{"id":90819,"date":"2026-10-22T12:41:28","date_gmt":"2026-10-22T10:41:28","guid":{"rendered":"https:\/\/webraketen.space\/?post_type=glossar&#038;p=90819"},"modified":"2026-10-22T19:17:16","modified_gmt":"2026-10-22T17:17:16","slug":"agentic-commerce","status":"publish","type":"glossar","link":"https:\/\/webraketen.space\/en\/guide\/agentic-commerce\/","title":{"rendered":"Agentic Commerce: Meaning, Process and To-Dos for Merchants"},"content":{"rendered":"<p>Agentic commerce (also agentic trade) refers to a new form of e-commerce in which autonomous or semi-autonomous <a href=\"https:\/\/webraketen.space\/en\/guide\/ki-agent\/\" data-internallinksmanager029f6b8e52c=\"13\" title=\"AI Agents - Definition, Function, and Practical Applications\">AI agents<\/a> handle purchasing processes and commerce-related tasks in whole or in part on behalf of customers and businesses. The agents search for products, compare offers, make a selection, and complete the purchase, including payment.<\/p>\n<p>In doing so, the buyer hands over steps they would normally handle themselves in classic e-commerce: research, product selection and checkout. In some cases, agentic commerce also provides support after the purchase.<\/p>\n<p>The agents work within AI interfaces such as AI Mode in Google Search, Gemini or Microsoft Copilot.<\/p>\n<p>Technically, they rely on checkout and payment protocols as well as tokenized payment data.<\/p>\n<h2>What does agentic commerce mean for merchants, and does it already affect them?<\/h2>\n<p>The hype around agentic commerce currently seems large. Product research is already shifting clearly toward AI. With ChatGPT, Gemini and similar tools, prospective buyers research options, have details explained and are shown alternatives.<\/p>\n<p>If purchases are made directly there in the future, merchants should also be present there, make their products purchasable for AI agents and admit agents as customers.<\/p>\n<p>For merchants, this shifts where customers find and buy products. Responsibility for the purchase, however, remains with the merchant: they are still the Merchant of Record and therefore responsible for fulfillment, returns and support. Buying directly within an AI agent has so far worked almost only in the US. In the EU, merchants can for now mainly prepare their product data as well as shipping and return information.<\/p>\n<p>For merchants, things change mainly in three places:<\/p>\n<ul>\n<li><strong>Who accesses the shop:<\/strong> The customer is represented by an agent that accesses the shop on the customer's behalf.<\/li>\n<li><strong>What purchasability depends on:<\/strong> Whether an agent can buy a product via Google's Checkout depends on fields in the product feed and details in the Merchant Center.<\/li>\n<li><strong>How the order arrives:<\/strong> The order data reaches the merchant's backend from the agent via a protocol, and the merchant accepts or rejects it.<\/li>\n<\/ul>\n<p>Whether this already affects a merchant depends on the market, as the section on the status in 2026 in Europe and Germany shows. For merchants in the EU, the focus is currently on preparation rather than live operation.<\/p>\n<h2>How does a purchase through an AI agent work from the merchant's perspective?<\/h2>\n<p>A purchase by an AI agent follows a fixed sequence in which the roles are clearly distributed. Here is one example based on ACP, as it is described for purchases in ChatGPT:<\/p>\n<ol>\n<li><strong>Order:<\/strong> The customer states what they are looking for.<\/li>\n<li><strong>Search:<\/strong> The agent searches for suitable products and compares offers.<\/li>\n<li><strong>Selection:<\/strong> The agent compiles suitable products, and the customer selects one.<\/li>\n<li><strong>Payment Method and Approval:<\/strong> The customer chooses their payment method and approves the payment. Only then does the payment service provider issue a token that applies only to this merchant and this cart total.<\/li>\n<li><strong>Checkout:<\/strong> ChatGPT transmits the order data, together with the token, to the merchant's backend.<\/li>\n<li><strong>Acceptance and Payment:<\/strong> The merchant accepts or rejects the order and processes the payment through its existing payment service provider.<\/li>\n<li><strong>Fulfillment:<\/strong> The merchant delivers, accepts returns and provides support.