Increasingly, there is no human between your shop and the buying decision, but an AI agent. ChatGPT, Gemini and Perplexity compare products, check prices and availability and increasingly add to the cart themselves. AI traffic to US retail sites grew 393% (Adobe Analytics) year over year in the first quarter of 2026. Whether your shop even appears in this agentic checkout comes down to a single question: can the AI read, understand and complete a purchase from your product data? This article shows how to operationalize product data, availability and checkout specifically for AI agents.
What agentic checkout means for your revenue
In a classic online purchase, a human navigates your shop, scrolls, compares and clicks. In agentic checkout, an AI agent takes over these steps: it researches in natural language, filters offers, assembles a cart and triggers the order. The human only sets the goal ("Find me waterproof hiking shoes under 150 euros, deliverable by Friday") and confirms at the end.
This shift is not a distant forecast. Gartner expects that by 2030 around 20% (Gartner) of digital commerce transactions will run through AI platforms, whether on-platform or via autonomous agents. McKinsey puts the global potential of agentic commerce at 3 to 5 trillion US dollars (McKinsey) annually by 2030, and Bain & Company expects 15 to 25% (Bain & Company) of US e-commerce by 2030. During the 2025 holiday season, industry analysis found that one in five (Adobe Analytics) Cyber Week orders involved an agent.
For revenue, the quality of this traffic matters most. AI referrals recently converted around 31% (Adobe Analytics) better than other traffic sources, and revenue per visit was 37% (Adobe Analytics) higher than for non-AI traffic. Those who are visible and buyable here benefit from highly qualified, purchase-ready visitors.
How the Universal Commerce Protocol (UCP) works technically and which standards sit behind it is explained in our foundational article on agentic commerce and UCP. This article is about practical operationalization: how to set up checkout, product data and UX so that agent visits actually turn into orders.
Why machine-readable product data decides conversion
An AI agent does not see your shop the way a human does. It does not read glossy photos and emotional copy, it extracts facts: name, price, currency, availability, rating, GTIN. If these fields are missing in structured form, the agent often cannot reliably process your product and, in case of doubt, recommends the competitor whose data is cleaner.
The numbers underline the gap: across an analysis of around 6 million URLs, AI-cited pages were almost three times (Ahrefs) as likely to carry JSON-LD as non-cited pages, and 53% (Ahrefs) of AI-cited pages overall used structured data. At the same time, complete Product schema remains rare: across all measured pages the Product type appears on only about 1.5% (HTTP Archive), leaving many catalogs hard for AI systems to capture.
Required fields
name, sku, brand, price, priceCurrency and availability are the minimum for an agent to capture an offer at all.
Trust fields
aggregateRating with ratingValue and reviewCount plus gtin help the agent rank your offer and prefer it over alternatives.
Service fields
shippingDetails and hasMerchantReturnPolicy answer exactly the questions agents check before a purchase: delivery time and returns.
First equip your top sellers with the full field set (including additionalProperty and gtin); for less strategic items a core set of seven fields is enough at first. This way you focus effort where it moves the most revenue. More in our guide to product data optimization with structured data.
Real-time availability: the underrated conversion lever
For human buyers an outdated availability note is annoying. For an AI agent it is a disqualifier. If an agent triggers an order for a supposedly available product that is in fact sold out, the entire autonomous process breaks. Such failures not only reduce current conversion but also the likelihood that the platform suggests your shop again in the future.
The effect of complete, concrete data is measurable: in a cross-platform study, Product and Review schema with populated fields such as price, aggregateRating and availability was cited at 61.7% (Fischman 2026), versus only 41.6% (Fischman 2026) for generic schema types like Article or Organization. The key is synchronization: stock, price and delivery time must flow from the same system that also serves your shop.
availabilityreflects real stock and updates with every sale- Delivery times are stated as a concrete range, not as vague text
- Prices including VAT and shipping are clearly assigned
- Variants (size, color) each carry their own availability and price data
- The feed updates automatically, not in manual overnight exports
An agent that finds reliable availability with you prefers to buy from you over a larger provider with uncertain data. Clean, current product data is therefore a competitive advantage that mid-sized shops can deploy consistently as well.
