Agentic Commerce: How AI Agents Buy Products via Shopify

Agentic commerce is changing how products get sold

AI agents are buying products. Not in two years. Now. According to Shopify's own figures, AI traffic to Shopify stores grew more than eight times year over year and orders from AI searches rose almost thirteen times, measured across the first quarter of 2026. The direction is clear: agentic commerce is growing into a serious purchasing channel, no longer an experiment.

This is not an abstract story. Dutch consumers spent a combined 35.7 billion euros online in 2025, with growth in product purchases and a decline in travel and events. A growing share of those transactions does not start on your homepage, not through Google Ads and not through an Instagram ad. They start in a chat window. A consumer types "find me a good leather laptop bag under 80 euros" and an AI agent searches online stores, compares options and sends the buyer on to the purchase.

This is called agentic commerce. Over the past year Shopify has built a complete ecosystem for it, and a lot has shifted since the first version of this article in June. Checkout inside ChatGPT itself has stopped, while Microsoft, Google and Meta now do make it possible for American buyers.

This article explains where that ecosystem stands on 28 September 2026, which protocols sit underneath it, how far the competition has come, and what it concretely means if you run an online store.

What has changed in agentic commerce since June?

Three things we noted as "coming soon" in June have since been decided, and not all in the expected direction. ChatGPT no longer handles checkout itself, Copilot does, and Google and Meta have joined as channels with direct checkout. The table sets our old expectation next to where things stand today.

The lesson is in the first row. OpenAI put it in its own words, as quoted by CNBC: "We've found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we're allowing merchants to use their own checkout experiences while we focus our efforts on product discovery." Buyers use AI to choose, and still like to pay at the store itself. That makes your own checkout more important, not less.

How do AI agents buy products?

An AI agent is not a chatbot. Where a chatbot answers questions, an agent carries out tasks. The difference is fundamental.

When a consumer asks ChatGPT to "find a red running shoe for wide feet, size 44, under 120 euros", the agent goes through several steps:

1. Discovery. The agent searches the product catalogues of online stores. With Shopify this happens through Shopify Catalog, which automatically supplies product data from eligible stores to the connected AI channels, with title, description, options, images, price and availability.

2. Evaluation. The agent compares products on specifications, price, availability, reviews and return policy. Not visually, but on the basis of structured data. JSON-LD schema, API responses, machine-readable product information.

3. Selection. Based on the user's criteria, the agent filters down to a shortlist. The agent can weigh in contextual factors: previous purchases, budget, location, delivery preferences.

4. Transaction. This is where the channels differ. In ChatGPT the buyer opens your own store's checkout in an in-app browser, or in a new tab on the web. In Microsoft Copilot, Google AI Mode and Gemini, and at Meta, the buyer can pay inside the channel itself with direct checkout, if your store is eligible.

Every step in this process depends on data. No images, no atmosphere, no brand story in the classic sense. Pure structured, machine-readable information.

That is the core of agentic commerce: your store now has two audiences. People who scroll and look. And machines that read data and decide.

Which AI platforms already sell products via Shopify?

Shopify bundles the channels under the name agentic storefronts, and switched them on for eligible stores on 24 March 2026. In the announcement Shopify wrote that millions of merchants could sell to ChatGPT users starting that week, managed from the Shopify admin, without separate apps or extra transaction fees. The channels are on by default for eligible stores.

ChatGPT. A discovery channel. The agent searches products, shows images, compares prices and sends the buyer to your own checkout. Because that checkout is yours, all your payment methods, discounts and branding work. Shopify's condition: your store sells to buyers in the US, although the store itself may be based elsewhere.

Microsoft Copilot. Built into Windows, Edge and Copilot.com. Microsoft launched Copilot Checkout in January 2026 in the US, with PayPal, Shopify and Stripe as partners. Shopify stores were enrolled automatically after an opt-out window. In April Microsoft extended Copilot Checkout to its mobile apps and added support for UCP. Payment is by credit or debit card.

Google AI Mode and the Gemini app. Google builds its checkout on UCP. According to Google Merchant Center, that UCP checkout is available to select merchants, for products eligible in the United States, Canada and Australia. Through Shopify the channel is only open to stores based in the US that sell to American buyers, and the rollout is still under way.

