Agentic Commerce: AI Agents as Purchase Mediators

Agentic commerce moves the purchase decision from the search results page into a dialogue with an AI agent. Instead of browsing online stores themselves, users state a goal, and the agent discovers, compares, and ideally buys the product. For B2B sales in the DACH region, this creates a new, machine-mediated visibility layer between brand and customer. This article puts the term, protocols, platforms, and measurement into context and provides an action plan for Austrian and DACH companies.

Why agentic commerce is becoming relevant now

The prerequisite for agentic shopping is simple: people have to use AI systems in their everyday lives. In the DACH region, that is now reality. In Germany, 67 percent of people aged 16 and over used generative AI such as ChatGPT, Microsoft Copilot, or Google Gemini, a jump of 27 percentage points from summer 2024 (40 percent at the time). Among users, ChatGPT leads with 43 percent, followed by Copilot with 39 percent, Gemini with 28 percent, and Perplexity with 1 percent. These Bitkom figures refer to Germany, not the entire DACH region, but they show the direction: the chat interface is becoming the standard interface. If you don't show up there, you don't make the shortlist.

This shift hits marketing in a sensitive spot. When the agent does the research and comparison, traditional touchpoints such as the product detail page or the paid click partly drop out of the customer's view. Instead, the brand has to appear in the generated answer itself. This is exactly where agentic commerce connects with generative engine optimization (GEO): visibility in the AI answer is the entry ticket to the agentic purchase process.

How it works: discovery, protocols, and platform storefronts

Agentic commerce is not the same as AI product search. It pays to separate three stages. Discovery: The agent finds and names products in its answer, but the purchase still happens in the online store. Shortlisting: The agent actively narrows down the options and makes recommendations. Autonomous purchase: The agent triggers the transaction inside the AI interface. Traditional e-commerce ends when the user clicks the buy button; agentic commerce moves this step into the agent. The first two stages are widespread today. The third is technically possible but not yet broadly established.

For the agent to sell anything at all, it needs an infrastructure layer. Several standards compete here:

• Agentic Commerce Protocol (ACP): Developed by OpenAI together with Stripe and published as open source. It governs how an agent initiates an order and processes the payment. ACP was the basis for Instant Checkout in ChatGPT. According to OpenAI, ACP remains the infrastructure that connects users with merchants across the entire purchase journey.
• Universal Commerce Protocol (UCP): Google's open standard for agentic shopping. In January 2026, Google announced that UCP would power a new checkout feature on eligible product listings in AI Mode and the Gemini app, letting customers buy directly from selected US merchants. Google names Shopify, Etsy, Wayfair, Target, and Walmart among its partners. According to trade press reports, the checkout was limited to eligible US merchants in May 2026 and at that point also appeared in regular search results for the first time. In the same month, Google announced plans to roll out UCP checkout in Canada and Australia, followed by the UK. Google did not mention the DACH region.
• Model Context Protocol (MCP): An open standard developed by Anthropic for connecting agents to tools and data sources, housed since December 2025 at the Agentic AI Foundation under the umbrella of the Linux Foundation. MCP is not a payment protocol. It is the generic bridge through which an agent can retrieve structured product and inventory data. ACP and UCP handle checkout, MCP handles data access. So the standards partly compete and complement each other on different levels.

On the interface, these protocols turn into visible storefronts. Google AI Mode introduces a new ad format called Direct Offers: merchants can show exclusive offers such as a 20 percent discount directly in AI Mode to users who are ready to buy, with Google using AI to determine when an offer is relevant. Brands taking part in the pilot include Petco, e.l.f. Cosmetics, Samsonite, and Rugs USA. OpenAI had introduced Instant Checkout in September 2025 for completing purchases directly in ChatGPT. In March 2026, OpenAI scaled it back, and purchases are now meant to run through merchant and partner apps inside ChatGPT. According to reports by The Information, hardly any merchants had signed on. Perplexity and Microsoft Copilot are building similar purchase-oriented interfaces. What matters for the DACH market: these pilot programs launch first in the US with selected merchants. The Austrian market therefore lags behind the US market in maturity.

GEO as a prerequisite: how your brand gets cited in AI answers

Before an agent can buy, it has to know and name the brand. Visibility in generative answers follows different rules than traditional ranking. The Princeton-affiliated study on generative engine optimization shows that targeted optimization can increase visibility in generative answers by up to 40 percent. The study notes that the most effective strategies vary by subject area, but it highlights substantively enriched content. In practice, this means the following for agentic commerce:

• Front-loading: The key information, meaning what the product is, who it suits, and what it costs, belongs in the first sentences, not at the end. AI systems prefer to extract clear, early statements.
• Answer islands: Self-contained answer blocks that fully answer a specific question are easier to cite than scattered information.
• Brand mentions instead of backlinks: For citability in AI answers, consistent mentions of the brand in a relevant context often count for more than the traditional backlink. Generative systems aggregate mentions across many sources.
• Statistics and quotations: According to GEO research, sourced figures and verifiable statements increase the likelihood of being used as a source.

