Agentic Commerce

What is agentic commerce – and what does it take to be visible?

Agentic commerce describes a shift in how products are found and bought online, where AI searches and shops on the consumer’s behalf. It is happening in two distinct stages, and right now they are at very different levels of maturity.

Step one: AI search. It is already happening.

Consumers are increasingly turning to ChatGPT, Gemini and Perplexity when searching for products. “Which running shoe suits me if I have wide feet and train three times a week?” The answer comes back as a recommendation. The products being recommended are the ones whose data the AI model can read, understand and assess as trustworthy.

This creates new demands on how your product data is structured, how your product descriptions are written, and whether your site is seen as an authority in its field. Get it right and your products can be recommended by AI in ChatGPT, Gemini, Perplexity and similar search services, with the optimisation delivering results across all of them at the same time.

Step two: AI transactions. It is coming, probably fast.

The next step is for the AI agent not only to recommend, but also to complete the purchase directly in the chat interface. That requires ecosystem-specific integrations and the right payment infrastructure, and this is where the picture becomes more complex and fragmented. Google, Amazon and OpenAI are each building their own transaction infrastructure with different protocols and requirements.

The functionality is being tested in the US now, and EU expansion has been communicated. Once it opens up, things are expected to move quickly. Your products could then be bought directly in an AI interface, without the customer ever leaving the conversation.

McKinsey estimates that AI agents could influence more than $3 trillion in global purchases by 2030. Gartner predicts that 40 per cent of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5 per cent in 2025.

How does AI choose which products to recommend?

Several factors determine whether an AI agent chooses to recommend your product. Among the most important are how well the product data answers the actual intent behind the question, how accurate and complete the structured product data is, and whether the site is seen as a credible authority in its category. A product with incomplete or inconsistent data is quietly filtered out, while well-structured products with rich content are prioritised in recommendations.

Why is structured data critical for AI visibility?

Schema.org/JSON-LD markup on product pages is the foundation for how an AI agent understands what a product is, who it is for, whether it is in stock and at what price. Without correct and complete structured data, the product is practically invisible to agents.

The most important data points include:

• GTIN (EAN/UPC): The single most important identifier. Without a valid GTIN, AI cannot reliably match the product to external reviews and comparison sources.

• Real-time stock status: InStock, OutOfStock and PreOrder must be updated in real time. LLMs deprioritise retailers with unreliable stock status.

• Ratings and reviews: A product with an average rating of 4.8 based on 200 reviews is almost always prioritised over one with no rating.

• Product variants: Size, colour and other variants must be structured with clear parent-child relationships. If an agent asks for “blue, size 43” and gets a generic answer for the whole product, it loses the context and chooses another retailer.

How does the Google Shopping feed affect visibility in AI search?

A well-optimised Google Shopping feed affects product visibility across several AI search services at the same time. For Google’s own AI surfaces, Gemini and AI Mode, the Merchant Center feed is the primary data source. But the impact goes further: a study by Search Engine Land in March 2026 showed that 83 per cent of the products ChatGPT recommends in its shopping results are sourced directly from Google Shopping data. In other words, a well-optimised product feed delivers value far beyond traditional Shopping ads.

Accurate stock status and pricing are basic requirements. If your feed shows a product as in stock when it is sold out, the AI agent risks recommending something that cannot be delivered, which damages the retailer’s trustworthiness over time. AI agents learn from these inaccuracies and deprioritise retailers with unreliable data. The same applies to physical stores: if a consumer finds a product through AI search and wants to know whether it is in stock at a nearby store, exactly the same demands apply to data quality and real-time accuracy.

Site authority and expertise influence which products are recommended

AI models do not just evaluate individual product pages. They assess the topical depth of the entire site. A retailer that consistently publishes guides, comparisons and expert content within a specific area builds up topic authority that the AI model learns to associate with that category. That means a site with deep knowledge of running shoes, with guides on running technique, surfaces and shoe types, gives its product recommendations a structural advantage over a generalist retailer with the same range. This is known as E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness), and it is a signal AI search is applying with increasing precision.

Ratings and reviews are a ranking factor

Structured customer reviews, correctly formatted in Schema.org markup with ratingValue and reviewCount, are a strong signal for AI visibility. A product with a high average rating based on many reviews is almost without exception prioritised over a comparable product with no review data. The AI model also extracts customers’ own words from reviews to understand a product’s strengths and weaknesses, and uses them in its responses.