Meta. New since June. Through Shopify, Meta surfaces such as Muse have direct checkout for stores that sell to buyers in the US, Canada or Mexico.

The pattern is clear. Every major AI player is building a shopping layer, and Shopify has a seat at all four tables. At the same time the table shows where it stops for now: at the American buyer.

The protocols under agentic commerce: MCP, UCP, A2A and NLWeb

Agentic commerce does not run on loose integrations. A protocol layer is emerging that defines how AI agents communicate with online stores. Four protocols form the foundation.

Model Context Protocol (MCP)

MCP was developed by Anthropic and is now supported by OpenAI, Google and Microsoft. It is the protocol AI models use to call tools and data sources. In December 2025 Anthropic placed MCP with the Linux Foundation's Agentic AI Foundation. At that point the protocol counted more than 97 million SDK downloads per month and 10,000 active servers.

Shopify has built its own MCP server: the Dev MCP Server (@shopify/dev-mcp). It makes the Shopify developer platform "AI-native". Developers can use natural language to set up Shopify apps, run GraphQL queries and generate validated code for Liquid templates, UI extensions and Hydrogen storefronts.

For Shopify development this concretely means: less digging through documentation, faster theming, fewer errors. The server runs locally and requires no authentication.

Universal Commerce Protocol (UCP)

UCP is the standard built specifically for agentic commerce. Google and Shopify started it, and the circle has grown considerably since. The official UCP site lists Google, Shopify, Etsy, Wayfair, Target, Walmart, Amazon, Microsoft, Meta, Salesforce and Stripe as co-developers for shopping. Its endorsers include payment players such as Visa, Mastercard, American Express, PayPal, Klarna and Adyen, and European retailers such as Zalando and Carrefour.

UCP defines how an AI agent requests product data, starts a checkout and handles a payment. It is the common language between agents and online stores. Where Meta was still testing in June, direct checkout on Meta surfaces now runs through Shopify.

According to the UCP specification, a store publishes its profile at /.well-known/ucp. That JSON file tells agents which capabilities the store supports and how its checkout works. On Shopify you do not have to build this yourself: Shopify handles the connection to the channels through agentic storefronts. Brands without a Shopify store can still put their products into Shopify Catalog through the Agentic plan.

Agent2Agent Protocol (A2A)

A2A was developed by Google and now sits with the Linux Foundation. The protocol governs communication between agents. Think of a scenario where your personal shopping agent negotiates a return with an online store's customer service agent.

Version 1.0, the first stable specification, was released on 9 April 2026. According to the Linux Foundation, more than 150 organisations now support the standard, including AWS, Google, Microsoft, Salesforce and SAP.

Natural Language Web (NLWeb)

NLWeb comes from Microsoft and was conceived by R.V. Guha, the creator of RSS and Schema.org. It makes websites queryable in natural language. Every NLWeb instance automatically functions as an MCP server.

The practical implication: websites with existing Schema.org markup can make their content available to AI agents without extra development. Shopify, Tripadvisor and Eventbrite were early adopters. Of the four protocols this is the quietest: since June we have not seen new commerce features appear on it.

Together these four protocols form the infrastructure of the agentic web. This is not about a single integration with ChatGPT. It is a whole layer being built on top of the existing web.

Why Shopify is ahead of WooCommerce, Magento and BigCommerce

Shopify's lead on agentic commerce is still large, but smaller than in June. The competitors are mainly building MCP servers. What they lack is distribution: nobody else puts your products into four AI channels at once without integration work.

WooCommerce got its MCP integration as a beta in its October 2025 release. It is still behind a feature flag: you have to switch it on and maintain it yourself. The June 2026 WooCommerce release replaced the old integration with fixed building blocks for products and orders that work the same across REST, MCP and CLI. WordPress 7.0 shipped on 20 May 2026 with an AI Client, the Abilities API and an MCP adapter, but aimed at site management, not at selling through AI channels. In April Microsoft also made its Brand Agents, the AI assistants for your own site, available to WooCommerce stores.