<\/li>\n<\/ol>\n<p>Steps five to seven are decisive for merchants, because the merchant remains the Merchant of Record, meaning the seller responsible for the transaction. They are therefore responsible for fulfillment, returns and support. The same applies to Google's UCP checkout. In the case of a wrong delivery or a return, the merchant is the point of contact and not the provider of the AI agent.<\/p>\n<table>\n<thead>\n<tr>\n<th>Participant<\/th>\n<th>Role in the process<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Customer<\/td>\n<td>Formulate the request, make a selection or set a framework, approve payment<\/td>\n<\/tr>\n<tr>\n<td>AI agent, e.g. ChatGPT<\/td>\n<td>Search and compare products, transmit order data to the merchant<\/td>\n<\/tr>\n<tr>\n<td>Payment service provider, e.g. Stripe<\/td>\n<td>Issue a token bound to the merchant and cart total<\/td>\n<\/tr>\n<tr>\n<td>Merchant as Merchant of Record<\/td>\n<td>Accept or reject the order, process payment, deliver, take over returns and support<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h3>Which steps stay with the customer: individual approval or mandate<\/h3>\n<p>How much the agent may do on its own is tiered. Users can always decide for themselves whether the agent must ask for permission before each purchase or may shop independently within a pre-approved framework. A comparable case is a person who takes along a shopping list and a card with a fixed limit: they select items in the store themselves and pay at the checkout. For merchants this makes a difference, because in the second case an order arrives without the customer being involved at the moment of purchase.<\/p>\n<p>The agent takes over search and comparison in both cases. Differences exist in three steps: who selects the product, who confirms the payment and who triggers the purchase. At these points, the level of autonomy of a specific offer can be read.<\/p>\n<table>\n<thead>\n<tr>\n<th>Step<\/th>\n<th>With Individual Approval<\/th>\n<th>With Open Mandate<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Search and compare products<\/td>\n<td>Agent<\/td>\n<td>Agent<\/td>\n<\/tr>\n<tr>\n<td>Select product<\/td>\n<td>Agent proposes, customer selects<\/td>\n<td>Agent, within predefined conditions<\/td>\n<\/tr>\n<tr>\n<td>Approve payment<\/td>\n<td>Customer, for every purchase<\/td>\n<td>Customer, once in advance for the framework<\/td>\n<\/tr>\n<tr>\n<td>Trigger purchase<\/td>\n<td>only after confirmation by the customer<\/td>\n<td>Agent, without customer involvement at the moment of purchase<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>The table shows both ends of the spectrum. For individual approval, there are Google's Native Checkout and purchases through the <a href=\"https:\/\/developers.openai.com\/commerce\" target=\"_blank\" rel=\"noreferrer noopener\">Agentic Commerce Protocol (ACP)<\/a> in ChatGPT. For the open mandate, there is the <a href=\"https:\/\/ap2-protocol.org\/\" target=\"_blank\" rel=\"noreferrer noopener\">Agent Payments Protocol (AP2)<\/a>.<\/p>\n<p>With Google's Native Checkout (<a href=\"https:\/\/developers.google.com\/merchant\/ucp?hl=de\" target=\"_blank\" rel=\"noreferrer noopener\">Universal Commerce Protocol, UCP<\/a>), the user discovers eligible products in AI Mode or Gemini and buys them directly there. To pay, the agent hands over to a Google interface. There, the user confirms the purchase themselves and pays with Google Pay, without leaving Google's shopping interface.<\/p>\n<p>AP2 goes furthest, as it recognizes pre-granted, cryptographically signed mandates. This framework is recorded in mandates that set, for example, a budget, an amount range, and an execution period. In AP2, such conditions can be specified as restrictions on a Payment Mandate. The agent then acts within this framework without the customer being involved at the moment of purchase.<\/p>\n<p>The choice between individual approval and a mandate has consequences. More room for the agent means less control over the individual decision.<\/p>\n<p>Individual approval takes more clicks, but it keeps every decision with a human.<\/p>\n<p>A mandate saves these steps, but within the framework, the agent makes decisions that the user no longer checks individually.<\/p>\n<h2>Where Does Agentic Commerce Stand in Europe and Germany in 2026?<\/h2>\n<p>Availability differs by region and layer. The following information reflects the status as of October 2026. Because providers launch features quickly and sometimes withdraw them, every statement should be read with its date.