Agentic Commerce: How to Prepare Your Online Store for AI Agents That Shop

Agentic commerce is when the product is chosen not by a human but by an AI agent: it reads your catalog, compares prices and stock, and the customer completes the purchase on your site. Being ready means a machine-readable feed, real-time inventory, and support for the ACP or UCP protocol.

A year ago this sounded like futurology. Today Adobe Analytics data is blunt: in March 2026 traffic from AI assistants converted 42% better than all other traffic. A year earlier — March 2025 — that same AI traffic converted 38% worse. That is not incremental growth, that is a full reversal. And during the 2025 holiday season AI-referred traffic to US retail sites grew 693% year over year.

If you run an online store, the question is no longer whether AI traffic will arrive. It is whether the agent will see your product when it does.

What agentic commerce means in plain terms

The classic path: a person googles, lands on a site, compares, adds to cart, pays. The agentic path: a person describes the job to an assistant ("I need a laptop under $1,200 with strong battery life, delivered this week"), the AI crawls catalogs itself, compares options by price, availability and shipping, and returns a short list — or places the order.

The key difference: a middleman now sits between your product and your buyer, and that middleman cannot guess. It does not scroll your beautiful landing page or notice your discount banner. It reads structured data: price, availability, GTIN, delivery time, return policy. No data, no product in the answer.

"ChatGPT is moving into shopping territory with confidence. For us as marketers that is a signal: optimize your SEO for AI. If your brand makes it into that answer, you get a fully warmed-up customer who has already compared you with competitors inside the chat and is ready to buy right now."

— Nikolay Shmichkov, "How to get into ChatGPT results and win warmed-up customers"

▶ How to get into ChatGPT results and win warmed-up customers

Why buying inside the chat stalled

In 2025 everyone expected checkout to move into the chat window. OpenAI even shipped Instant Checkout — buying without leaving ChatGPT. In March 2026 it quietly pulled the feature. The official line: it "did not offer the level of flexibility we aspire to provide." The unofficial reason is more interesting: according to CNBC, Walmart measured that checkout inside ChatGPT converted roughly three times worse than a click-through to walmart.com.

The conclusion the industry landed on has a name — "discover in AI, buy on site": AI owns discovery, the store owns the payment. The assistant does what it does better than a human (grind through hundreds of listings), and the money moves where the business keeps control, analytics, upsells and loyalty.

For a store owner that is good news. You do not have to hand checkout to someone else's platform. You only have to be visible during the selection step.

ACP and UCP: the two protocols to know

For an agent to talk to a store, both sides need a shared language. Two standards matter now.

ACP (Agentic Commerce Protocol) — an open standard from OpenAI and Stripe. It connects your catalog straight to the assistant: the agent sees live prices, stock and variants, and hands the order to the merchant. The spec is public, with implementations in Shopify, Etsy and other platforms.

UCP (Universal Commerce Protocol) — Google's answer, announced in January 2026 at NRF together with Shopify, Etsy, Wayfair, Target and Walmart. By Google Marketing Live in May 2026 it gained Universal Cart — a cart that spans multiple retailers — plus BNPL via Affirm and Klarna. You wire UCP up through Merchant Center: Google's docs.

Then there is the payment layer: in June 2026 Visa opened its network to agents paying any Visa-accepting merchant; Mastercard has Agent Pay. Arguing over which protocol wins is pointless. The practical move: ask your platform (Shopify, WooCommerce, Magento) whether ACP and UCP are supported, and switch on whatever is available. Both need exactly the same underlying data.

What a store owner should do: 5 steps

1. Clean up the product feed. This is the foundation. Minimum per item: GTIN/MPN, brand, exact price with currency, availability (in stock / out of stock), shipping time and cost, return policy. An agent that cannot verify stock simply will not recommend the product — being wrong is expensive for it.

2. Make the product page machine-readable. Product schema with offers, price, priceCurrency, availability, aggregateRating, review. Adobe notes plainly in 2026 that retail sites are lagging behind AI traffic precisely because their pages are hard for machines to parse. More in our guide on getting cited in AI Overviews and the 2026 e-commerce SEO checklist.

3. Sync inventory in real time. Once a day is not enough. An agent that sends a customer to an out-of-stock item learns not to recommend you again.

4. Give the agent facts it cannot infer from the card. Compatibility, dimensions, materials, warranty, who the product suits and who it does not. In agentic search the winner is not the prettiest photo but the richest set of verifiable attributes.

5. Collect reviews and UGC. Agents look for consensus: they cross-check your claims against what people say. A store with no reviews is, to an AI, a store with no evidence.

How to tell whether agents have found you

GA4 has had a dedicated AI Assistant channel since May 2026, but it does not catch everything by default. Check your traffic acquisition report for referrers chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com. Separately, look in server logs for GPTBot, OAI-SearchBot, PerplexityBot, Google-Extended: if they never show up, your catalog is invisible to agents. And make sure you have not blocked them in robots.txt "just in case" — that single line quietly cuts off your best-converting channel.

FAQ

Is agentic commerce only for large retailers?

No — the opposite. The agent does not know you are small. It compares price, availability and data. A small store with a clean feed will outrank a big one with a broken feed.

Do I need checkout inside the chat?

No. The 2026 model is "discover in AI, buy on site": AI helps choose, the customer pays on your site. Instant Checkout was pulled precisely because in-chat conversion was worse.

Which matters more, ACP or UCP?

Both rest on the same data: feed, prices, stock, schema. Start with the data — hooking up a specific protocol is then a platform-level task.

Does AI traffic really convert better?

Per Adobe, in March 2026 it converted 42% better than non-AI traffic and produced 37% more revenue per visit. The reason is simple: the customer arrives pre-sold, having done the comparison in the chat.

Bottom line

Agentic commerce does not ask you to reinvent your store. It asks for discipline in data: an accurate feed, honest stock levels, complete Product schema, reviews, and open doors for AI crawlers. The protocols handle the rest. Whoever fixes their data in 2026 gets the warmed-up customer; whoever does not simply disappears from the short list the AI shows the buyer.

Want an audit of your feed and product pages for AI search? Let's talk about your project — we will check whether agents can see you, and what is in the way.