Magento (Adobe Commerce) does not have agentic commerce in production yet. At Summit 2026 Adobe announced a Commerce MCP for agentic experiences from discovery to checkout, but Adobe's own documentation marks the feature as "Coming Soon". Until then, agent readiness on Magento means custom development.

BigCommerce is moving. Since May every live BigCommerce store can switch on the Storefront MCP through Early Access. An agent can use it to search the catalogue, fetch variants, fill a cart and create a checkout link. Direct connections to ChatGPT, Copilot and Gemini like Shopify's are not something BigCommerce offers out of the box.

If you are choosing a platform for a new store right now, or considering a migration, this is the table that matters. Shopify is not by definition the best platform for every situation. For the broader trade-off we wrote Shopify vs WooCommerce: which platform fits your brand. But for agentic commerce readiness, Shopify is currently the only platform with a complete stack.

At Oase Creative we build Shopify stores and now include agentic readiness as standard in every new build. Not as an upsell. As part of a store that is ready for how e-commerce works in 2026.

What does agentic commerce mean for the Dutch market?

Short answer: discovery through AI works for Dutch buyers, paying inside the chat does not yet. The Dutch e-commerce market has specific characteristics that add nuance to the agentic commerce story.

Paying inside the chat is still American

In June we wrote that ChatGPT Instant Checkout was live for Etsy in the US and that Shopify would follow soon. That did not happen: OpenAI ended the feature in March. The direct checkout that does exist, in Copilot, Google and Meta, only applies to buyers in the US, and at Meta also Canada and Mexico. Google's UCP checkout itself is also available for Canada and Australia outside Shopify. Shopify, Microsoft and Google give no date for Europe.

What makes a European rollout harder is no secret:

• Payment preferences. Dutch consumers pay with iDEAL, not credit cards. Belgian consumers use Bancontact. Direct checkout in Copilot and Google currently works with payment cards.
• Strong customer authentication. European payment rules require extra verification steps for online payments, and those have to work just as well inside a chat window.
• Phased rollout. The AI channels roll out checkout country by country. The US first, Europe later.

Product discovery already works

Checkout may not be live in the Netherlands yet, but product discovery through AI does work. When a Dutch consumer asks ChatGPT or Copilot for product recommendations, your store can appear in the results through the open web: crawlers read your pages and your structured data. Taking part in ChatGPT and Copilot through Shopify Catalog is also possible from the Netherlands, as long as you sell to American buyers.

This is where the opportunity is right now. Not in the transaction layer (that is coming), but in the discovery layer (that is running). How to see whether AI channels are already sending you buyers is covered in measuring AI referrals in GA4.

Your Shopify store is already talking to agents

Every Shopify store now automatically serves an /agents.md, plus /llms.txt and /llms-full.txt: files in which an AI agent reads how your store works, with your store name, sitemap, policies and the addresses where product information can be found. You do not need an app for this, and you can adjust the content with a Liquid template.

This is a concrete opportunity for Dutch stores. When we checked 18 Dutch Shopify stores, all 17 with a real file were still on the factory default. Whoever writes that file themselves gives agents a better answer than the competitor who leaves it alone.

Dutch online stores that invest in agentic readiness now are building the same kind of lead as the early SEO investors ten years ago. With one difference: the adoption curve is steeper. SEO had years to mature. Agentic commerce compresses that into months.

The Shopify AI Toolkit: what does it do and why does it matter?

The AI Toolkit is Shopify's connection between AI coding assistants and the Shopify platform. Shopify published it in April 2026. According to Shopify's developer documentation it works with Claude Code, Codex, Cursor, Visual Studio Code and Antigravity CLI, among others.

What it does:

• Developer docs and API schemas searchable from your code editor, so the assistant works from current reference material
• Code validation of GraphQL queries, Liquid templates and Shopify Extensions against Shopify's own schemas
• Store management through the Shopify CLI, inside the authenticated store context, straight from your editor

The toolkit is open source under the MIT licence and available on GitHub.

For developers this is a productivity leap. No tabs full of documentation. No hunting for the right API call. The editor knows what Shopify can do and validates your code straight away.

