Methodology version 0.1

So messen wir

The check assesses what an AI agent sees when it visits a shop or website without logging in. It reads the HTML as the server delivers it, the structured data, robots.txt, sitemaps and the known agent endpoints. It never logs in, never submits forms, never adds anything to the basket and never changes anything. The result is a readiness indicator based on public signals, not a certification.

Findability30 %

Are AI agents allowed to find your pages and products – and can they?

Readability45 %

Can agents understand and compare your products and content?

Actionability25 %

Can agents trust you and act on behalf of their users – buy, enquire, get in touch?

How the score is calculated

  1. Each check gets a value between 0 (not met) and 1 (met), with partial points in between.
  2. Checks that cannot apply to a site are left out. Informational checks never count.
  3. Each of the three levels is the weighted average of its checks (0 to 100).
  4. The overall score is the weighted average of the three levels.
  5. Certain gaps cap the overall score, however good the rest is.
  6. If the scan is blocked, there is no overall score, but the result “limited assessment”.

Shop or website

The check detects by itself whether it is looking at an online shop or a website without a shop, such as a B2B website presenting products and services. Some checks differ depending on the mode. A score of 70 for a website therefore doesn’t mean the same as for a shop.

Findability

30 % of the score

Are AI agents allowed to find your pages and products – and can they?

Access for AI agents

Check ShopWebsite Full points Partial points Effort
AI search crawlers allowed 25 % 32 % All search and retrieval crawlers of AI services (e.g. OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-SearchBot) may read the home page, category and product pages. Proportional to the crawlers allowed. 0 points if the four most important are all blocked. low
robots.txt available 4 % 5 % A valid text file, not an HTML page, is served at /robots.txt. – low
Usage rules for AI 4 % 5 % The robots.txt contains at least one Content-Signal line stating what content may be used for. – low
Secure connection 8 % 11 % The site loads over HTTPS, and HTTP redirects to HTTPS. The site loads over HTTPS, but HTTP doesn’t redirect. medium
Catalogue publicly visible 12 % – Products and prices are visible without logging in. Products are visible, prices only after logging in. This is often a deliberate B2B decision; the report puts it in context. high

Finding the content

Check ShopWebsite Full points Partial points Effort
Sitemap reachable 8 % 11 % A readable XML sitemap can be reached via robots.txt or the standard path. – low
Products in the sitemap 8 % – A dedicated sitemap for products. Product URLs only appear in a general sitemap. low
Product pages findable 21 % – At least 3 product pages found via links or the sitemap. 1 to 2 product pages found. medium
Offering pages findable – 26 % At least 3 pages about products or services found via links or the sitemap. 1 to 2 such pages found. medium
Language and country versions marked up 4 % 5 % Multilingual pages reference each other correctly and reciprocally via hreflang. hreflang present, but incomplete or without return links. Not applicable to single-language sites. low
References in the HTTP header 4 % 5 % The home page’s HTTP header points to an API, feeds or documentation (Link header as per RFC 8288). – low

Readability

45 % of the score

Can agents understand and compare your products and content?

Product data

Check ShopWebsite Full points Partial points Effort
Product data per schema.org 17 % – Product or ProductGroup as structured data on at least 80 % of the product pages checked. Proportional, with a deduction. medium
Machine-readable price and availability 14 % – Price, currency and availability in the product data of all product pages checked. Price and currency alone give 60 %. Not applicable if prices are deliberately behind a login. medium
Unique product identifiers 9 % – GTIN/EAN or manufacturer part number (MPN). A shop-specific item number (SKU) alone gives half. medium
Brand and company 6 % – Brand in the product and an Organization entry for the website. One of the two. low
Reviews 6 % – Reviews (aggregateRating or review) in the product data. Proportional to product pages with reviews. medium
JSON-LD format 3 % – Complete product and offer data as JSON-LD, the format agents prefer. Proportional; Microdata is recognised but doesn’t count here. medium

Content and structure

Check ShopWebsite Full points Partial points Effort
Machine-readable company – 16 % Organization with name, logo, address and contact as structured data. Organization present, but incomplete. low
Structured products and services – 12 % Offering pages describe products or services as Product or Service per schema.org. Proportional to pages. medium
Product content sufficient for a recommendation 11 % – Title, description of 150 characters or more, product image and price or availability. Proportional to product pages and elements. medium
Offering content sufficient for a recommendation – 20 % Offering pages with title, description of 150 characters or more, image and technical features or a data sheet. Proportional to pages and elements. medium
Content readable without JavaScript 11 % 16 % Product data, prices and main text are already in the HTML the server delivers. Parts only appear after JavaScript runs. Most AI agents then can’t see them. high
Clean canonicals 6 % 8 % All pages checked state their own canonical URL without tracking parameters. Proportional to pages. low
Breadcrumb navigation 6 % 8 % BreadcrumbList as structured data. Breadcrumbs only as HTML. low
Questions and answers 3 % 8 % FAQ content with FAQPage markup. FAQ page without markup. medium