Product content determines whether you appear in the conversation

ChatGPT and similar systems are heavily text-driven. They match against natural language in conversational queries. Product descriptions that explain who the product is for, in what context and what makes it unique are what determine whether you appear when a customer asks “find a gift for a 50-year-old who enjoys the outdoors”.

Short product titles and technical specifications without context are no longer enough.

Serkan Selcuk, Product Owner at Viskan, explains: "This is about machine readability as infrastructure. AI agents quietly exclude products with incomplete, inconsistent or ambiguous data. It is not SEO optimisation in the traditional sense. It is a new requirement for data quality."

That is the foundation for being visible in AI search. The next step is understanding what it takes for your products to also be bought directly in an AI interface.

What does it take for AI agents to buy products directly in an AI interface?

For AI agents to complete purchases, your platform needs to support the ecosystems where those transactions happen. Today, Google, OpenAI and Amazon are building parallel transaction infrastructures with their own protocols and their own requirements. They are not compatible with each other, which means an integration with one ecosystem does not automatically cover the others. 

Google: the strongest position in agentic commerce

Google is currently the most complete player in agentic commerce and the ecosystem where development has come furthest. It is in Google’s AI-driven search surfaces, namely AI Mode in Google Search, AI Overviews and the Gemini app, that products can already be recommended and bought via AI agents. The direct purchase function is currently available in the US.

The foundation for this is Universal Commerce Protocol (UCP), which Google launched in January 2026 together with Shopify, Target, Walmart and Wayfair among others. More than 20 global partners have signed on, including Adyen, Visa, Mastercard and Stripe.

UCP makes it possible for an AI agent to present the product, manage the basket and complete payment via tokenised solutions such as Google Pay, directly inside AI search or the Gemini app.

In March 2026, Google updated UCP with new capabilities: agents can now add multiple products to a basket at the same time, fetch real-time data directly from the retailer’s catalogue, and connect identity to loyalty programmes to automatically apply the correct customer price.

Google’s ecosystem is the most mature for agentic commerce today. But there is one important limitation: the direct purchase function is being tested in the US. An EU launch has been communicated, but no date has been set.

OpenAI/ChatGPT: a strategic shift

On 4 March 2026, OpenAI confirmed that it is discontinuing Instant Checkout, the feature that enabled direct purchases inside the ChatGPT interface, and is instead focusing on merchants building dedicated apps within the platform. The decision is based on concrete data: ChatGPT users research products at scale but complete their purchases elsewhere. Early data showed that integrated checkout performed three times worse than sending the customer on to the retailer’s own site.

Today, ChatGPT works as a powerful research and discovery layer, while the transaction happens on the retailer’s platform. That is an important shift to keep an eye on: ChatGPT is a massive traffic source with 900 million weekly users, but its transaction role is being reshaped.

For transactions, the direction points towards Amazon. With the strategic partnership between OpenAI and Amazon, where Amazon is investing $50 billion and AWS becomes the primary cloud provider, and Amazon’s parallel launch of Shop Direct in March 2026, the likely development is that purchase flows in ChatGPT will eventually be channelled through Amazon’s commerce infrastructure rather than through direct integration with OpenAI. This is not yet fully implemented, but it is the direction the market is moving in.

Amazon: one to keep on your radar

Amazon has a limited position in the Swedish market today, but it is playing a growing role in the global agentic commerce landscape. Through the Shop Direct programme, launched in March 2026, external merchants can sync product data with Amazon’s AI infrastructure and reach customers via Rufus and Buy for Me.

What sets Amazon apart from the other players is that it runs a completely separate ecosystem. It is not part of UCP or ACP. Its own rules, its own terms. And in light of the partnership with OpenAI, it is likely that Amazon’s infrastructure will also shape how transactions flow in ChatGPT over time.

Amazon is not yet a priority channel for Nordic retailers, but it is an ecosystem you need to keep an eye on.

Serkan Selcuk, Product Owner at Viskan, says: "Three ecosystems, three separate sets of requirements for product data and transaction flows. That is the reality Nordic retailers need to deal with. A platform that only optimises for one of them leaves gaps competitors can fill."

Oskar Nilsson, Online Growth Specialist at Viskan, adds: "What we do know is that AI agents are already indexing and evaluating products. The work you do today on structured data and product quality pays off regardless of which protocol wins, and that is the best starting point for anyone who wants to be ready."

What are UCP and MCP – and what do the protocols make possible?

Behind agentic commerce are two fundamental protocols worth understanding at a high level.

