Agentic Commerce — How to Sell When the Buyer Is an Agent (ChatGPT Checkout, ACP, AP2, UCP)

Let me start with the thing you won't read in most "AI shopping revolution" pieces, because they were written before March: OpenAI pulled back from native Instant Checkout. "Buy it in ChatGPT", launched on February 16, 2026 — a purchase completed entirely inside the chat — was rebuilt in March: products still appear in the conversation, but you complete the transaction in the merchant's store, through an in-app browser or a separate tab, on Shopify's or Etsy's checkout. CNBC called it plainly a revamp after a rough start; Shopify talks about "agentic storefronts". The first lesson of this post for any store follows directly: the AI transaction layer is in motion, and anything you build exclusively for one checkout variant can go stale within a quarter.

In February 2026 OpenAI launched "Buy it in ChatGPT" — and in March it pulled back from native checkout, pivoting to agentic storefronts: the purchase completes in the merchant's store, not in the chat. The AI transaction layer is in motion, but the direction is settled: the ACP (OpenAI/Stripe), AP2 (Google) and UCP protocols are already standardizing how an agent finds a product, pays and places an order. What a store should do today to avoid burning budget on a moving target: the product feed as the zero-risk investment, API readiness, and a cool-headed decision matrix — join now or wait deliberately.

The second lesson matters just as much: the direction hasn't changed. The protocols are public and actively developed (ACP from OpenAI and Stripe, AP2 from Google, UCP with Walmart and Target on board), the product feed spec is stable, and agents shopping on the user's behalf are the logical extension of the trend I described in SEO and GEO for e-commerce — from product visibility in a carousel to placing the order is just one protocol step. This post covers what to do today: where to invest with zero risk (the feed), what to prepare in advance (the order API), what not to do at all — and how to make that call on a real-world stack, not in San Francisco.

The timeline: from Instant Checkout to agentic storefronts

To judge the risk properly you need the sequence of events — it's shorter and more instructive than the noise suggests:

The pattern in this timeline is one you know from every platform: the maximal variant first (native checkout), then a correction to the variant merchants will accept. Because it was the merchants who pushed back — a native checkout inside the chat meant giving up the customer relationship, the data and control of the cart. The "product in the chat, purchase at your store" model is a compromise the merchant wins more from: agent-driven visibility, transaction and data as before. Is it the end state? I doubt it — the protocols keep evolving (the stable April 2026 spec covers checkout, payment delegation, cart, feed, orders and MCP integration), and Stripe and PayPal are building for full agent-led transactions. But it's today's reality, and today's reality is what you play against.

The protocol stack: ACP, AP2, UCP — who's behind what

Three acronyms fly around interchangeably in agentic commerce writing, and they cover different layers:

• ACP (Agentic Commerce Protocol) — OpenAI + Stripe, open source (Apache 2.0, September 2025). It defines the full loop: how an agent reads the offer (feed), manages a cart, delegates payment with a narrowly scoped token and places the order with the merchant, who remains the merchant of record — the crucial part: you are the seller, not OpenAI.
• AP2 (Agent Payments Protocol) — Google, the agent-payments layer: mandates and authorization for what an agent may buy on the user's behalf. Adyen and other payment players are building their integrations on it.
• UCP — developed with a retailer coalition (Walmart, Target, Shopify, Etsy and twenty-plus others): the offer-discovery and purchase-intent layer, meant as a common language between agents and store catalogs.

The practical translation: the feed and product data are the common denominator of all three. The protocols compete over payments and orchestration, but every one of them first needs to know what you sell, at what price, and whether it's in stock. Which is why…

The product feed — the only zero-risk investment

If you remember one thing from this post: a solid product feed works no matter which checkout variant wins. It feeds today's shopping carousels (recall the number from the e-commerce post: ~83% of ChatGPT's carousel data comes from Google Shopping), it's the entry point to ACP, and it fuels agents' comparisons. The ACP feed spec converges with what you should have for Google anyway:

1. 1.Identity: stable product IDs, GTIN/EAN, brand — without them the agent can't join your offer to the rest of what it knows about the product.
2. 2.Content: titles and descriptions written for buyers' questions, not keyword stuffing — the agent reads them like a human, only faster.
3. 3.Commerce: price with an ISO 4217 currency, real-time availability, variants (size/color), delivery costs and times.
4. 4.Compliance flags: eligibility markers — what may be sold through an agent (regulated categories, age limits, shipping restrictions).
5. 5.Trust: the returns policy, reviews with ratings — an agent recommends what the data tells it to, and returns and ratings data weigh a lot.
6. 6.Freshness: a feed refreshed hourly, not weekly — a stale price in an agent's answer is a cancelled order and lost system trust.