The agentic commerce sales channel in Shopware
In spring 2026, Shopware introduced a dedicated agentic commerce sales channel with version 6.7.10, initially as a beta. It is a new top-level sales channel that supports multiple AI platforms and thus unifies future integrations. Merchants can use it to generate dedicated product feeds, starting with a feed in JSONL format for AI platforms, and to track the business impact of AI-generated traffic.
The audience is substantial: ChatGPT counts around 800 million (Shopware) users, of whom more than 100 million (Shopware) actively use the tool for product research. AI-driven merchant traffic rose 15-fold (Shopware) comparing Q1 2026 to Q1 2025, per Shopware. The channel targets platforms such as ChatGPT, Gemini, Perplexity and others.
| Aspect | Classic shop | Agent-ready shop |
|---|---|---|
| Data delivery | HTML for humans | structured feed plus schema |
| Availability | periodic export | real-time synchronization |
| Checkout | form in the browser | agent-ready endpoint |
| Success measurement | sessions and clicks | AI channel as its own channel |
| Recommendation by AI | random | data-driven and predictable |
We work with the freely available Shopware base and set up the sales channel individually for your shop. Integration with your inventory system happens via open interfaces such as our shipping and logistics APIs and JTL-Wawi connection.
Agent-ready checkout: what is technically different
A checkout built for humans assumes visual interaction: buttons, multi-step forms, pop-ups, sometimes a cookie banner. For an AI agent these are obstacles. An agent-ready checkout instead provides a machine-addressable endpoint through which the agent can handle cart, shipping options and order programmatically, ideally following an open protocol.
Maturity still varies. The infrastructure for fully autonomous purchases, from payment through identity to authorization, is still young across the industry. One analysis shows that conversion of agent-driven purchases currently lags that of affiliate traffic by as much as 86% (ecommerceguide), simply because many shops were not built for agents. This is exactly where the opportunity lies for merchants who set up cleanly early.
{
"id": "SW-2048",
"title": "Hiking Shoe Pro GTX",
"brand": "Example Brand",
"price": "129.00",
"price_currency": "EUR",
"availability": "in_stock",
"inventory_quantity": 42,
"gtin": "4012345678901",
"shipping": { "country": "DE", "service": "standard", "days": "1-3" },
"return_policy": { "days": 30, "method": "free_return" },
"rating": { "value": 4.6, "count": 218 }
}What matters is that every record is complete and free of contradictions in itself. An agent makes decisions in fractions of a second and without asking back. If a field is missing or feed and shop contradict each other, your offer falls through the cracks. Unlike a human, an agent interprets nothing charitably: a missing VAT figure is not a cosmetic flaw, it means the total price is not computable and the offer therefore not comparable.
From data silo to agent-ready feed: the implementation
In practice, agent readiness rarely fails because of the platform, but because of distributed data. Prices sit in the shop, stock in the inventory system, delivery times with the carrier and reviews in yet another system. An AI agent, however, expects this information bundled and consistent in a single record per product. The first task is therefore to define a reliable data source that brings together all relevant fields and feeds them to the sales channel.
This step aligns with good data management in classic retail and therefore pays off twice. A central product data management forms the foundation that feeds both your shop and the AI feed. How to build this data ownership is described in our article on PIM strategy and product data ownership. This way you avoid human and machine seeing contradictory information.
- Bundle the data source: consolidate master data, prices, stock and reviews into one leading source
- Complete the fields: check required and trust fields per product, starting with the top sellers
- Output the schema: embed structured data server-side into the product pages
- Connect the sales channel: link the agentic commerce channel with the feed and run it as its own source
- Secure real time: update stock and prices event-driven instead of in an overnight export
- Measure and refine: continuously monitor aborted agent purchases and data gaps
Technically, it is worth looking at the performance of the data path. When agents and upstream AI services retrieve your product data, fast, cacheable delivery matters, an aspect we explore in our article on Store API caching for headless Shopware. A lean, well-cached data interface reduces latency and load alike.
A separately maintained AI feed that diverges from the shop creates more problems than it solves: contradictory prices, stale stock, aborted purchases. The feed should originate from the same source as your shop. This keeps information consistent and the maintenance effort manageable.