Universal Commerce Protocol (UCP) – the transaction layer

UCP is the open protocol that makes direct purchases in AI interfaces possible, launched in January 2026 by Google and Shopify with support from 20+ partners. In theory, it works as a common language: instead of each platform having to build separate integrations with each AI player, the protocol is implemented once and becomes available to everyone who supports it. In practice, the picture is more complex. Google, Amazon and OpenAI are building parallel ecosystems with their own protocols and their own terms, which means separate integrations are still needed to reach all three.

In practice, UCP means an AI agent can create a checkout session, add products, fetch totals including shipping and VAT, and communicate order status, all headless, without the customer leaving the conversation.

Model Context Protocol (MCP) – the deep data layer

While UCP handles the transaction, MCP handles the context. MCP is an open protocol that gives AI agents structured real-time access to deeper data in an e-commerce platform, the kind of information that never makes it into a product feed.

Concrete examples of what MCP makes possible:

• "Order more of that coffee I bought at Christmas." The agent reads the customer’s order history, identifies the exact product and prepares the basket. Zero friction for repeat purchases.

• "Is this jacket available in size L in a shop near me?" The agent uses the customer’s location, checks store stock in real time and replies: "Yes, it is available at Knalleland. Shall I reserve it for collection in an hour?"

• B2B: "Is spare part X compatible with machine Y?" The agent reads technical specifications and compatibility data in real time, information that is rarely available in a product feed.

The difference is clear: without MCP, an LLM is a highly intelligent static reference point. With MCP, it becomes an active shop assistant with real-time access to everything the platform knows.

MCP + UCP = the full buying journey. MCP handles discovery and context. UCP handles the transaction. Together, they cover the complete agentic commerce journey from question to completed purchase.

What requirements does agentic commerce place on payment solutions?

Agentic commerce assumes payment solutions can handle a new type of transaction: one initiated by an AI system rather than by a human clicking a button. It may sound like a small difference, but it creates entirely new requirements for how payment flows are built.

What works today

The foundation is tokenised payment solutions such as Google Pay and Apple Pay. Tokenisation means the AI agent never handles actual card details. Instead, it works with cryptographic tokens tied to a specific transaction. That protects payment information and makes it possible to complete purchases without the customer needing to be active at every step.

Several leading payment providers have launched frameworks specifically for agent-initiated transactions. A core principle is that systems need to be able to verify that the AI agent completing a purchase is legitimate and acting within the limits set by the customer, in much the same way a retailer today verifies a customer’s identity.

What is needed next

For agentic commerce to work seamlessly, your payment solution needs to be able to accept payments from several different AI ecosystems such as Google Pay, Apple Pay and, over time, Amazon’s infrastructure, without you as a retailer needing separate payment flows for each channel.

Several fundamental questions still do not have answers. How should an AI agent approve a purchase when European legislation requires a human to confirm the transaction? And how should banks and retailers know that the purchase is initiated by a legitimate AI agent rather than a malicious bot? The industry is actively working on these questions, but the standards are not yet in place.

Why does agentic commerce not work the same way in Europe as it does in the US?

Agentic commerce is technically possible in the US, but it faces a different regulatory reality in Europe. The frameworks were not designed for agentic commerce, and several fundamental questions still do not have answers.

Data protection and consent

When an AI agent acts on behalf of a consumer and handles personal data such as preferences, order history and payment credentials, complex questions arise around legal basis, consent and data processing. Who is the data controller when the purchase happens inside Gemini? How is the customer’s consent documented when the agent acts on her behalf? What happens to data shared with an AI agent if the customer withdraws consent? Today, these questions do not have unambiguous answers, and they are being actively worked on at regulatory level in the EU.

Payments and liability

European payment services legislation requires the customer to actively confirm a payment, a requirement that becomes more complicated when an AI agent initiates the purchase. At the same time, liability questions remain unresolved: today, the retailer is in most cases still the Merchant of Record, meaning the legally responsible party for the transaction, returns and disputes. But as AI agents take on a more independent role in the buying journey, the question becomes who actually carries responsibility if the agent buys the wrong product or acts outside the customer’s instructions. The answers are not yet there, either in the courts or in legislation, and payment providers, platforms and lawmakers around the world are actively working to solve this.

A fragmented ecosystem

The major players are each building their own ecosystem with their own standards. On top of that, each ecosystem must also comply with European legislation around data protection, payments and AI. That adds another layer of complexity, since regulatory requirements in the EU are significantly more extensive than in the US.

This is what will happen next – probably fast

Agentic commerce is in a phase that resembles the breakthrough of e-commerce in the late 1990s: the technology exists, the infrastructure is being built, early adopters are testing, and scaling will follow soon. Europe and the Nordics are close behind the US.

