← All cases

E-commerce / price comparison

Price comparison in AI assistants: 100 million products in Claude, ChatGPT and Copilot

How billiger.de stays visible where product search is moving, without giving up its referral model: one integration in three assistants, price comparison in chat, purchase in the shop.

Client
solute GmbH / billiger.de
Year
2026
Service
Agentic commerce: scoping, market analysis, development, testing, store submission
3 assistants Claude, ChatGPT, Copilot, one integration
16 % of shops already with a cart prepared in chat
4 weeks from analysis to first store submission

More than 3 million people compare prices on billiger.de every month.

That search increasingly starts not on the website or in Google but in chat with ChatGPT, Claude or Copilot.

For solute GmbH we brought billiger.de there: anyone who asks their AI assistant of choice about a product sees current offers from all retailers in one comparison and reaches the shop via click-through. The integration is live; the first click-outs and purchases via chat have been measured.

For retailers listed on billiger.de that means: their offers appear in the AI assistant without extra effort on their side, and the click goes straight to their shop.

Only for the next step, the cart prepared in chat, the shop itself must be technically ready; so far that applies to 16 %.

Starting point

billiger.de earns from referral: compare, click, buy from the retailer.

As search moves into AI assistants, the portal must be present there, otherwise the click happens without billiger.de even though billiger.de holds the valuable real-time data on valid prices and the best offer.

At the same time, major platforms announced their own checkout features in spring 2026 while standards were still in flux.

solute wanted three answers:

  1. Which standard will hold?
  2. What role does billiger.de play in chat?
  3. How many connected retailers are technically ready for an assistant to prepare a purchase?

Approach

Instead of betting on announcements, we measured. The integration uses an open standard all three assistants speak and shows the price comparison as a unified card in the conversation, with every retailer name as a partner link. We deliberately left checkout with the retailer: paying in chat is not mature in Europe yet. Where shops support it technically, the integration pre-fills the cart for checkout.

Delivery

  • Defined billiger.de’s role in the assistant and platform rollout order together with solute
  • Automated checks on all 1,371 shops in the retailer directory: 224 respond to an AI assistant, 217 of those support the flow through to a pre-filled checkout (as of July 2026)
  • Built a prototype on real shops through to checkout, derived the no-sign-in price comparison variant from it, and tested in Claude, ChatGPT and Microsoft 365 Copilot
  • Production on billiger.de’s domain, privacy notice legally reviewed by solute, submission packages for all three stores prepared; Microsoft store validation passed

Outcome

The same integration runs in Claude and ChatGPT; Microsoft 365 Copilot is next. Today users add it themselves in a few clicks; once it is approved in Anthropic's and OpenAI's directories it will be discoverable there directly. The referral model stays intact: purchase happens in the shop. Retailer checks are repeatable, so solute always sees how many shops are ready for assistant-led purchase and can decide on standards based on measurements rather than announcements.

What we learned

  • Agent-ready retailers are the exception. 16 % of shops, almost exclusively on Shopify. Anyone planning agentic commerce should measure their retailer base first.
  • Start without sign-in, add accounts later. The same integration can do both: price search without an account, personal features with sign-in.
  • Start store submissions early. Company accounts, verification and review times take longer than the tech. Not every platform is open: Gemini has no submission path, Copilot only reaches work accounts.

“Anyone who asks ChatGPT, Claude or Copilot about a product in future will get offers from billiger.de, and we remain the guide that helps customers to the best deal. From the first analysis through to store submission, wonk supported us end to end and implemented the vision early.”

Technology

MCP (Model Context Protocol) for access to the assistants · MCP App for the interactive price card · UCP (Universal Commerce Protocol) for the cart at supporting retailers. Without the card, the language model changes content and layout; in testing, partner links were missing then too.

For standards and platforms see Agentic commerce: where AI-assisted shopping really stands.

Is your catalogue findable in AI assistants, and what is missing before a prepared checkout? 30-minute intro call, no obligation, we will clarify whether and how we can help.