How our checks work — and where their limits are

This page exists so you can check our claims. It describes four checking and measurement methods. The methodology depends on the method: a technical analysis of a website is something different from observing AI answers under test conditions. Each section states which products it applies to. Implementation services such as Merchant Center Repair or Shopify automation have their scope on the respective product page.

Which methodology applies to which product

Pick the product you are interested in. The limits stated in one section do not apply to the others.

BFSG Technical Check (free quick check and €79 report)
Automated technical analysis of public web pages in a browser.
To the section
Product Intelligence (free preview and €49 report)
Read-only check of one product from your connected Shopify store and its public product page.
To the section
Public shop-data measurement (basis of the example report)
Five checks on publicly retrievable shop data. It does not query AI assistants.
To the section
AI Visibility Audit (€699)
Manual observation of AI answers under documented test conditions, plus a review of your product pages.
To the section

Two different kinds of statements

Technical website analysis (BFSG Technical Check, Product Intelligence, public shop-data measurement) reads what a page or shop publishes. Observations of AI answers (AI Visibility Audit only) record what a given AI platform answered to a given prompt at a given time. These are not interchangeable, and one never substitutes for the other.

BFSG Technical Check

Applies to the free quick check and the €79 report. It is an automated technical analysis, not a legal assessment and not a certificate of conformity.

What is checked

Pages: The free quick check covers the homepage. The full report covers up to 10 pages — homepage, category and product pages, plus cart and checkout entry where they can be found automatically.
Desktop and mobile: Every page is loaded at a desktop size (1366×768) and a mobile size (390×844). The mobile run identifies itself as a mobile device.
Automated rule checks: Rules from axe-core for WCAG 2.0 and 2.1 levels A and AA and WCAG 2.2 AA. Only findings the rules confirm are counted as findings.
Partial keyboard checks: Focus visibility, keyboard traps and a skip link, with at most 60 tab stops per page. Anything ambiguous is not counted as a finding but listed as a manual check point.
Purchase path: Cart and checkout entry are included only as far as they can be found automatically. To reach the cart, the scanner may click one generic “add to cart” control on a product page; it then only views the cart and checkout pages. It never submits forms, enters data or presses anything that could place an order. If no clearly safe control exists, the purchase path is reported as not automatically tested.
Possible outcomes: A check can finish completely, partially, be blocked by the website, or fail. The result says which; nothing is filled in for pages that could not be checked.

What it is not

Not every barrier is detectableAutomated tests detect only part of all barriers.
No manual review by specialistsThe list of manual check points is a checklist. You or specialists of your choice carry out that review; it is not part of the product.
No testing with assistive technologyThe check does not use screen readers or other assistive technology.
No legal advice, no certificateThe result is not a legal assessment, makes no statement on warnings or fines, and is not a certificate of conformity.

How the scanner behaves

The quick check only starts after you confirm that you are authorised to have the website checked.
The scanner identifies itself: Mozilla/5.0 (compatible; AIProCraftBFSGScanner/1.0; +https://www.aiprocraft.de)
The mobile run uses: Mozilla/5.0 (Linux; Android 14; Mobile; compatible; AIProCraftBFSGScanner/1.0; +https://www.aiprocraft.de)

Product Intelligence

Applies to the free preview and the €49 report for one product of your connected Shopify store.

For the product you pick it reads, read-only, only the catalogue fields the check needs, and fetches the product's public page as any visitor sees it: title, meta description, canonical URL, heading structure, product description and structured data. Nothing is written back to your store and no order or customer data is retrieved.

What Product Intelligence does not do

It does not query AI assistantsProduct Intelligence does not query ChatGPT, Claude, Perplexity or Gemini and does not evaluate their answers. Observing AI answers is the subject of the separate AI Visibility Audit (see below).
It does not compare against competitorsProduct Intelligence does not check other shops.

Public shop-data measurement

Applies to the example report on this site and to checks that read only publicly retrievable shop data.

The limit of this measurement

This measurement reads only what a shop publishes publicly. It does not observe whether an AI assistant mentions, recommends or knows the shop — that is not part of it. Observing AI answers is done only in the AI Visibility Audit, under the conditions described below.

Concretely: if your robots.txt blocks GPTBot, that is an instruction you published. It does not mean your shop is “invisible to AI assistants”. Those two statements are not the same, and this measurement only makes the first.