What platforms and payment solutions need to deliver

Agentic commerce places demands on several fronts at once. Your e-commerce platform determines whether your products are correctly structured and machine-readable. Your PSP determines whether your payment flows can handle agent-initiated transactions. And you as a retailer create the conditions for visibility: ratings and reviews, rich product content and a site that is seen as an authority in its category. It all connects.

The right questions to ask your platform provider are:

• Do you support automatic generation of Schema.org/JSON-LD on all product pages?
• Do you offer real-time integration with the Google Merchant API, not just scheduled XML feeds?
• How do you handle MCP exposure of product catalogue and order data?
• What roadmap do you have for UCP implementation?

And for your PSP:

• Do you support tokenised payments compatible with Google Pay and Apple Pay for agent-initiated transactions?
• How do you handle SCA requirements in an agentic flow?
• Do you support UCP’s Agent Payments Protocol (AP2)?

The early window of opportunity

The retailers and platforms that are ready with correct product data, real-time integrations and tokenised payments when agentic commerce launches in the EU will be in position to build an advantage that is hard to close. Those who wait until everything is fully established and proven risk starting stage three when competitors have already finished stage one.

Serkan Selcuk, Product Owner at Viskan, predicts: "The starting gun has already gone off. The retailers and platforms that have not started building the right infrastructure will fall behind when agentic commerce reaches scale in the Nordics. Those who are ready early will build a lead that is hard to catch."

Viskan Next Generation – built for agentic commerce

Viskan Next Generation is being developed with agentic commerce as a core requirement. The platform automatically generates structured data in line with Schema.org on all product pages, integrates with the Google Merchant API for real-time product feed updates, and handles ratings and reviews in the correct structured format. Everything that determines whether your products are visible and recommended in AI search. This includes real-time store stock, making it possible for an AI agent to answer the question “is this product available in a shop near me?” directly in the conversation.

MCP implementation is part of the roadmap, giving AI agents structured real-time access to product catalogue, stock and order data beyond what product feeds can deliver.

Viskan’s AI agent Mímis adds another layer: it creates and optimises product content and generates product-level FAQs, strengthening the contextual depth and E-E-A-T signals that AI search engines reward. Through its connection to Google Search Console, Mímis identifies which searches you are underperforming on, produces suggestions and creates content that strengthens product visibility. LLMs prioritise fresh content, and Mímis makes it possible to update your site frequently with minimal effort.

We are actively following transactional agentic commerce via UCP ahead of the EU launch. When direct purchases in AI interfaces open up for the European market, we will make sure Viskan Next Generation is ready.

Common questions about agentic commerce

We help you build a successful e-commerce business.

Get in touch with us at Viskan and we will discuss which decisions could be crucial for your e-commerce business going forward.  

Sources

Google Blog: New tech and tools for retailers to succeed in an agentic shopping era, 11 January 2026 – blog.google

Google Blog: Universal Commerce Protocol updates improve AI shopping for retailers, 19 March 2026 – blog.google

Shopify Engineering: Building the Universal Commerce Protocol (2026) – shopify.engineering

TechCrunch: Google announces a new protocol to facilitate commerce using AI agents, 12 January 2026 – techcrunch.com

CNBC: OpenAI's first crack at online shopping stumbled. It's preparing for the next wave, 20 March 2026 – cnbc.com

GeekWire: Filings: How Amazon's $50B OpenAI deal actually works, 1 March 2026 – geekwire.com

Digital Commerce 360: Amazon opens up new AI-enabled Buy for Me, Shop Direct options for merchants, 11 March 2026 – digitalcommerce360.com

Osborne Clarke: Agentic Payments: a new challenge for Europe's payments ecosystem, 6 March 2026 – osborneclarke.com

Electronic Payments International: The critical role payment service providers will play in making agentic commerce reliable and trusted, 4 March 2026 – electronicpaymentsinternational.com

Mastercard: Trusting AI to buy: Agentic commerce that's secure, transparent, January 2026 – mastercard.com

Search Engine Land: New Finding: ChatGPT Sources 83% of Its Carousel Products from Google Shopping, March 2026 – searchengineland.com

TechCrunch: OpenAI raises $110B in one of the largest private funding rounds in history, February 2026 – techcrunch.com

McKinsey: The agentic commerce opportunity: How AI agents are ushering in a new era for consumers and merchants, October 2025 – mckinsey.com

Gartner: Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, August 2025 – gartner.com