ChatGPT Shopping for a Magento store: the US and EU paths

The question arrives in one of two shapes. From a US merchant: "how do we get into ChatGPT Shopping?" From a European one: "why can't we?" Both deserve the same answer, which is that eligibility is a programme decision, not an engineering one — and that almost everything worth doing about it is the same work either way.

What actually changed in 2026

Three facts, because the rest of this page does not make sense without them:

1. OpenAI retired Instant Checkout in March 2026, roughly five months after launching it. In-chat purchase converted poorly against the alternative — Walmart measured chat checkout at about a third of the conversion of sending the shopper to its own site, and Etsy reported agent traffic below 1% of total. The consensus moved to "discover in AI, buy on the merchant's site."
2. UCP consolidated the protocol layer. Announced at NRF in January 2026 by Google and Shopify, joined by Amazon, Meta, Microsoft, Salesforce and Stripe through April. By June 2026 it counted 8,000+ verified stores — roughly 99% of them Shopify, which tells you exactly how much of that number is platform default rather than merchant effort.
3. OpenAI's merchant programme remains US-only. No amount of correct implementation changes that from outside the US.

The practical consequence for a Magento merchant is that the checkout protocols are the volatile part and discovery is the part that works today. Build for discovery; stay ready for checkout.

If you are a US merchant

The path exists and it is mostly feed work:

1. Get the product data right first. Everything downstream reads it. Feeds are rejected for the same reasons across every programme: missing GTIN or MPN, prices that disagree with the storefront, absent availability, variants that do not resolve to a buyable SKU. In our benchmark of 1,854 live Adobe Commerce and Magento storefronts, 54% of reachable product pages fail the structured-data check — that population is the competition, and it is not a high bar.
2. Google Merchant Center, which feeds AI Mode and Gemini. Live today, no waiting list, and the discipline it imposes on your feed is the same discipline the other programmes want.
3. The Perplexity merchant programme and the OpenAI product feed, on the specifications as published.
4. Check your edge before you check anything else. A blocked crawler is a store that does not exist. If your WAF answers AI shopping agents with a 403 — which is common, because most bot rules predate them and vendors group training crawlers with shopping agents — then feeds, schema and endpoints are all sunk cost.
5. Set up measurement now. Roughly 71% of ChatGPT-referred visits arrive in GA4 as Direct, Google's AI Mode deliberately sends noreferrer, and GA4's "AI Assistant" channel (added 13 May 2026) classifies on the Referer header, so it understates by design. Server logs are the honest source; SKU-cohort differential measurement is how you defend the budget. Data not collected this year cannot be collected retroactively next year.

None of these steps require Instant Checkout, and none of them are wasted if the checkout layer changes shape again.

If you are an EU merchant

You cannot enrol in the US-only programme. You can do everything else, and the everything else is where the traffic is:

• Google AI Mode and Gemini through Merchant Center — open to EU merchants, and the largest AI discovery surface that is.
• Perplexity — its merchant programme is the second live channel.
• Feed and catalog hygiene — the work that makes you eligible on the day OpenAI opens, done in advance rather than in a panic. Enrolment windows are not the constraint; a clean catalog is, and it takes months, not days.
• On-site agent readability — llms.txt, a UCP manifest, an MCP endpoint, complete product JSON-LD, an explicit AI-crawler policy. This is what an agent that arrives without a programme sees, and it is the majority of arrivals.

One thing to be clear about, because it is where European merchants are most often misled: being in ChatGPT's answers and being in ChatGPT's merchant programme are different things. The model browses. It reads pages, and it reads structured data on those pages. A European store with clean data and an open edge is cited in AI answers today, with no programme membership at all. The programme adds a commerce surface, not the ability to be found.

What we honestly do not know

Whether or when OpenAI opens the programme outside the US. Anyone who tells you a date is guessing. That uncertainty is the argument for the feed-first order of work above — every step of it pays off in Google and Perplexity now, and none of it becomes worthless if the programme never opens or opens with different requirements.

Keep the volume honest too. Peer-reviewed measurement (Kaiser & Schulze, Marketing Science 45(4), April 2026; 973 stores, 6 billion sessions) puts LLM-referred traffic under 0.2% of total, converting above paid social and below organic search, with no meaningful difference in average order value. Checkout.com put agent involvement at about 3% of transactions in June 2026. Growth rates in this channel are quoted without denominators far too often. It is a cheap channel to be readable in and an expensive one to bet a quarter on.

Where to start

Run the free scan: eight checks against your live storefront, no charge, no call, and it will tell you within a minute whether your edge blocks agents and whether your product data survives contact with one. Our own reference storefront, Nord Trail, passes all eight on plain Magento and is re-scanned every morning by that same scanner.

If you want the whole path owned end to end — data, feeds, programme applications, edge policy, measurement — that is the Readiness Sprint: five business days, fixed scope, published price, no replatforming. If you want the tooling and your own team to drive, Agentic Commerce Suite ships the feeds for Google Merchant Center, Perplexity and OpenAI, schema repair and the agent endpoints, at $399 once per installation.

Questions

Can a European store appear in ChatGPT at all?

Yes. Being cited in an answer needs readable pages and an edge that serves the model's crawler, not programme membership. The US-only restriction applies to the merchant programme and its commerce surface.

Should we set up a US entity to enrol?

That is a tax, legal and fulfilment decision with consequences far outside this channel, and on today's volumes it is not one we would make for AI shopping alone. Ask again when the channel is a material share of revenue rather than a fraction of a percent.

Is a product feed to ChatGPT worth it if Instant Checkout is gone?

Yes, for discovery. The feed is how the model knows your prices, availability and identifiers; the ACP specification is explicit that "feed prices are discovery-time signals" and that checkout responses remain authoritative. Nothing about the retirement of in-chat purchase makes your catalog less worth reading.

We already sell through Google Shopping. Are we done?

You are a long way ahead, and reusing that feed is the fastest route to the other programmes. What Merchant Center does not check is what happens on your own site when an agent arrives — structured data on the page, llms.txt, an MCP endpoint, and whether your WAF lets it in at all. That is what the free scan measures.