Plus the on-site layer: structured data Product/Offer consistent with the feed — an agent visiting the product page to verify the offer must see the same numbers.

The real-world stack: Shopify, BaseLinker and the rest

Shopify does the most for you: stores on the platform are being enrolled into the ChatGPT integration in bulk (over a million merchants), with checkout and payments staying on Shopify's side. If you're on Shopify, your work is product data quality and the visibility decision — not integration.

BaseLinker and the Presta/Woo/IdoSell stack — the reality of many European stores — has no "enable ChatGPT" button today. That's a smaller problem than it sounds, because what's worth doing pays off everywhere: a first-class feed to Google Merchant Center (with BaseLinker you do it once for all channels — covered in e-commerce automation with BaseLinker), structured data on product pages, fast and stable stock APIs. When agent integrations reach your platform — and they will, the spec is open — you'll join in weeks, not quarters. The post-order operations layer (statuses, returns, communication) is already covered in order and fulfillment automation — an agent's order is no different from a human's there.

Join now or wait — the cool-headed matrix

The question isn't "is agentic commerce coming" but "should your store be in the first wave". Honest criteria:

Play now if: you sell parameter-comparable products (electronics, supplements, parts — where the agent genuinely chooses), you have a price edge or unique inventory, your margin absorbs potential intermediary fees, and your feed and stock data are in order. The first wave gets disproportionate visibility — as always.

Wait deliberately if: you sell on relationships and advice (an agent flattens that edge into a table), your margins are thin (the intermediary layer's fees are still forming), or your customer ends up on a marketplace anyway. "Deliberately" means: you do the feed and the data regardless, and you revisit the transactional-integration decision quarterly.

The risks, named: losing the customer relationship (the agent is loyal to the user, not the store), price pressure (comparison is zero-cost), transaction data flowing through an intermediary and — familiar from the zero-click era — the risk of building on a borrowed channel. So that post's rule applies here too: agent channels as an additional leg of sales, never the only one.

How to measure it from day one

Before dedicated reports exist, you measure with what you have — the toolkit is in AI traffic analytics: referrals from chatgpt.com/copilot/perplexity to product pages, that traffic's conversion versus organic (expect it higher — the buyer arrives post-decision), orders via in-app browsers (user agents in server logs), and product presence in carousels for a fixed set of shopping prompts ("buy me X under $50"). One number worth reporting to the board monthly: the share of order value from AI sources — a fraction of a percent today, but its trend, not its level, tells you when to shift priorities.

A step-by-step implementation plan

1. 1.Feed audit — field completeness (GTIN, variants, delivery), refresh cadence, price consistency across feed ↔ page ↔ schema.
2. 2.Structured data on product pages — Product/Offer/Review synchronized with the feed; an agent verifying the offer must see one truth.
3. 3.Merchant Center and Bing Webmaster — both product indexes feed agent answers; the second is nearly uncontested ground in most European markets.
4. 4.An agent test of your own store — ask ChatGPT and Perplexity to find and compare your products; write down what they get wrong.
5. 5.The matrix decision — first wave vs deliberate waiting; put a quarterly review date on it.
6. 6.API readiness (if you play now) — stock and orders accessible programmatically; on Shopify, review the ChatGPT integration settings.
7. 7.Measurement from today — an AI traffic segment on product pages and the share of order value from AI sources in the monthly report.
8. 8.Watch the protocols, not the headlines — changes in the ACP/AP2/UCP specs say more about direction than press releases do.

---

I prepare stores for agent-driven sales end to end: from the feed and structured-data audit, through BaseLinker/Merchant Center integrations, to the decision matrix and measurement. I do this as part of e-commerce SEO and AI and AI optimization (GEO), with the operations layer tied into AI automation. Get in touch — I'll start with an agent test of your store: what agents see, what they miss, and how much that costs you in sales today.

Worth reading next:

• SEO and GEO for e-commerce — product visibility in AI shopping
• BaseLinker + AI — e-commerce automation in practice
• Order and fulfillment automation with AI
• Zero-click and declining organic traffic — a brand strategy