<\/p>\n<h3>The Major Launches Since 2025<\/h3>\n<ul>\n<li><strong>September 2025:<\/strong> OpenAI and Stripe introduce Instant Checkout in ChatGPT and the ACP.<\/li>\n<li><strong>January 2026:<\/strong> Google announces UCP at NRF 2026. As of October 2026, UCP checkout runs as an early access program for selected merchants.<\/li>\n<li><strong>March 2026:<\/strong> OpenAI announces that it will move Instant Checkout into merchant apps, that is, toward purchases through the merchants' connected apps.<\/li>\n<\/ul>\n<h3>Providers and Pilot Projects in Europe<\/h3>\n<p>The documented European cases come mainly from banks, payment service providers and card networks. According to Mastercard, Worldline and ING completed an agent transaction live in which all parts of the payment process were in Europe. Mastercard describes such purchases as controlled, consent-based scenarios in which consumers authorize an agent to buy.<\/p>\n<p>Further reported projects from 2026:<\/p>\n<ul>\n<li><strong>Santander and Mastercard:<\/strong> Both describe a payment via Mastercard Agent Pay as the first live payment in Europe fully executed by an AI agent.<\/li>\n<li><strong>Worldline and ING (June 2, 2026):<\/strong> They also describe their transaction as the first agent payment in Europe in production operation, with explicit consumer consent.<\/li>\n<li><strong>Nordea and Mastercard (June 5, 2026):<\/strong> They report an agent transaction in Finland.<\/li>\n<li><strong>PayPal and Hey Savi (June 2, 2026):<\/strong> They are launching an agentic commerce platform in the United Kingdom, with checkout in the app. The first merchant is the Debenhams Group.<\/li>\n<li><strong>Adyen:<\/strong> The European payment service provider launches Adyen Agentic in June 2026.<\/li>\n<li><strong>Visa:<\/strong> The Visa Agentic Ready program starts first in Europe, with Revolut, Barclays and Commerzbank among others. The institutions are to test and validate agent-initiated transactions.<\/li>\n<\/ul>\n<p>That several companies claim to have made Europe's first agent payment shows the nature of these figures: they are self-reported statements by the parties about individual transactions, not independently verified market data. For OpenAI, a secondary source reports that Instant Checkout was expanded to Europe at the end of August 2026. OpenAI has not yet confirmed this.<\/p>\n<h3>Status in Germany, Austria and Switzerland<\/h3>\n<p>In the German-speaking region, buying through AI agents is barely usable so far. UCP checkout is, according to <a href=\"https:\/\/support.google.com\/merchants\/answer\/16837055?hl=en\" target=\"_blank\" rel=\"noreferrer noopener\">Google<\/a> so far only for eligible products in the US, Canada and Australia. Whether ACP is enabled for EU merchants is not documented. According to Adyen, Europe is still largely in an early test phase: AI supports search and the cart, and the customer approves the purchase themselves. According to the digital consultancy diconium, autonomous checkout is currently not available in Europe, because legal requirements on data protection, payment transactions and regulation slow down implementation.<\/p>\n<p>Commerzbank, a German bank, is participating in the Visa Agentic Ready test program. According to Mastercard, banks from Germany, Austria, and Switzerland have also completed controlled live agent transactions, including Deutsche Bank, DZ Bank, Erste Bank \u00d6sterreich, and Viseca.<\/p>\n<h3>How Many Agent Purchases Are Already Happening<\/h3>\n<p>How many transactions through AI agents already run in Europe or Germany is an open question: no verified figures exist. Only reports about individual transactions and pilot projects are public. According to a secondary source, fully autonomous agent purchases with payment remain a fringe phenomenon in 2026. According to eMarketer, Europe is largely in a co-pilot phase, in which consumers prefer to have AI assist them while shopping rather than hand over the transaction to it.<\/p>\n<p>The common studies measure different things, and their figures cannot be compared directly. What matters is whether they capture the use of AI for research and advice or the willingness to let an agent make the purchase itself. Only the latter matches the term in its narrow sense.<\/p>\n<ul>\n<li><strong>AI Use When Shopping:<\/strong> In a Deloitte survey of 13,500 European consumers, more than half already use AI when shopping, according to commercetools. Germany, the UK and France are at the lower end, with 48 to 52 percent. What is measured here is mainly research and comparison.