But the toolkit is more than a developer tool. It is a signal. Shopify is not only investing in agentic commerce for consumers. It is investing in agentic development. The whole chain, from building to selling, is becoming AI-native.

For SEO the same principle applies on the sales side. Shopify Catalog makes your product data readable for AI channels, and crawlers read your pages as well. Schema.org markup on your store is no longer just a way to be found in AI search engines. It is also your agentic commerce foundation. Structured data is your AI storefront.

How do you prepare your store for agentic commerce?

Start with your product data, because that is what agents read in every channel. The protocols are in place. The channels are live. What should you concretely do now?

Product data as a strategic asset

This is the most important shift. Product data is no longer a technical detail you leave to your ERP system. It is a strategic asset.

AI agents do not read marketing language. "Premium quality" is worthless to an agent. "Full-grain leather, 2 mm thick, water-repellent, 350 grams" is usable. The first 60 to 100 words of your product description should be factual. Product type, materials, dimensions, specifications. Only then comes the brand story. Shopify also asks you to put any relevant legal disclosures in the first 6,000 characters of your product description.

Every product variant (size, colour, version) needs its own structured data with current availability. An agent that recommends size 44 while it is sold out quickly learns to avoid your store. If your product titles, descriptions or categories live in custom fields such as metafields, you can use Shopify Catalog Mapping to tell Catalog where to fetch them.

Structured data as the foundation

JSON-LD Product schema on every product page. Not optional. Product, AggregateRating, MerchantReturnPolicy, all fields filled in: name, description, price, availability, brand, SKU, GTIN/EAN.

On top of that: check what your store says about itself. On Shopify that is your /agents.md, and your policy pages (terms of service, privacy, returns) must be filled in to take part in the AI channels. If you are not on Shopify, a UCP profile at /.well-known/ucp is how you tell agents which capabilities your store supports and how its checkout works.

A fast and predictable checkout

In ChatGPT the buyer pays in your checkout, opened in an in-app browser. Every extra step, every slow script and every forced account costs you conversion right there. A fast Storefront API and a checkout without obstacles are as much agentic readiness as your product data. The points that go wrong most often are covered in checkout optimisation: Baymard's 32 points.

The 90-day roadmap

• Day 1 to 30: Product data audit. Implement JSON-LD schema on all product pages. Set up real-time data sync between your inventory and your storefront. Check in your Shopify admin under Sales channels > Agentic which channels are switched on.
• Day 31 to 60: Rewrite your /agents.md. Document your Storefront API. Optimise cart and checkout for ChatGPT's in-app browser.
• Day 61 to 90: Measure AI traffic and AI orders per channel. Test the full flow from discovery to checkout, on mobile and desktop.

This is not a theoretical roadmap. It is based on what Shopify merchants are doing right now to become findable for AI agents.

The future of agentic commerce: what is still coming?

Agentic commerce is at the beginning, and 2026 showed that the direction is not a straight line. This is what is coming, and what we phrase more carefully after this year.

Autonomous purchases. Right now the consumer still confirms every purchase. The market expects consumers to eventually hand repeat purchases over to agents entirely. "Order my coffee every month, find the best price." The end of Instant Checkout does show that buyers talk about this faster than they actually do it.

Agent-to-agent negotiation. Through the A2A protocol, stable since April, consumer agents can start negotiating with store agents. On price, delivery time, bundles. The consumer only sees the end result.

Brand identity for machines. Branding becomes twofold. Your visual identity for people. And your data identity for machines. Consistent, reliable, fast data becomes the equivalent of a strong brand. Agents learn which stores supply reliable information and give them preference.

Consolidation towards Shopify. The Agentic plan has been publicly available since March. Brands that do not run a Shopify store but do want to be findable for AI agents can use it to add their products to Shopify Catalog and keep their existing store. This lowers the barrier to take part and grows Shopify's network.

Checkout coming to Europe. Direct checkout works in the US now, and at Google also in Canada and Australia. Europe is the logical next step, but no date has been announced. Do not plan on it, but make sure your data is ready when it comes.

The brands that start on agentic readiness now are building transaction history and trust signals with AI agents. That is a compounding effect latecomers cannot catch up on by spending more money later. Just like SEO, only faster.