Is your shop agent-ready? · Article 5 · 2026

What an agent finds on your product pages: 34 shops checked

Whether an AI agent can work with your shop is decided on the product page. That is where price, GTIN and stock are. We wanted to know what an agent finds there today, and on 3 November 2026 we retrieved the product pages of 34 shops: DIY chains, opticians, electronics retailers, and manufacturers and dealers of industrial sensors. The check worked for 30 of them. The results fall into three groups, and the sector makes a large difference.

How we checked

Our checking program identified itself as "DewlineCheck", with a link to our website. It posed neither as an AI provider's crawler nor as a browser. It read the robots.txt, drew 100 product pages at random from the sitemap and retrieved each page once, with a pause of one and a half seconds and without JavaScript. If a shop already refused the robots.txt, or answered the first five product pages with a security check, we stopped. On each page we looked for three things:

  • the price in the visible text,
  • structured product data following schema.org (JSON-LD) with price, GTIN, availability, brand, shipping and returns,
  • whether the price in the data appears in the text exactly as it is.

This is how an agent reads a page when it retrieves it without rendering it in a browser. According to a measurement by Vercel and MERJ, the crawlers of OpenAI and Anthropic do not execute JavaScript. They do not see anything that only the browser adds to the page.

One limitation up front: many shops let the crawlers of large providers through, based on their name or their network addresses. Our checker shows what happens to an agent that is on none of these lists. Examples are a customer's shopping agent, a procurement agent in B2B or the tool of a new provider.

The result

reading path in place reachable, empty without JavaScript no access for the checker DIY and garden 3 shops 2 1 Optical 5 shops 4 1 Electronics and mail order 9 shops 1 8 Industrial sensors 13 manufacturers and dealers 1 2 10 Own measurement, 3 November 2026: 30 shops, up to 100 product pages each from the sitemap, without JavaScript, checker with its own identifier.

Group 1: no access, 19 shops. 19 of 30 shops gave our checker no product page:

  • Electronics and mail order: Coolblue, Cyberport, Euronics, Galaxus, MediaMarkt, notebooksbilliger, Otto, Saturn.
  • Industrial sensors: Automation24, Balluff, Baumer, Conrad, Contrinex, Festo, Leuze, Pepperl+Fuchs, RS, SICK.
  • Optical: Brille24.

Most of them refused the robots.txt already, with status code 403 or the security check of a content delivery network ("Just a moment …"). Balluff, RS and Brille24 delivered the robots.txt and then answered the first five product pages with a security check. Otto answered every product page with error 400. SICK delivered an empty shell of four kilobytes. Contrinex is the only sensor manufacturer that publishes a profile under the Universal Commerce Protocol, the very description agents are meant to use to find its catalogue. After 23 pages, Contrinex gave our checker no more.

Group 2: reachable, empty without JavaScript, 4 shops. Here the agent gets a page, but no product. toom delivers the price only in a meta tag meant for social networks; description, GTIN and availability are built by the browser. expert delivers two megabytes per page and no price in the text. Keyence delivers structured product data, but without price and without GTIN. At wenglor we found neither a price nor product data.

Group 3: reading path in place, 7 shops. At Hornbach, the price was in the text on 99 of 100 pages. All 100 pages had structured product data with GTIN, 86 of them with availability. On 97 of 99 pages the price from the data appeared in the text exactly as it is. Next to Hornbach are the opticians Fielmann, Mister Spex, eyes + more and Apollo, the DIY chain hagebau and the sensor manufacturer ifm. Even in this group, something is missing almost everywhere:

  • Only Mister Spex and hagebau had shipping costs and returns in their structured data.
  • A GTIN was missing at Fielmann, eyes + more and ifm.
  • At hagebau and ifm, the price was only in the data and not in the text.

Mister Spex was the most complete: price in the text, GTIN on 94 of 100 pages, availability, brand, shipping and returns on all of them.

Who is asking

Our checker identifies itself as an unknown program. A program that presents itself as the crawler of a large AI provider may get further at some shops. If a shop checks only the name and not the address the request comes from, a program that merely calls itself GPTBot gets through as well. The operators publish their address lists so that shops can check exactly that.

A statement about agent purchasing therefore needs two pieces of information: who was asking, and which pages. A homepage says little about the product pages where the agent buys. Hornbach shows this: its robots.txt closes cart, customer account and similar areas to all visitors, but no AI crawler, and the product pages are fully readable.

What retailers can do with this

The order follows from the three groups.

  1. Access. Shops in group 1 first need a rule on which agents see which pages. Verified bots can be recognised by their operators' address lists and signatures and let through with quotas. A blanket block also keeps out your customers' agents. (Article 6)
  2. Readable page. Price, availability and the core data belong in the HTML the server delivers. Where JavaScript builds the page today, a pre-rendered version helps. It must show the same facts as the page in the browser. Google treats a version with different content for bots as cloaking.
  3. Complete data. Structured product data with GTIN, price, availability, shipping and returns. Shipping and returns were missing at 5 of 7 shops with a reading path. Google requires shipping information in Merchant Center for Germany, and the checkout under the Universal Commerce Protocol requires a returns policy.
  4. Check that prices match. The price in the data, in the text, in the feed and at checkout must match, and keep matching. At Hornbach it matched on 97 of 99 pages. If it differs, an assistant may name a different price from the one the checkout charges.

What this check does not show

  • It is a sample of 100 pages on one day. In large ranges, another part may look different.
  • We did not render the pages in a browser. Agents that operate a browser see more. Article 1 describes what that looks like.
  • Whether the crawlers of large providers get through only shows in the shop's server logs.
  • We did not check Google Shopping or mentions in answers here.

Sources

  • Own measurement, 3 November 2026: 34 shops, 30 of them checkable, up to 100 product pages each from the sitemap, without JavaScript, checking program with its own identifier, one request per page, 1.5 seconds pause. OBI and Lensbest (sitemap not reachable), Edel-Optics (no product pages in the sitemap) and Turck (too many requests) could not be checked. Raw data per shop is available on request.
  • Address lists of the crawler operators, for example OpenAI (openai.com/searchbot.json) and Google (developers.google.com/crawling/ipranges), retrieved 3 November 2026.
  • Vercel and MERJ, "The rise of the AI crawler", December 2024: the crawlers of OpenAI and Anthropic do not execute JavaScript.
  • Google Search Central, guidelines on cloaking; schema.org, Product and Offer with shippingDetails and hasMerchantReturnPolicy.
  • Requirements of Google Merchant Center, Microsoft Merchant Center, OpenAI and the Universal Commerce Protocol for product data, evaluated in our research of 1 November 2026.

What does an agent see on your product pages?

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

Book a call