What is measured

Five checks, all on publicly retrievable data. No account, no Shopify access, no AI evaluation.

AI crawler access: We read your robots.txt and determine which of the four major AI crawlers are allowed to read the shop. A block is only reported when the entire shop is disallowed — an excluded cart is not a block.; Source: /robots.txt
Structured product data: On product pages we check whether schema.org/Product data is present and complete (price, availability, brand, description).; Source: Product pages (HTML)
Titles and meta descriptions: We check whether the page title and meta description exist and are a sensible length.; Source: Product pages (HTML)
Product count: We determine your product count via /products.json or sitemap.xml. Where only part is determinable we report it as a lower bound rather than asserting an exact figure.; Source: /products.json, /sitemap.xml
llms.txt: We check whether an llms.txt exists. We do not claim it has any effect — no provider publicly commits to reading it.; Source: /llms.txt

Crawlers checked: GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot.

At most 20 product pages are fetched per check. Results for larger catalogues are labelled as an extrapolation.

What this measurement does not cover

These five checks cannot determine the following. They are listed so it is clear what this measurement does not tell you.

Whether an AI assistant mentions your shopThis measurement does not query ChatGPT, Claude, Perplexity or Gemini and does not evaluate their answers. That observation is part of the AI Visibility Audit only.
Rankings or positionsIt does not measure a position in any search result and does not promise one.
Visitors or revenueIt has no access to your analytics and does not extrapolate a revenue effect.
Comparison against competitorsThese five checks do not look at other shops and build no leaderboards. Competitor observations are part of the AI Visibility Audit.
The quality of your copyIt checks whether fields are present and complete — not whether a text is well written.

The possible outcomes

Not every check always produces a result. Rather than guess, we state which state a check is in.

Checked: The check ran and its result is in the report.
Unavailable: The file or page could not be retrieved. We report that as “could not check” and derive no result from it.
Not permitted: Your robots.txt disallows our own crawler. We respect that and stop the check.
Failed: The retrieval failed technically. That is reported too, and never counted as a finding.

How we behave as a crawler

We respect your robots.txt for our own fetches too. If you exclude us, we do not check.
We identify ourselves clearly: AIProCraftScanner/1.0 (+https://www.aiprocraft.de/scan; visibility check on behalf of the site owner)
The number of checks per connection is limited so your shop is not burdened.
Only domain, timestamp and a result summary are stored — no personal data.
If you later connect Shopify, we request read access only (read_products, read_orders, read_all_orders). There is no write access.

AI Visibility Audit

Applies only to the AI Visibility Audit (€699). This is where AI answers are observed — manually, under the conditions below, and documented in a written report.

How the audit is carried out

Manual observation: We query the platforms ourselves and read the answers. The audit is not an automated scan and does not rely on a score from a tool.
Tested platforms: ChatGPT Search, Google AI Overviews / AI Mode, Perplexity and Gemini. Coverage depends on each platform's accessibility and the results it actually returns at the time of delivery. Claude is only included if separately ordered as an add-on.
Prompt selection: The prompts are derived from the objective and the competitors you give us for your store. The prompt set is chosen per audit; we do not state a fixed number of prompts.
Mention, citation, recommendation: These are recorded separately: a mention is your brand or product appearing in an answer; a citation is your store or product pages named as a source where the tool shows sources; a recommendation is your product being actively recommended rather than only mentioned in passing.
Accuracy of what AI says: We note whether the information an AI tool states about your products is correct.
Competitor observations: We note which competitors appear instead of your store in the answers we observed.
Review of your product pages: Separately from the AI answers, we review how well your product pages support AI understanding: attributes, variants, price, availability, shipping, returns, FAQ and structured data.
Written baseline report: The findings are documented as a baseline report in your client portal. The audit includes three monthly re-checks against that baseline at no extra cost; continuing beyond that is optional and quoted separately.

Limits of these observations

An observation is a snapshotAI answers change over time and can differ between runs. An observed answer describes what that platform answered to that prompt when we asked; it does not represent what every user is shown.
No access to private conversationsWe see only the answers we obtain ourselves. We have no access to what other people ask or are told by AI assistants.
Platform access variesA platform may return no result for a store or a prompt. We do not claim every platform returns a result for every store.
No guaranteed outcomeWe do not control or guarantee how third-party AI systems rank, cite or recommend anything, and we do not promise a result after you act on our recommendations.

AI Visibility Audit