- In agentic commerce, customers delegate product selection to AI. The products that get recommended are those with complete, consistent and verifiable data.
- OpenAI scaled back its in-chat checkout in March 2026: merchants use their own checkout, and ChatGPT focuses on product search via the Agentic Commerce Protocol.
- Google enables purchases directly in AI Mode and Gemini via the Universal Commerce Protocol, and Microsoft does the same in Copilot via Copilot Checkout. Both are currently limited to markets outside Europe.
- In all three systems, the merchant remains the seller. That makes the product feed your store's most important shop window.
If you work in e-commerce, you've already noticed: the familiar levers are less reliable than they used to be. Costs per click are rising, SEO delivers less predictable results, users click less, and buying decisions happen earlier. At the same time, new terms keep appearing: AI Overviews, conversational commerce, agentic checkout, shopping in ChatGPT.
Dismissing this as a passing trend underestimates what's happening. Software no longer just prepares buying decisions; it helps make them. That's exactly what agentic commerce describes. In this guide, we explain how the major systems work as of October 2026 and the steps that make your store the one that gets recommended.
In agentic commerce, your customers delegate the choice to AI
Agentic commerce doesn't mean users stop making decisions. They delegate. Instead of “I'll google, compare ten stores and decide”, they say: “Find me the best product for what I need.”
The model interprets the intent, weighs the alternatives and compares prices, availability, reviews and technical specs. In the end, it recommends one specific option and often leads straight to the purchase. If you're not recommended, you're not on page 2 or in position 7. You're simply not in the answer.
For AI assistants, your data matters more than your marketing message
Many teams respond with more content, more keywords and more campaigns. That misses the point. Language models don't click, don't read the marketing promises in your hero section and don't respond to emotional copy.
They evaluate data quality, consistency, verifiability and third-party validation. Agentic commerce is therefore first and foremost a product data and strategy task, and only then a marketing one.
Three platforms set the rules, each on its own terms
Three systems shape the market. Each changed significantly during 2026, and each follows a different business model.
| Platform | Checkout in the assistant | Merchant's role | Availability (October 2026) |
|---|---|---|---|
| ChatGPT (Agentic Commerce Protocol) | Product discovery first; checkout usually in the merchant's own store | Merchant supplies feed and promotions and uses its own checkout | Product search for ChatGPT Free, Go, Plus and Pro users |
| Microsoft Copilot Checkout | Checkout directly in Copilot | Merchant remains the seller and keeps the customer relationship and data | English-language merchants selling to buyers in the US |
| Google (Universal Commerce Protocol) | Checkout directly in AI Mode and Gemini via Google Pay | Merchant remains the seller | Selected merchants in the US, Canada and Australia |
ChatGPT focuses on product discovery
OpenAI first launched a native checkout in ChatGPT with the Agentic Commerce Protocol (ACP). In March 2026, it changed course: the first version of the checkout had not offered the flexibility it wanted, so merchants can now use their own checkout while OpenAI concentrates on product search.
Since then, merchants have shared product feeds and promotions via ACP, including through providers such as Salesforce and Stripe. According to OpenAI, Shopify merchants appear in ChatGPT through the Shopify Catalog with no extra work. If you want a deeper integration, you can build your own app in ChatGPT.
Microsoft Copilot sells inside the assistant
With Copilot Checkout, buyers complete their purchase without leaving Copilot. The merchant remains the seller and, according to Microsoft, keeps the customer relationship and the data. Microsoft currently charges no commission. You need a Microsoft Merchant Center account and a product feed. For now, only English-language merchants selling to US buyers can take part.
Google links Merchant Center and checkout
The Universal Commerce Protocol (UCP) is an open standard covering the entire shopping journey, from product search to post-purchase service. Built on it, selected merchants can add a buy button to AI Mode in Google Search and to Gemini. Payment runs through Google Pay, and the merchant remains the seller. The button only appears for products that carry the Merchant Center attribute native_commerce (checkout_eligibility). It is currently available for products in the US, Canada and Australia.
Who owns the customer data decides your margin
The key economic question is whether you keep the customer relationship after an AI-driven purchase. Without an email address and consent, there is no retargeting, no newsletter and no repeat business from lifecycle marketing, and a large share of customer value disappears.
All three systems currently keep the merchant as the seller of record. Even so, check for each platform which data you receive and how you are allowed to use it. Agentic commerce spans several business models, each with a different impact on profit.
We build the foundations that make your store trustworthy to AI systems
Before you plan any integration, ask one basic question: would a machine trust your store? Agentic systems vet merchants rigorously, and rightly so.
You sell directly and own the responsibility
Agentic commerce only works with clear accountability. Google and Microsoft explicitly require the merchant to remain the seller. You should be the legal seller, process payments yourself and own returns and complaints.
If your business relies on pure dropshipping, unclear fulfillment or layered marketplace models, you will not be favored. Agentic systems want an accountable seller, not a supply chain.
In-assistant checkout is launching outside Europe first
As of October 2026, direct checkout in the assistant is limited to North America and Australia: US buyers at Microsoft; the US, Canada and Australia at Google. Product search in ChatGPT, AI Mode and Gemini, however, already matters for your customers in Germany.
If you sell internationally, plan your participation now rather than waiting for the features to reach Europe.