<\/li>\n<li><strong>Trust in Delegated Purchases:<\/strong> A study by Sopra Steria of 8,400 consumers in eight countries, including Germany, found in February 2026, according to a secondary source: 55 percent have heard of agentic commerce, but only 13 percent say they really know about it. 41 percent do not trust any existing provider with a delegated purchase.<\/li>\n<li><strong>Market Forecasts:<\/strong> The McKinsey figure, which states that agentic commerce could \u201corchestrate\u201d 3 to 5 trillion US dollars worldwide by 2030, is a forecast, not a market size.<\/li>\n<\/ul>\n<h2>To-dos for merchants: product feed, protocols, checkout and agents<\/h2>\n<p>Currently, the most reliable primary documentation comes from Google (UCP and Merchant Center) and applies for now to products in the United States, Canada, and Australia, but not the EU. In the EU, the focus should therefore be on preparing the data foundation: product feeds, shipping and return information, and legal texts. The following points show what purchasability specifically depends on once activation takes place.<\/p>\n<h3>Prepare product feed and Merchant Center<\/h3>\n<p>Whether an agent can buy a product through UCP checkout depends on fields in the product feed and on information in Merchant Center:<\/p>\n<ul>\n<li><strong><code>native_commerce(checkout_eligibility)<\/code>:<\/strong> The attribute contains a Boolean and defaults to FALSE. If it is missing or set to FALSE, the product is not checkout-eligible.<\/li>\n<li><strong>Product-<code>ID<\/code>:<\/strong> The ID in the feed must match the product ID that the checkout APIs expect. If the two differ, the custom attribute <code>merchant_item_id<\/code> maps them to each other.<\/li>\n<li><strong>Excluded categories:<\/strong> For products from categories that Google excludes from checkout, <code>native_commerce(checkout_eligibility)<\/code> must remain empty or be set to FALSE.<\/li>\n<li><strong><code>consumer_notice<\/code>:<\/strong> For items subject to legal warning requirements, this attribute is mandatory.<\/li>\n<li><strong>Return Policy:<\/strong> The return conditions must be stored in Merchant Center.<\/li>\n<li><strong>Customer Support:<\/strong> A customer support contact is mandatory.<\/li>\n<\/ul>\n<p>Beyond the feed, price, stock and delivery time should match across all systems. As merchant of record, the merchant is responsible for real-time inventory checks, according to Google.<\/p>\n<h3>Which protocols are relevant for merchants<\/h3>\n<p>The protocols can only be meaningfully compared when they are assigned to their layer. An agent purchase runs through four layers:<\/p>\n<ul>\n<li><strong>Discovery:<\/strong> The agent finds products. Adapting for AI search affects only this layer, that is, discoverability in AI answers, not the execution of the purchase.<\/li>\n<li><strong>Checkout:<\/strong> The agent and the merchant's checkout exchange cart, shipping and order details. ACP and UCP operate here, and according to Google, UCP also covers discovery and post-purchase support.<\/li>\n<li><strong>Payment Authorization:<\/strong> The payment is released in a verifiable way. AP2 and the card network tokens operate here.<\/li>\n<li><strong>Settlement:<\/strong> The money flows through the card networks.<\/li>\n<\/ul>\n<p>Agentic checkout is a subset of agentic commerce. It refers only to the completion phase of an agent purchase, i.e. cart, shipping, payment and confirmation, without discovery and comparison of products. Within this layered model, agentic checkout covers the checkout layer, on which ACP and UCP operate, and payment authorization, for example via AP2.<\/p>\n<table>\n<thead>\n<tr>\n<th>Protocol<\/th>\n<th>Layer<\/th>\n<th>Purpose<\/th>\n<th>Behind the Protocol<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>ACP (Agentic Commerce Protocol)<\/td>\n<td>Checkout<\/td>\n<td>Purchases from within ChatGPT<\/td>\n<td>OpenAI and Stripe, introduced in September 2025<\/td>\n<\/tr>\n<tr>\n<td>UCP (Universal Commerce Protocol)<\/td>\n<td>Checkout, according to Google also discovery and post-purchase support<\/td>\n<td>Agent actions in Google AI Mode and Gemini<\/td>\n<td>Google, co-developed with Shopify; according to a secondary source also with Etsy, Wayfair, Target and Walmart<\/td>\n<\/tr>\n<tr>\n<td>AP2 (Agent Payments Protocol)<\/td>\n<td>Payment authorization<\/td>\n<td>Prove customer consent