Agentic commerce 2026 | shopping with AI assistants

As of 23 October 2026

AI assistants have gone from novelty to everyday tool in a short time. If you use them daily to research, compare and write, you know the situation: the assistant finds the right product, but at the point of purchase the conversation ends with a link. Why doesn't it help with the next step as well? Comparing prices and buying directly in the chat would be the obvious extension. The big providers announced such features a while ago.

Anyone following the topic runs into three central questions:

Which standards exist, and what are they suited for?
How can price comparisons be integrated into AI assistants, and where can you actually buy directly in the chat today?
Which of these options are already available in Germany, and which might follow?

Between announcement and real-world use, a lot of time often passes, and not every announced solution lasts. So we looked at which standards and solutions already exist, what they can be used for, and how far their practical rollout has really come.

What happened to Instant Checkout in ChatGPT

OpenAI set the direction for 2026, though not quite as planned. In September 2025, together with Stripe, the company launched Instant Checkout in ChatGPT, initially in the US and with Etsy as the first major partner. The technical foundation was the jointly developed Agentic Commerce Protocol (ACP), which is meant to connect AI assistants with retailers' checkout systems. In March 2026, however, OpenAI changed course: instead of offering purchase completion exclusively inside ChatGPT, the company put more weight on retailers' own online shops again.

What was dropped was not the channel but an approach: the single buy button that let customers pay directly in the conversation moved into the background. More interesting than the change of course is what came after. Walmart is still reachable through ChatGPT, but with its own assistant and a checkout flow it designs itself. Access to customers remains, while control over product presentation and purchase completion lies more with the retailer again.

Other providers take similar approaches. Claude, for example, integrates Instacart: the basket can be assembled in the conversation, while the purchase is completed through the provider's account.

ACP remains relevant despite the change of course. It continues to be maintained as an open specification and is picked up by other platforms. Microsoft, for example, uses it for its Copilot Checkout.

A trend emerges: AI assistants increasingly take over product search, comparison and preparation of the purchase, while retailers continue to play a central role in completing it. Which technical standards enable this cooperation and how they differ is shown in the overview below.

Which standards make AI shopping possible

For AI assistants to find and compare products and to prepare or complete purchases, different systems have to work together. Several technical standards have emerged for this, each taking on different tasks.

UCP: the flow of commerce

While ACP came out of the collaboration between OpenAI and Stripe, the Universal Commerce Protocol (UCP) developed as another approach to standardising AI-assisted commerce. UCP was introduced in January 2026 by Google and Shopify, with companies such as Etsy, Wayfair, Target and Walmart involved. It aims to connect the different steps of online retail, from product search through the basket to checkout and order management. In April 2026, Amazon, Meta, Microsoft, Salesforce and Stripe joined the Technical Council as further major companies. Stripe also joined the Governing Council.

UCP is not just a theoretical standard. Shopify activated the corresponding interfaces platform-wide at the end of May 2026, without merchants having to do anything themselves. This allows AI assistants to retrieve products and baskets via the shop domain without a separate partnership. Additional requirements apply to completing the purchase: registered applications can generate a pre-filled payment page, while direct completion inside an AI assistant is initially reserved for selected partners.

AP2: the power of attorney for the agent

While UCP describes the flow of commerce, the Agent Payments Protocol (AP2) deals with a different question: how can an AI agent prove that it is acting on behalf of a person and may trigger a purchase within certain limits?

For this, AP2 provides cryptographically signed mandates that define, for example, spending limits, recipients and validity period. In April 2026, Google handed the protocol over to the FIDO Alliance, which is among other things behind the passkey standards. AP2 does not execute payments itself but provides the technical basis for authorising them. Broad productive use is still to come.

MCP: connecting external services

Another important role is played by the Model Context Protocol (MCP). It allows AI assistants to connect external tools and services. This makes it possible, for example, to integrate product searches or price comparisons directly into a conversation.

The division of labour at a glance

The standards thus take on different tasks: ACP and UCP create technical foundations for AI-assisted commerce, AP2 governs the authorisation of purchases, and MCP enables the connection of external services. The actual payment processing remains with the payment providers.

How products and price comparisons reach AI assistants

The technical standards are the foundation. For companies, though, a practical question remains: how do their own products or a price comparison actually get into an AI assistant?

One option is external integrations, for example via the Model Context Protocol (MCP). They let AI assistants access services that search current offers from different retailers and present matching results directly in the conversation. How such integrations are made available differs by platform, however:

Anthropic (Claude): There is a public directory where developers can submit their integrations. Users can also add their own MCP connections directly, even if they are not officially listed.
OpenAI (ChatGPT): Developers can submit their applications for inclusion in the app directory. Once approved, they can be used inside ChatGPT.
Google (Gemini): Google relies more on selected partnerships. There is no public directory for external MCP integrations so far. For retailers, the Shopping Graph also plays an important role, through which Google supplies product information to its search and shopping offerings.
Microsoft (365 Copilot): Custom agents can be published. They are aimed mainly at companies and can help with research, offer comparison or procurement, for example. Use within an organisation usually requires approval from IT administration.

