than silently reinterpret the shopper's request.
ACP and UCP Are Protocols, Not Guaranteed Distribution
Two important names are the Agentic Commerce Protocol, or ACP, and the Universal Commerce Protocol, or UCP. Both address machine-readable commerce interactions, but they are not interchangeable labels for universal access to every shopping assistant. A protocol's capabilities and a platform's currently enabled customer experience are different things.
Agentic Commerce Protocol
The Agentic Commerce Protocol was developed by Stripe and OpenAI as an open standard for interactions between buyers, AI agents and businesses. Its published guidance explains that merchants retain their role as merchant of record and can connect agent-facing checkout flows to their existing commerce infrastructure.
OpenAI's commerce documentation describes product feeds, merchant-controlled checkout sessions and delegated payment handling. The merchant validates the order, calculates the applicable totals, processes payment through its payment service provider and decides whether to accept the order. A checkout shown in ChatGPT does not make ChatGPT the seller.
Importantly, the protocol is open to implementation, while access to Instant Checkout in ChatGPT is subject to platform approval. OpenAI's production guidance describes approved-partner participation and launch testing. Implementing ACP does not automatically list your products or enable every country, product category or payment method.
Universal Commerce Protocol
Google's UCP documentation describes a framework for commerce integrations, including checkout and related capabilities. Its native checkout guide is a useful reminder that an AI-assisted journey need not end with an autonomous agent handling sensitive payment details.
The documented native flow lets a user, optionally assisted by an agent, build a checkout session before handing control to Google's checkout interface. The user then supplies sensitive fulfilment and payment details and submits the order without the agent participating in that final step. Do not describe that flow as an AI independently spending money on the user's behalf.
These official sources were reviewed on 9 October 2026. Before committing development time, verify the provider's current merchant eligibility, regional availability, supported products, integration version and commercial terms. UK retailers should not assume that a US launch announcement establishes UK checkout availability.
What UK Retailers Should Prepare First
Start with the data and operational foundations your store already depends on. An agent cannot reliably resolve contradictory prices, ambiguous variants or a delivery promise that your fulfilment system cannot honour.
✓Stable product and variant identifiers, with clear names, descriptions, sizes, materials and other relevant attributes.
✓Current prices and currency, with applicable VAT treatment and unavoidable delivery costs reflected correctly in the buyer-visible total.
✓Reliable stock information, destination restrictions and realistic delivery options.
✓Consistent product feeds, public product pages and checkout responses, so a recommendation does not lead to a different offer.
✓Clear seller identity, contact details, returns information, terms and privacy information that remain available in the supported purchasing journey.
✓An order-management process that can handle confirmations, tracking, cancellations, refunds and customer-service enquiries from the new channel.
Schema markup can help describe public products, but it does not create a checkout API, certify a payment integration or qualify a merchant for a platform programme. Treat those as separate workstreams with separate owners.
For UK consumer sales, review the government's online and distance-selling guidance, including required pre-contract information and applicable cancellation rights. The exact obligations and exceptions depend on what you sell and how you sell it. An AI interface does not remove the retailer's responsibilities; obtain appropriate compliance advice for your business.
Authorisation and Security Must Stay Explicit
The language model should not be the authority on the final price or whether a payment succeeded. Validate product identifiers, quantities, address eligibility and totals on the merchant's systems. A page, tool response or product description must not be allowed to override the customer's purchase constraints or the checkout's security rules.
Use the payment provider's supported credential-handling flow rather than giving an agent unrestricted access to card details. OpenAI's documented delegated payment model uses constraints such as an amount allowance and an expiry. The controls available in your own integration depend on its provider and implementation.
Tokenisation is not a blanket exemption from payment-security obligations. OpenAI's production guidance explicitly distinguishes integration approaches that may change PCI scope. Have your payment provider and compliance advisers review card-data exposure, authentication requirements, fraud controls and the specific UK payment flow.
Keep an auditable record of the selected offer, authorised total, checkout state and order result, with appropriate privacy and retention controls. Never record raw payment credentials in analytics or ordinary application logs.
Failure Handling Matters as Much as the Happy Path
An impressive demonstration can hide problems that only appear when a customer retries a request, changes an address or buys the last available item. Before launch, test failures as deliberately as successful purchases.
Stock changes. — The order must not succeed on the basis of an old product-feed snapshot when the selected variant is no longer available.
Changed totals. — Recalculate shipping, taxes and discounts when the relevant inputs change, and apply the appropriate review or authorisation rules.
Payment declines and authentication. — Show the actual outcome and provide a supported recovery path. An attempted payment is not a confirmed order.
Retries and timeouts. — Use idempotency and durable order state so a repeated completion request does not create a second charge or duplicate fulfilment.
After-sales events. — Make sure cancellation, tracking and refund updates refer to the correct order and remain consistent across the merchant and platform.
OpenAI's production checklist includes recoverable errors, stock failures, payment declines, signed order updates and idempotency testing. Those are useful evaluation questions even when you choose another platform: ask your developer or commerce provider to demonstrate the equivalent behaviour.
How to Measure an Agentic Checkout Pilot
Keep discovery metrics separate from transaction metrics. A rise in AI mentions can be encouraging, but it does not establish checkout usage or profitable sales.
Track attributable checkout sessions, validated offers, payment attempts, confirmed orders, declines, cancellations and refunds using the channel information your integration actually provides. Follow those through to margin and fulfilment costs rather than reporting gross order value alone.
Define conversion-rate denominators clearly. A rate based on people who start checkout is not directly comparable with a rate based on every product recommendation. Compare similar products, customers and reporting periods before claiming that an AI channel outperforms your website.
Use your server-side order records as the authority on completed purchases. Browser analytics may not observe every step of an embedded checkout. Do not infer a sale from a conversation or assume that a missing referral session proves no agent-assisted order occurred.
For the discovery baseline, our AI visibility tools guide explains repeated prompt testing and citation evidence. Those observations complement checkout reporting; they do not replace it.
A Sensible Next Step for a UK Store
Begin with an eligibility and readiness assessment, not a promise to launch everywhere. Confirm which supported platform and merchant programme match your market. Then review product data, payment support and the operational requirements before choosing a limited pilot.
Select a manageable product range, document the expected buyer approval flow and test it in the provider's sandbox. Include stock changes, address changes, declines and retries. Keep your normal storefront and support process available while learning from the new channel.
Agentic checkout may reduce the handoff between product discovery and purchase, but any conversion benefit needs to be measured. The strongest preparation is accurate data, clear authority to transact and reliable fulfilment—not a claim that AI will sell your products automatically.
Frequently Asked Questions
Q: What is agentic checkout?
A: It is the transaction stage of an AI-assisted shopping journey, where a supported integration connects product selection to the merchant's checkout systems and enables an appropriately authorised purchase.
Q: Does agentic checkout mean an AI can buy anything without asking?
A: No. The supported flow must respect the buyer's authorisation and the merchant's validation and payment controls. Some current experiences deliberately hand final payment and submission back to the user.
Q: Does adding Product schema enable agentic checkout?
A: No. Product schema describes public product information. Transaction support requires a supported commerce integration, merchant eligibility where applicable, payment handling and reliable order processing.
Q: Who handles refunds in OpenAI's documented checkout flow?
A: The merchant remains the merchant of record and handles refunds and chargebacks. The integration must also keep the platform's order information synchronised.
Want to connect AI product discovery with a realistic commerce-readiness plan? Panovista Marketing can help you assess product information, visibility gaps and the next steps to discuss with your commerce and payment providers.