Guides for agents

Check ShopWebsite Full points Partial points Effort
llms.txt 6 % 8 % A structured Markdown file at /llms.txt explains the site to language models. File present, but without structure. low
AGENTS.md 3 % 4 % A guide for agents is available at /AGENTS.md. Only at /agents.md in lower case. low

Actionability

25 % of the score

Can agents trust you and act on behalf of their users – buy, enquire, get in touch?

Trust and legal

Check ShopWebsite Full points Partial points Effort
Clearly identifiable merchant 12 % 19 % Company name plus address or tax/registration number in the Organization data (per country, e.g. USt-IdNr., HRB, NIP, KRS, SIREN, SIRET). The same details only as text in the legal notice or the “Mentions légales”. low
Contact channels 8 % 19 % Phone or email as structured data or as a tel:/mailto: link. Only a contact form. low
Shipping and returns 12 % – Half each – shipping and returns as structured data (OfferShippingDetails, MerchantReturnPolicy). A text page with at least 500 characters gives half of each half. medium
Legal pages 8 % 12 % Privacy policy and terms (shop) or privacy policy and legal notice (website) reachable and filled with content. One of the two. low

Agentic Commerce

Check ShopWebsite Full points Partial points Effort
Universal Commerce Protocol (UCP) 17 % – Versioned UCP profile at /.well-known/ucp with services, coverage of catalogue, basket, checkout and order, plus payment methods. Graded by how complete the profile is. high
Agentic Commerce Protocol (ACP) 12 % – Signs of an ACP checkout (OpenAI/Stripe) can be found publicly. – high
Product feed findable 8 % – A product feed (e.g. for Google Merchant or ChatGPT) is publicly linked or announced. – medium

Enquiries and contact

Check ShopWebsite Full points Partial points Effort
Enquiries possible – 25 % Offering pages lead clearly to an enquiry or quote (form, enquiry button, contact person). Only a general contact page. medium

Interfaces for agents

Check ShopWebsite Full points Partial points Effort
MCP server 8 % 6 % A publicly announced MCP server through which agents can use the catalogue or functions. – high
API catalogue 4 % 6 % Linkset at /.well-known/api-catalog (RFC 9727). – medium
OAuth/OIDC discovery 4 % 6 % Discovery document with issuer, through which agents can sign in on behalf of users. – medium
A2A Agent Card 4 % 6 % Agent Card with name and skills. File without skills. medium

What limits the score

  • Agents find no product pages, neither via links nor via the sitemap.29
  • Agents find no pages about products or services.29
  • The robots.txt blocks the most important AI search crawlers.49
  • No product page has machine-readable product data.49
  • Price and availability are not machine-readable anywhere.69

Levels

  • Pioneerfrom 90
  • Agent-readyfrom 70
  • Readablefrom 50
  • Room to growfrom 30
  • Invisible to agentsfrom 0

Context (not weighted)

Shop or website

Detected and shown; determines which checks apply.

Platform

Shopware, Salesforce Commerce Cloud, Shopify, Magento and others are detected and shown.

Bot protection

Detected firewalls (e.g. Cloudflare, Akamai, DataDome) and whether they let our scanner through.

Training crawlers

Whether crawlers for AI training (e.g. GPTBot, ClaudeBot, Google-Extended) are blocked. This is a legitimate decision and is never scored.

AI Catalog

Reported while the standard is still young.

When the scan can’t get through

Limited assessment: we show what we could still check, but no overall score.
  • A firewall or CDN turned our scanner away. Whether this also affects AI agents depends on its configuration.
  • The robots.txt blocks our scanner. We respect that.
  • The robots.txt couldn’t be retrieved. We don’t scan without known rules. Another attempt usually works.
  • The robots.txt asks for a very long pause between two requests (crawl-delay above 5 seconds). A check that respects it would take many minutes. We respect it and therefore do not run a full check.
Unreachable: the site couldn’t be loaded the way an agent would load it.
  • The home page doesn’t respond, or doesn’t respond with a web page.

Limits of the method

  1. Closed B2B catalogues are barely visible to agents – or to us. The score then describes public visibility, not the quality of the business.
  2. Content that only appears via JavaScript doesn’t count. We do show it separately, though, so you can see what agents miss.
  3. Large catalogues are assessed using a sample of up to 6 product pages.
  4. Agentic Commerce is young. Checks are added and reweighted; scans months apart are not directly comparable.