Best practices: structured data and feed hygiene for AI agents

Once the brand is being cited, the agent needs machine-readable facts to compare and buy. The following measures form the technical foundation:

• Maintain a clean Product schema: Structured data based on schema.org/Product with price, availability, brand, and unique identifiers (GTIN, MPN) gives the agent reliable attributes instead of guessed values.
• Keep merchant feeds up to date: Product feeds must reflect price and stock changes promptly. An agent that recommends a product that is no longer available damages brand trust immediately.
• Keep availability and price attributes consistent: Contradictory prices between feed, schema, and store lead the agent to classify the source as unreliable.
• Provide llms.txt: The llms.txt file in the root directory serves as a machine-readable guide for AI systems. It can point to key product information, terms, and data sources and so make it easier to be included in agentic processes. The standard is voluntary and not every provider reads it, but it is part of feed hygiene for machine-readable online stores.

For B2B providers in the DACH region, this means in concrete terms: first get your data foundation in order; only then does piloting ACP or UCP pay off. A technically clean feed is the condition for appearing in agentic purchase journeys at all.

Common mistakes and the question of trust

The most common strategic mistake is assuming that consumers will immediately hand autonomous purchasing over to AI. The data clearly contradicts this. According to a Gartner survey of 322 US consumers, the willingness to let AI make purchase decisions on its own peaked at 11 percent, and only in low-risk categories such as personal care and household supplies. Respondents are more open to shortlisting: 31 percent would let AI narrow down the selection for household supplies, 28 percent for personal electronics. These figures come from a US survey and cannot be transferred one-to-one to the DACH region, but the pattern holds: trust in AI for research and shortlisting, reservations about autonomous purchases.

The same pattern showed up in ChatGPT. According to a source cited by The Decoder, users researched products there but rarely bought. The Decoder sees this as one reason why OpenAI scaled back its own checkout.

The strategic consequence is human-in-the-loop. Your conversion strategy should keep the human at the decision threshold instead of replacing them. If you optimize for the next, purchase-intensive stage, you should also invest in the shortlisting stage that dominates today.

A second mistake is neglecting platform dependency. When the agent controls the storefront, the platform controls the terms. Discount formats such as Direct Offers can create margin pressure, and dependence on a single provider is a real risk. The Bitkom finding that 68 percent consider Germany too dependent on the US and China for AI reflects this concern at a societal level. If you pilot agentic commerce, you should deliberately safeguard your own data sovereignty and your direct relationship with the customer.

Metrics and measurement: AI referral traffic and new GEO KPIs

AI-mediated traffic is considered closer to a purchase because the agent has already clarified the need. You can only prove this assumption if you track the traffic cleanly. In GA4, you can isolate referrals from ChatGPT, Perplexity, Gemini, or Copilot via referral sources and regex filters and analyze them as a separate channel group. Only then can you see whether AI sessions actually convert better.

Beyond traditional rankings, GEO-specific metrics are emerging:

• Share of voice: How often the brand is named in AI answers to relevant questions, relative to the competition.
• Citation rate: The share of answers in which your own source is cited or linked.
• AI-attributed orders: Orders that can be traced back to an AI session or an agentic checkout.

These metrics are still young and not standardized, but they form the core of reliable agentic commerce reporting. If you set them up early, you can quantify the value of AI visibility compared with the traditional funnel.

B2B and DACH specifics: procurement, data protection, maturity

In B2B, agentic purchasing extends beyond the consumer space. Gartner predicts that AI agents will mediate more than 15 trillion US dollars in B2B spending by 2028 and that 90 percent of all B2B purchases will be handled by AI agents within three years. These figures are an international forecast, not a DACH-specific status quo, but they mark the scale of the upheaval in procurement.

At the same time, people remain central to the B2B process. According to Gartner, 69 percent of B2B buyers prefer to validate AI-generated insights with sales reps. The survey does not name a region, so it is not established as a DACH finding. It does support the thesis that in B2B, AI complements sales rather than replacing it: human sellers act as trusted validators.

For Austria and the DACH region, two further factors come into play. First, the regulatory context: data protection under the GDPR and the requirements of the EU AI Act shape how agentic purchase processes may handle data. Second, maturity: the main platform storefronts launch in the US. Whether and when they reach the DACH region is open. Austrian companies therefore have a window to build their data foundation and visibility before agentic purchase journeys become widely available locally. Note: the regulatory information is for general guidance only and does not replace legal advice.

Practical action plan and further reading

For DACH companies, a pragmatic sequence emerges. First, build visibility in AI shopping by optimizing content according to GEO principles (front-loading, answer islands, sourced figures, consistent brand mentions). Second, establish feed hygiene: keep Product schema, merchant feeds, and price and availability attributes clean, and provide llms.txt. Third, pilot ACP and UCP in a targeted way as soon as the platforms reach the DACH market, starting with lower-risk product ranges. Fourth, set up measurement: isolate AI referral traffic in GA4 and track share of voice, citation rate, and AI-attributed orders. Fifth, actively manage risks, especially margin pressure from discount formats and platform dependency, and safeguard the direct relationship with the customer.

Agentic commerce is therefore less a single tool than a shift in the interface. Companies that make their data machine-readable now and are present in AI answers lay the groundwork for showing up at all in the next stage of online retail.