through signed checkout and payment mandates<\/td>\n<td>Developed by Google, announced in September 2025, in the standardization process of the FIDO Alliance since April 2026<\/td>\n<\/tr>\n<tr>\n<td>MCP (Model Context Protocol)<\/td>\n<td>General, not commerce-specific<\/td>\n<td>Connecting AI agents with tools and data sources<\/td>\n<td>Also used in the commerce stack<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Payment authorization is a separate layer and not a replacement for a checkout protocol. Even an agent connected via MCP still needs ACP or UCP for checkout according to a secondary source, and AP2 or Visa's Trusted Agent Protocol (TAP) for authorization. According to a secondary source, UCP is an open-source standard, while ACP is an open standard with access for approved partners.<\/p>\n<p>The specifications are currently changing quickly. At the time this guide was written, AP2 is available in version 0.2. According to a security analysis, this version adds new features and addresses some earlier weaknesses. Every implementation must therefore be checked against the current specification version, not the first announcement.<\/p>\n<h3>Native Checkout or Purchase in Your Own Shop?<\/h3>\n<p>Merchants choose between two models. With native checkout via UCP or ACP, the customer buys directly in the AI interface. The merchant remains merchant of record, with all obligations for fulfillment, returns and support. Which protocol is needed depends on the target interface: for ChatGPT it is ACP, and for Google AI Mode and Gemini it is UCP.<\/p>\n<p>The second pattern is: discover in AI, buy in your own shop. In March 2026, OpenAI announced that it would move Instant Checkout into connected merchant apps. The ACP protocol should be viewed separately from this, so this shift should not be read as the end of the standard.<\/p>\n<p>For shops on platforms such as Shopify, much depends on the platform itself. According to Gladly, UCP support is being rolled out directly via Shopify. The real work then lies in knowing which of one's own platforms deliver this and in switching it on.<\/p>\n<h3>Recognize Agents Instead of Blocking Them<\/h3>\n<p>On the merchant side, the line between a legitimate agent and a fraud bot is blurred, because according to Adyen, a legitimate shopping agent can look just like a scraper or an exploit bot. However, legitimate agents can be identified rather than blocked across the board as bots. UCP requests are recognizable by the header <code>UCP-Agent:Profile<\/code> and by a User-Agent that begins with <code>Google\/UCP<\/code> begins. According to the UCP specification, merchant profiles themselves are located under <code>\/.well-known\/ucp<\/code>, where agents retrieve them.<\/p>\n<p>For card payments, Visa introduced the Trusted Agent Protocol, launched with Cloudflare: merchants check an agent's signature against a directory from Visa. This allows merchants to decide whether a request comes from a known agent. Bot defenses that reject such requests without checking, however, also exclude customers who are willing to pay.<\/p>\n<h2>What gives customers trust and what merchants can do about it<\/h2>\n<p>An agent purchase requires customers to entrust their payment data and a purchase decision to the agent. According to McKinsey, cited by diconium, only 30 percent of European consumers would let an AI place an automatic order. It is therefore relevant for merchants which protective mechanisms exist, where they end and who is liable for wrong purchases.<\/p>\n<h3>How payment data is protected in agent purchases<\/h3>\n<p>Anyone who hands their payment data over to AI may be concerned about its security.<\/p>\n<p>With ACP using the Shared Payment Token and with AP2, the agent does not receive real card data. There it works with proxies that are bound to a merchant, an amount or an approved framework. According to Google, its UCP checkout currently uses the card numbers that users have saved in Google Wallet (FPANs) in the Google Pay flow. For merchants, little changes in processing, because they handle payment through their existing payment service provider.<\/p>\n<p>Two types of token replace the card number in agent purchases:<\/p>\n<ul>\n<li><strong>Shared Payment Token (SPT):<\/strong> Stripe issues a token that allows the AI application to trigger a payment without disclosing payment data. The token applies only to a specific merchant and a specific cart total, and the AI application never receives the real card data.