Is Your Checkout Agent-Ready? A GCC Merchant Audit Checklist (UCP, ACP, Visa TAP, Mastercard Agent Pay)

An agent-ready checkout is not an in-chat buy button. For a GCC merchant in 2026 it means four things: AI agents can discover accurate product data, verified agents get past your bot defences, agents pay with tokens instead of card numbers, and you keep evidence for disputes. Score yourself on the 10-point checklist below. Most merchants we talk to land in the middle, and the gaps are usually fixable in weeks.

The protocol news moves fast, so this post sticks to what has been announced by the networks and platforms themselves, and flags where things are still open.

Why should GCC merchants start with discovery, not checkout?

Because the in-chat checkout bet has already wobbled. OpenAI launched ChatGPT Instant Checkout in September 2025 on the Agentic Commerce Protocol (ACP) it built with Stripe. By March 2026, industry reporting said OpenAI was pulling back: only a small number of merchants had gone live, keeping prices and stock accurate across merchants proved hard, and shoppers tended to research in ChatGPT and then buy on familiar sites. The reported new direction is discovery inside ChatGPT, with the purchase completed in the merchant’s own app or website.

That is good news for most GCC retailers. You don’t need to hand your checkout to a chat surface. You need agents to find you, read you correctly and send the shopper to a checkout that works.

Google and Shopify took a different path with the Universal Commerce Protocol (UCP), announced at NRF on 11 January 2026, which lets shoppers buy inside Google AI Mode and the Gemini app while the retailer stays merchant of record. Native checkout there has rolled out market by market, starting in the US, and we haven’t seen a GCC launch date. Either way, both protocols start from the same place: a clean, current product feed. Our explainers on UCP and ACP cover the mechanics.

What does the agent-readiness checklist look like?

Score each line 0 (not started), 1 (partial) or 2 (done). The maximum is 20.

How to read your score:

How do you let good agents in without letting scrapers in?

This is where many GCC merchants fail without noticing. Bot defences tuned to block automation will happily block a legitimate shopping agent too, and the shopper just sees a failed task.

Visa’s answer is the Trusted Agent Protocol (TAP), built with Cloudflare on top of Web Bot Auth. In plain terms, registered agents sign each HTTP request using HTTP Message Signatures (RFC 9421), and publish their public keys in a directory. The signature carries a timestamp and a nonce to stop replays, plus a tag that says whether the agent is browsing or paying. Cloudflare says Mastercard has built Web Bot Auth into Mastercard Agent Pay, and that American Express will use it too.

What that means for your audit:

• If you’re on Cloudflare, it validates these signatures for you, and it has described managed rules to allow registered Visa and Mastercard agents while your other bot rules still apply.
• If you’re on another CDN or WAF, ask your vendor directly whether they verify Web Bot Auth signatures today. If not, you’re relying on user agents and IP lists, which are easy to spoof.
• Either way, decide your policy: what verified browsing agents can do, what verified paying agents can do, and what happens to unverified automation.

One caveat: Web Bot Auth is still an IETF draft, so expect details to change. Build the policy now and keep the implementation behind your CDN or WAF vendor.

How should agents actually pay you?

With tokens, not card numbers. Mastercard Agent Pay, announced in April 2025, introduced Mastercard Agentic Tokens, built on the same tokenization behind card-on-file and mobile wallets. Visa’s equivalent sits in Visa Intelligent Commerce, which uses Visa’s Token Management Service.

Both networks already have UAE reference points. Mastercard said its first Agent Pay transaction outside the US happened in the UAE in November 2025, with Majid Al Futtaim and the fintech Dataiera, and VOX Cinemas ticket booking described as an early use case. Visa announced what it called the first live Visa Intelligent Commerce implementation in the region with Aldar in December 2025, and launched its Agentic Ready issuer programme in the UAE in May 2026 (we covered what issuers must build).

As a merchant, you rarely integrate with the networks directly. The practical question is for your PSP or acquirer: do you accept network tokens today, and when will you support agentic credentials from Visa and Mastercard? Get the answer in writing.

The other reason tokens matter is PCI scope. If an agent types a full card number into your checkout, the card data flow just got longer and harder to defend. Our sister practice explains the scoping logic in detail in PCI DSS scope for agentic commerce.

What PCI evidence does an agent checkout need?

The same as any e-commerce payment page, with one more reason to have it. PCI DSS requirements 6.4.3 (authorise and inventory every script on the payment page) and 11.6.1 (detect unauthorised changes to payment page scripts and security headers) apply to merchants filling SAQ D or going through an onsite assessment. SAQ A merchants instead confirm an eligibility criterion that their site is not susceptible to script attacks.

Why it matters more with agents: agent traffic multiplies the automated sessions hitting your checkout, and your acquirer will ask how you know the page those sessions load is the page you intended. A current script inventory and change alerts are the evidence. If you can’t produce them, score row 8 as zero.

Who pays when an agent’s purchase is disputed?

Be honest with yourself here: nobody knows yet. We have not found a published Visa or Mastercard rule that specifically assigns chargeback liability for agent-initiated purchases, so your existing dispute rules apply by default. Mastercard’s own launch materials talk about helping clarify agentic transactions that look unfamiliar to the cardholder, which tells you the networks see this as a real risk.

The new dispute pattern to expect is “my agent bought it, but I didn’t want it”. Your defence is evidence:

1. Agent identity: the verified signature or agent ID on the session.
2. Authorisation: the token used and any authentication result returned by your PSP.
3. What the agent saw: price, delivery date and returns terms at the time of order.
4. What the customer was told: the order confirmation email or SMS, with timestamp.
5. Fulfilment: delivery or collection proof.

Most of this you already have. The work is tagging agent orders (row 9) so you can pull the pack quickly (row 10).

How do you sequence the fixes?

1. Fix the feed and policies first. It helps you with every agent surface and with ordinary search.
2. Audit your bot rules. Find out what happens today when a signed agent request hits your checkout.
3. Get your PSP’s token roadmap in writing.
4. Close the PCI payment-page evidence gap.
5. Tag agent orders and define the dispute pack.
6. Pilot with one protocol or network programme where your PSP is ready, not all of them at once.

Book a fixed-fee agent-readiness checkout audit

We run a fixed-scope agent-readiness checkout audit for GCC merchants. We score your store against this checklist with real evidence: feed and policy data, how your WAF treats signed agent traffic, your PSP’s token and agentic credential support, payment-page script controls and your dispute evidence. You get a scorecard, a prioritised fix list your engineering and payments teams can execute, and a recommendation on which protocol or network pilot to try first.

See our agentic payments consulting service, or get in touch to scope an audit.