Agent Channel · Article series · 2026

Is your shop agent-ready?

Your customers ask AI assistants what to buy and where. To answer, the assistant reads shops, price comparison sites and feeds, recommends a product and a retailer, and on request puts the item in the cart. Today this happens through the same interface your customers see, or not at all if the shop turns the program away. An Agent Channel is a separate entrance for agents into your shop. It recognises who is asking and gives each agent the data you intend for it.

What happens today

Three observations from our own tests in November 2026:

  • Assistants recommend what they could read. They take the price from the product page or from a price comparison site, often from an older snapshot. ChatGPT told us: "The available retailer prices are from September, however." (Article 1)
  • Agents already buy. At our request, ChatGPT put a cordless drill kit in an online retailer's cart. To do so, it opened a browser in the cloud. The retailer saw a visitor like any other. (Article 1)
  • Many shops turn new agents away. Of 30 shops whose product pages we retrieved with our own checking program, 19 did not hand over a single product page. Only 7 delivered price and product data without a browser having to build the page first. (Article 5)

Two levels

Agent-Ready. The shop is readable for agents and hands them data selectively.

  • Recognition and policy. Every request is assigned a level, from an unknown program to an agent with a customer account. One rule applies to the CDN, the robots.txt and the interfaces, and the business team writes it.
  • Reading path and agent catalogue. First, readable product pages with price, GTIN and stock. Then search, product, price and stock as an interface, described to the outside world in the Universal Commerce Protocol and fed internally from the same data as your feeds.
  • Selective delivery. List price and availability in public, more depth per credential, quotas per operator.
  • Cockpit. Requests per operator and level, price parity across all channels, rules rolled back with one click.

Store Agent. An agent run by the retailer answers the questions of incoming agents.

  • It answers questions on fit, compatibility, delivery and returns from your product data. Price and stock come from the interface, never from the language model.
  • Your pricing system calculates offers by your rules: volume tiers, customer group, price floor. The Store Agent passes them on.
  • A manifest sets out what the Store Agent may do and what stays with a human, per customer group, marketplace and channel.
  • Cost prices and price floors are calculated by models that run in-house.

The Store Agent talks to your customers' agents. You still reach the customers themselves through your shop.

For technical product ranges

In B2B, customer prices are behind the login. That can stay as it is. The buyer links their customer account to their assistant once, and from then on the agent asks with the buyer's permissions. It sees exactly what the logged-in buyer sees. The standard for this is OAuth 2.1, described for agents in the authorisation part of the Model Context Protocol (specification of 28 July 2026) and in the Identity Linking of the Universal Commerce Protocol (version of 25 August 2026). ChatGPT, Claude, Gemini and Copilot support this flow. (Article 10)

Where to start

With the product pages. For every item in your range, there are eight questions to ask:

  1. Is the item in the catalogue, with price, GTIN and delivery time?
  2. Can an agent reach the product page without a security check?
  3. Is the price in the HTML, without JavaScript?
  4. Is the structured data complete, and does it match the page?
  5. Is the item in Google Shopping, with the right price?
  6. Do AI bots retrieve the page, and what do they get?
  7. Do assistants name the item, and at which retailer?
  8. Does the price they name match your price at the time of the answer?

We answer the first five questions from the outside, without access to your systems. For the sixth we need your server logs, for the seventh a measurement panel. The eighth connects both with your shop and your pricing.

The series

  1. It is easy to test which shop an AI assistant recommends
  2. Agents buy through the front door
  3. Every web page gets two readers
  4. Two blocks: visibility and agent purchasing
  5. What an agent finds on your product pages
  6. Who gets what: selective delivery
  7. Price comparison sites as suppliers to AI
  8. UCP, ACP, MCP, AP2: the standards, explained for retailers
  9. The Store Agent
  10. Technical product ranges: the buyer's agent is already here
  11. The Proof of Value in 90 days
  12. Five legal questions before the Agent Channel

Dewline GmbH, Hamburg, builds and runs integration and pricing platforms for retailers with their own platform.

What does an agent see on your product pages?

We look at it together with you, on your own pages.

Book a call