<\/li>\n<li><strong>Network Tokens:<\/strong> The card networks also work with tokens for agents. As part of Agent Pay, Mastercard offers so-called \u201cAgentic Tokens.\u201d<\/li>\n<\/ul>\n<p>Payments that an AI agent authorizes, triggers or executes on behalf of a user or company are called agentic payments. They form the payment layer of agentic commerce, but extend beyond retail, for example to micropayments between machines or invoice payments in businesses. Technically, they rely on mechanisms such as Shared Payment Tokens, network tokens and signed mandates.<\/p>\n<h3>What Prevents Unwanted Charges<\/h3>\n<p>AP2 works with two signed mandates. The Checkout Mandate secures what is bought, and the Payment Mandate secures the payment for this checkout. Both exist in closed form for a specific cart or in open form, which records conditions in advance. This results in two modes: the customer approves the finished cart in real time, or they set up open mandates in advance, within which the agent later acts without them.<\/p>\n<p>In both cases a double safeguard arises, because the merchant cannot charge more than approved and the agent cannot spend without authorization. An SPT limits the merchant and amount of a single purchase, an open mandate limits a framework across several purchases, and the signed mandates serve as proof of consent in a dispute. The greater an agent's leeway, the more important tightly set limits become.<\/p>\n<h3>Where Protection Ends<\/h3>\n<p>Signed mandates secure the transaction data, i.e. items, amount and approval, from the signature onward, but not the context that shapes the selection beforehand. In an indirect prompt injection, instructions are hidden in web pages or product texts, and the agent adopts them when reading. In this way, product choice and ranking can be influenced even though the mandates are valid. The agent then buys with formally correct authorization, but the wrong product or one that has been manipulated to be preferred.<\/p>\n<p>For AP2 version 0.1, earlier work described replay attacks (reuse of a valid approval) and prompt injection. A security analysis of version 0.2 (Aviv et al., arXiv 2608.23858) notes that v0.2 addresses some of these problems. Inputs that shape the transaction before the signature, however, remain outside the protection. The semantic attack surface therefore lies before the signature, and better cryptography does not close it.<\/p>\n<p>For customers, tight limits and individual approvals for higher amounts therefore remain sensible, because none of the protective mechanisms prevents a manipulated selection. For merchants, it follows that their product texts and product data are part of the context the agent reads and on the basis of which it selects.<\/p>\n<h3>Open liability question for wrongful purchases<\/h3>\n<p>Who is liable for an agent's wrong decision depends, according to <a href=\"https:\/\/www.haendlerbund.de\/de\/ratgeber\/recht\/agentic-commerce\" target=\"_blank\" rel=\"noreferrer noopener\">H\u00e4ndlerbund<\/a> probably under the general rules of civil law and depends on the individual case. If the agent acts within the given parameters, its decision is attributed to the user, and the contract is generally valid. If the merchant caused the wrong decision with incorrect product data, the merchant may be liable.<\/p>\n<p>Toward the customer, the merchant is liable for everything that concerns the contract: price, product and delivery. According to H\u00e4ndlerbund, the provider of the AI agent is liable to the merchant for error-free AI infrastructure, correct interfaces and data security.<\/p>\n<p>Under German law, avoidance of the contract is mainly possible in cases of technical transmission errors (comparable to \u00a7 120 BGB). If the agent exceeded the set limits, such as a price limit, avoidance may be considered in individual cases. The right of withdrawal remains in place for consumers even for purchases made by an agent, but it does not apply to businesses. As stated in the H\u00e4ndlerbund article cited above, it remains unclear whether software falls under the Product Liability Act and how the final EU liability rules will be designed.<\/p>\n<h3>Where Payment Regulation Falls Short<\/h3>\n<p>European payment regulation has not yet been written for agents. According to Adyen, it still assumes that a person authorizes every individual transaction. The Payment Services Directive PSD2 requires strong customer authentication, and according to a secondary source, it is open how far this authorization can be delegated to an agent. This fits poorly with open mandates, in which the agent later buys without the customer.