Your payment flow has to be frictionless
An agent-led purchase does not tolerate friction. Redirects, extra confirmations and slow authorizations cost you sales. The systems rely on established payment providers such as Stripe, PayPal and Google Pay. If you already use them, you have the shortest path.
Your product feed is your new storefront
In agentic commerce, your website is no longer where the decision starts. Your feed is. If it is incomplete, inconsistent or outdated, you will not be recommended, however good your store looks. See how we build feeds on our product data feed page.
Product titles say what the product is
Creative product names don't help here. A good title states what the product is, who it is for and what it is used for, for example: “Noise-cancelling over-ear headphones – Bluetooth, 40 h battery, travel and office”. Language models favor clarity because it lowers the risk of a wrong recommendation.
A price mismatch costs trust
Language models compare your feed, your website and third-party sources. A price mismatch erodes trust. Keep prices current, always state the currency and avoid hidden conditions.
Report availability honestly
“In stock” means available now. “Coming soon” or “expected to be available” do not belong in this field. The systems penalize inaccurate availability data.
Images back up what the text promises
Images now also serve as evidence for the data in your feed. Use high-resolution images that show the product from several angles, without overlays or text. Language models use them to verify color, shape, material and variants.
The description is your strongest argument
The product description is specification, decision aid and fact source in one. It covers materials, dimensions, use cases, limitations and care instructions. Exaggeration hurts you.
Unique identifiers for every product
For products with a barcode, include the GTIN in the feed. Without a unique identifier, a system struggles to match your offer against other sources.
Structured data keeps your feed and website consistent
Product, Offer and Organization form the basis; AggregateRating and BreadcrumbList are useful additions. Google's AI answers don't require them, but they help every system match your feed with your website. Inconsistency breeds mistrust. Read more in our article on Generative Engine Optimization.
How we approach each platform on its own terms
ChatGPT: recommendations through product search
ChatGPT is the agent with the strongest recommendation effect. It recommends the product that best matches the query, then sends the buyer to your store.
- Shopify: Product data flows in through the Shopify Catalog. Check that the sync is complete and your products display correctly.
- Other store platforms: Product feeds via ACP, either directly or through providers such as Stripe and Salesforce.
- Paid reach: Product feed campaigns in ChatGPT Ads put your range into the answers as ads as well. Learn more in our guide to ChatGPT Ads.
Commercially, ChatGPT is most attractive for products with high order values and clear differentiation.
Microsoft Copilot: in-assistant checkout with the customer relationship intact
Copilot combines search, agent and checkout. According to Microsoft, you keep the customer relationship and the data, so you can reach those customers again. For merchants selling to US buyers, it currently strikes a good balance between reach and control.
Google: performance driven by data and price
Google's system is data-driven and highly price-focused. In return, you keep full control and can build on your existing Merchant Center setup. It suits interchangeable products where price and delivery time decide. See how we manage Google Shopping for profit on our Google Shopping page.
Language models find products through structured content and third-party validation
Language models read differently from people. They favor structured content, comparability and third-party validation.
Content that supports a buying decision
Detailed product pages, guides to specific use cases, comparison articles, pros-and-cons lists and spec tables work well. Generic blog posts, SEO filler and thought leadership without facts achieve little.
Independent reviews are essential
Language models trust trade media, specialist blogs, YouTube test videos and independent tests. Your job is to get your products in front of these sources and enable genuine reviews without influencing them.
For on-site reviews, quality counts
What matters are verified purchases, details on real-world use and honest criticism. Language models read negative reviews too and reward transparency.
Verifiable claims make your products worth recommending
Language models check claims. If you write “best product”, “sustainable” or “market leader”, you need to be able to prove it.
What works: clear use cases, a precisely defined target audience and candid trade-offs, for example: “Ideal for commuters with small ears – less bass, but lightweight.” What doesn't: generic superlatives, unsupported comparisons and greenwashing. The systems filter out marketing claims without substance.
We choose the strategy based on your product type
| Product type | Focus | Rationale |
|---|---|---|
| Interchangeable products | Google, price and availability | Efficiency decides here; a ChatGPT recommendation rarely tips the balance. |
| Differentiated products | ChatGPT and Copilot, content and reviews | Here, the recommendation matters more than price, and your brand becomes an advantage. |
| New products without data or reviews | Google campaigns, partnerships with testers | Fast external validation lays the groundwork for future recommendations. |
Your margin decides which platform comes first
Which platform pays off depends on your margin per order and the value of repeat customers.
- Prioritize ChatGPT and Copilot if your order values are high, your products clearly stand out and your profit relies little on repeat purchases via email.
- Prioritize Google if price is decisive, fast delivery matters and customer lifetime value from repeat purchases is important.
- Use all platforms if you sell on Shopify, carry several product categories and have the resources to split the work cleanly.
We make this call with you based on your margins and customer value. Learn more on our Agentic Commerce page.
Five rules for succeeding in agentic commerce
- Product truth beats marketing.
- Feed quality determines visibility.
- Speed beats perfection.
- Your data strategy decides your margin.
- Starting early builds trust.
Conclusion: agentic commerce is changing how buying decisions are made
Agentic commerce is neither a single feature nor a passing trend. It is a new way of making buying decisions, and the platforms keep adapting their models, as OpenAI's change of course in March 2026 shows.
So the question is not whether your store takes part, but how deliberately. Up-to-date, complete product data makes it easier to prepare for additional sales channels. Check which interfaces and requirements apply for each platform.