The differences show that the technical connection alone is not enough. It also matters whether and how each platform makes external services accessible to its users.

A pure price comparison needs no purchase or payment process of its own. Offers can be presented directly in the conversation while the actual purchase takes place with the retailer. Whether such services are available to users depends on the integration options and approval processes of the respective AI platform.

Where buying directly in the AI assistant already works

Buying directly through AI assistants is now a reality, although the offerings are so far concentrated in the US market. The big platforms take different approaches:

Google: In mid-October 2026, Google introduced Universal Cart, a shared basket for Google Search and the Gemini app. Users can collect products from different retailers and then either pay directly with Google Pay or complete the purchase in the respective retailer's online shop.
Microsoft: Copilot Checkout, launched in January 2026, allows purchase completion directly within the conversation. It is supported by Shopify, PayPal and Stripe, among others. Shopify merchants are in principle included automatically unless they object to participating.
Amazon: Amazon renamed its shopping assistant Rufus to "Alexa for Shopping" in May 2026. It helps with product search and comparing offers and, through functions such as "Buy for Me", can also place orders with external retailers.

The AI assistant supports product selection and purchase completion, while the retailer still sells the goods. Payment runs through established payment services such as Google Pay, PayPal or Stripe.

And what about Germany?

In Germany, AI-assisted product searches and price comparisons are already possible. Direct purchase completion inside an AI assistant, however, is not yet widely available. The corresponding features are so far concentrated in the US market, and no concrete launch date for Germany is known.

A major challenge is payment processing. In Europe, requirements such as PSD2 have to be taken into account, in particular secure authorisation of payments by the user. First pilot projects show, however, that progress is being made here too. Until buying directly in the chat becomes everyday practice, AI assistants can already help with product selection and purchase preparation.

Why payment providers play a central role

Alongside AI platforms and retailers, payment providers are gaining importance too. Even if an assistant can find products and prepare a purchase, the actual payment step needs secure and reliable infrastructure.

Companies such as Adyen, Stripe, PayPal and Klarna have therefore deliberately extended their offerings with features for agentic commerce. Adyen, for example, supports several of the emerging protocols, while Stripe was involved in developing the Agentic Commerce Protocol and also supports the further development of UCP. PayPal works with Microsoft and Perplexity, among others, and Klarna is integrated into Google's shopping offering.

It is becoming clear that payment providers are not committing to a single standard. Instead, they are building solutions that can work with different AI platforms and commerce protocols.

This is particularly relevant for the European market. Many of the shopping offerings so far are strongly oriented towards the US market and card-based payment methods. In Germany, however, customers expect additional options such as PayPal, SEPA direct debit or buying on invoice. These payment methods, like the regulatory requirements, have to be integrated into the new flows.

For broad adoption in Germany, it is therefore not enough for buying in the AI assistant to work. Payment processing also has to meet local requirements.

What companies can implement today

Even though direct purchase completion in Germany is still limited, retailers and price comparison sites can already create important prerequisites for AI-assisted commerce today. Three stages can be distinguished.

1. Make products discoverable

The first step is to make products and offers accessible to AI assistants. This can be done through structured product data, existing commerce platforms or external integrations. MCP in particular makes it possible to bring services such as product searches and price comparisons directly into the conversation. Neither a payment function of your own nor a fully integrated checkout is required for this.

2. Prepare the purchase

In the next step, the assistant can go beyond pure product search. If an online shop supports the relevant interfaces, baskets can be assembled and buyer or delivery details prepared for checkout, for example. The user is then forwarded to the online shop and completes the purchase there.

This approach already offers practical value: product selection and preparation happen in the AI assistant, while the actual order and payment are handled through the retailer's existing infrastructure.

3. Complete the purchase directly in the chat

The third stage is fully integrated purchase completion, where the user no longer has to leave the AI assistant. Such solutions already exist but are not yet widely available for the German market. Companies should therefore follow the development of platforms, payment standards and regulatory requirements without making their whole strategy depend on this feature.

For retailers and price comparison sites, this leads to a clear priority: first, products and offers should be discoverable in AI assistants. Then purchase preparation can be integrated. Direct payment completion can follow once the technical and regulatory conditions are in place.

Conclusion: AI-assisted commerce does not start at checkout

The development shows that agentic commerce is long past being a mere announcement. With ACP, UCP, AP2 and MCP, technical foundations are emerging for cooperation between AI assistants, retailers and payment providers. At the same time, platforms still differ markedly in which functions they offer and how open they are to external services.

While direct purchase completion is advancing mainly in the US, product search, price comparison and purchase preparation are already possible in Germany as well. That is the opportunity companies can use in the short term: they can make their offers available in AI assistants today, without waiting for the fully automated purchase process.

What matters, then, is less which features are announced and more which ones already work reliably.

Would you like to know how your products or your price comparison can be integrated into AI assistants? We help you assess the technical options and build the right integrations. Get in touch for an initial conversation.

Agentic commerce: where shopping with AI assistants really stands