<\/p>\n<p>How agent purchases fit with the GDPR still appears unresolved as of today. Legislation is lagging behind technical development here. Courts may also have to decide this on a case-by-case basis.<\/p>\n<h2>How Does Agentic Commerce Differ from Conversational Commerce?<\/h2>\n<p><a href=\"https:\/\/webraketen.space\/en\/guide\/conversational-commerce\/\" data-internallinksmanager029f6b8e52c=\"30\" title=\"Conversational commerce\">Conversational commerce<\/a> (auch C-Commerce) bezeichnet den dialogbasierten Online-Handel. Hier werden Produkte und Dienstleistungen direkt \u00fcber Konversationstools wie Chatbots (ChatGPT, Gemini, Claude), Messaging-Dienste oder Sprachassistenten in Echtzeit gesucht, beraten und gekauft.<\/p>\n<p>To distinguish agentic commerce from C-Commerce, what matters is whether the action is delegated or the dialogue is.<\/p>\n<p>In conversational commerce, the purchase happens in conversation (chat, voice, messenger), but the human makes every decision themselves. A chatbot that only recommends products is therefore advice, after which the customer goes to the checkout themselves.<\/p>\n<p>In agentic commerce, the agent carries out the transaction. The conversation is therefore only an interface: \u201cagentic\u201d refers to acting on someone's behalf, not to the conversation.<\/p>\n<table>\n<thead>\n<tr>\n<th>Model<\/th>\n<th>Who selects?<\/th>\n<th>Who triggers the purchase?<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Classic e-commerce<\/td>\n<td>Human<\/td>\n<td>Human<\/td>\n<\/tr>\n<tr>\n<td>Conversational commerce<\/td>\n<td>Human, supported in dialogue<\/td>\n<td>Human<\/td>\n<\/tr>\n<tr>\n<td>Agentic Commerce<\/td>\n<td>Agent, within the framework set by the human<\/td>\n<td>Agent, depending on level, after approval<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>The term itself is not used uniformly, however. The narrow, buyer-side meaning is used by, among others, the provider Eco: unlike a recommendation engine or a chat assistant, an agent carries out the transaction itself.<\/p>\n<p>In addition, some providers use the term for AI agents in retail in general. Salesforce, for example, describes agents that act on behalf of users or companies and that also manage inventory, automate customer service or personalize recommendations. This broad meaning lacks the feature \u201cacting on behalf of the buyer\u201d. For specialist texts and vendor claims on the topic, it is therefore first necessary to check which meaning is intended.<\/p>\n<p>An AI shopping assistant that only recommends, compares and advises therefore does not fall under the narrow meaning. That changes as soon as it executes purchases, as Amazon envisions with \u201cBuy for Me.\u201d In practice, the models often overlap, because a dialogue comes first and execution follows. For classification, the last step matters, not the interface.<\/p>","protected":false},"excerpt":{"rendered":"<p>Agentic Commerce (auch agentischer Handel) bezeichnet eine neue Form des E-Commerce, bei der autonome oder halbautonome KI-Agenten Kaufprozesse und handelsbezogene Aufgaben ganz oder teilweise im Auftrag von Kundinnen und Unternehmen \u00fcbernehmen. Die Agenten suchen Produkte, vergleichen Angebote, treffen eine Auswahl und wickeln den Kauf samt Bezahlung ab. Der K\u00e4ufer gibt dabei Schritte ab, die er [&hellip;]<\/p>\n","protected":false},"author":5,"featured_media":0,"template":"","class_list":["post-90819","glossar","type-glossar","status-publish","hentry"],"_links":{"self":[{"href":"https:\/\/webraketen.space\/en\/wp-json\/wp\/v2\/glossar\/90819","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/webraketen.space\/en\/wp-json\/wp\/v2\/glossar"}],"about":[{"href":"https:\/\/webraketen.space\/en\/wp-json\/wp\/v2\/types\/glossar"}],"author":[{"embeddable":true,"href":"https:\/\/webraketen.space\/en\/wp-json\/wp\/v2\/users\/5"}],"version-history":[{"count":14,"href":"https:\/\/webraketen.space\/en\/wp-json\/wp\/v2\/glossar\/90819\/revisions"}],"predecessor-version":[{"id":90864,"href":"https:\/\/webraketen.space\/en\/wp-json\/wp\/v2\/glossar\/90819\/revisions\/90864"}],"wp:attachment":[{"href":"https:\/\/webraketen.space\/en\/wp-json\/wp\/v2\/media?parent=90819"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}