Trust and UX for human and machine at once
A common misconception is that the agent-ready shop only needs to work for machines. The opposite is true: at the end of many agent chains stands a human confirmation, and agents increasingly evaluate signals that used to be read purely by humans, such as reviews, return policies and transparency.
Acceptance is growing fast: 38% (commercetools) of US consumers have already used generative AI for shopping, and 52% (commercetools) plan to. Trust signals you structure cleanly therefore work twice over, towards the agent and towards the human who confirms at the end. How to build these signals deliberately is shown in our article on trust signals in the online shop.
Clear return rules
Structured hasMerchantReturnPolicy data answers the agent's return question before it buys and builds trust with the human.
Consistent ratings
Aggregated ratings with value and count give the agent a robust quality signal instead of unstructured prose.
Fair checkout logic
No hidden costs, no misleading scarcity cues; more in our article on dark patterns and fair checkout.
Clear delivery information
Delivery country, shipping method and delivery time as structured data, not as a footnote in body text.
If AI cannot read your store, it will not recommend it.
Paraphrased from Shopware on agentic commerce
Measure visibility: are you being recommended at all?
You cannot steer in agentic commerce what you do not measure. Unlike classic click traffic, part of the value creation ends up in the AI platforms themselves. You therefore need your own measurement points: how often does your shop appear in AI answers, with which products, for which queries?
The dynamics are enormous. In the 2025 holiday season alone (November to December), AI traffic to US retail sites grew 693% (Adobe Analytics) year over year. Those who establish a baseline today recognize early which product categories agents pick up and where data quality still falters. Methodologically this connects to measuring LLM visibility and share of model.
- Establish a baseline: in which AI answers does your assortment appear today?
- Track the AI channel separately: run the agentic sales channel as its own source
- Close data gaps: fill missing fields for the most-requested products first
- Monitor availability errors: use aborted agent purchases as an early warning signal
- Iterate: regularly check feed, schema and checkout against real agent queries
Many hurdles for agents are the same ones that deter humans: missing data, unclear availability, slow pages. A systematic analysis uncovers both, see our article on AI-assisted shop analysis.
This is how your agent-ready shop could look:
Elektronik-Shop
This article is based on data and statements from: Adobe Analytics (AI traffic Q1 2026, AI conversion and Cyber Week 2025), Gartner (digital commerce transactions forecast), McKinsey (agentic commerce potential), Bain & Company (US e-commerce share), Shopware (agentic commerce platform and release 6.7.10), Ahrefs (correlation analysis of schema and AI citations, around 6 million URLs), Fischman 2026 (cross-platform schema citation study), HTTP Archive (Web Almanac, prevalence of Product schema), commercetools (Agentic Commerce Stats 2026) and ecommerceguide (e-commerce statistics 2026). The figures cited are snapshots and may vary depending on timing and methodology.
In agentic checkout, an AI agent such as ChatGPT, Gemini or Perplexity takes over the steps a human would otherwise perform in the shop: comparing, adding to cart and triggering the order. The human sets the goal and confirms at the end. For this to work, product data and checkout must be machine-readable.
No, the principle applies to any modern shop system. Since version 6.7.10 Shopware offers a dedicated agentic commerce sales channel, which usually simplifies implementation. We work with the freely available Shopware base and connect product data and checkout individually.
That depends on your setup. Those who maintain their own cleanly structured data and a clearly attributed AI channel keep control over assortment, prices and measurement. In practice, an early, owned data strategy is the best protection against appearing as just an interchangeable supplier.
This cannot be promised across the board, as it depends on assortment, competition and platform behavior. AI referrals are, however, considered highly qualified: they recently converted around 31% (Adobe Analytics) better than other traffic sources. Those who structure cleanly early typically position themselves ahead of hesitant competitors.
As a rule, first the required fields name, price, currency and availability, then trust fields such as ratings, GTIN, shipping and return policy. Start with your highest-revenue products and expand step by step to the entire assortment.
The infrastructure for fully autonomous purchases, from payment through identity to authorization, is still young across the industry. That is why many agent chains still end with a human confirmation. A clean data and checkout foundation, however, already creates the prerequisites to use future maturity levels without a rebuild.