Ten company websites: what information we found for AI enquiries

What we actually checked

On 1 November 2026, we examined ten public company websites from the DACH region: four from Switzerland, three from Germany and three from Austria. The targeted selection includes bike workshops, restaurants, woodworking businesses, hotels, driving schools, and tailoring and shoe repair businesses, among others. It is not a representative market sample. We did not check company sizes against a formal SME definition.

For each website, we defined specific offer or contact pages and looked at seven questions separately: Who is the provider? What exactly do they offer? How do you find out the price? What conditions are stated? How do you get in touch? What next appointment or registration step is visible? What structured data is in the HTML?

Nine websites could be evaluated for content. One responded with HTTP 403 to the documented request; its content remains unknown. Three candidates that were initially not technically evaluable were replaced once and logged as exclusions. Access restrictions were not circumvented. We did not send any enquiry or book any appointment.

Sources: Veloplus – checked offer page · Hiltl – checked offer page · Gerold Kälin AG – checked offer page · Hotel Luise – checked offer page · Fahrschule Schlegel – checked offer page · fluidbike – checked offer page · Fahrschulen RAINER – checked offer page · Änderungsschneiderei Anna Magzanov – checked offer page · Schuhservice Yildiz – checked offer page

A price and a price path are different findings

On four of the nine evaluable websites, we found concrete figures for described services. On two others, the path to an individual quote was explicitly explained. On three websites, we only saw references to price lists, menus or a booking process whose target content we did not check in this limited pilot. This explicitly does not mean that there are no prices there.

For an automated comparison, this distinction is crucial: a starting price is not a guaranteed final price. A linked PDF is not yet an extracted offer. And getting in touch can be the right next step if material, effort or condition must be assessed first.

Sources: Veloplus – checked offer page · fluidbike – checked offer page · Fahrschulen RAINER – checked offer page · Änderungsschneiderei Anna Magzanov – checked offer page · Gerold Kälin AG – checked offer page · Schuhservice Yildiz – checked offer page · Hiltl – checked offer page · Hotel Luise – checked offer page · Fahrschule Schlegel – checked offer page

An appointments heading does not yet confirm a booking

At fluidbike, service packages and additional material costs were described in concrete terms. Under the appointments heading, however, it explicitly said “coming soon”. In the supplementary browser run, we could therefore see a price but not confirm an online appointment action that was already available. Phone and email are a different contact route.

The joinery Gerold Kälin, by contrast, describes the path to a non-binding quote and shows direct contact details. That too is a clear next step. A business does not have to offer a calendar if an individual project needs to be clarified first.

On other pages, we found reservation widgets, external booking links or a registration. We did not go through these as far as availability or completion. Both browser extractions were carried out by the same Codex agent; they are not independent comparative tests of several AI models.

Sources: fluidbike – checked offer page · Gerold Kälin AG – checked offer page · Hiltl – checked offer page · Hotel Luise – checked offer page · Fahrschulen RAINER – checked offer page

Structured data does not replace a clear description of services

On five of the nine evaluable websites, the selected HTML documents contained JSON-LD or microdata types. On four, we found no such types there. This says nothing about all subpages and is not a check of whether the existing data is complete, correct or suitable for special search results.

The types found range from general website information to offers and price details. The mere presence of a website schema therefore does not mean an agent can correctly compare scope and conditions. Likewise, a missing type in a document should not automatically be rated as a bad website.

Sources: Veloplus – checked offer page · Hiltl – checked offer page · Gerold Kälin AG – checked offer page · Hotel Luise – checked offer page · Fahrschule Schlegel – checked offer page · fluidbike – checked offer page · Fahrschulen RAINER – checked offer page · Änderungsschneiderei Anna Magzanov – checked offer page · Schuhservice Yildiz – checked offer page

What you can check in your own offer

The examples do not produce a ranking, but they do raise concrete questions for your own website. Start with an actual service and follow the path from first understanding to the right enquiry.

• Are the provider, location and responsible contact shown together with the offer?
• Are the result, scope and important exclusions described concretely?
• Is an amount given, including unit and surcharges, or is the path to a quote explained?
• What information, documents or items do you need before confirming?
• Is the next step actually available, and is it labelled correctly?
• Is structured information maintained from the same current facts as the visible text?

What this pilot does not prove

Deterministic HTML fetching measures neither the capabilities of all AI agents nor visibility in ChatGPT, Claude or Google. The two browser tasks also allow no statement about success rates in the market. Further tests must document different agents, concrete customer tasks and identical success criteria separately.

Our finding is narrower: for real offers, price information, quote path, contact and an executable next step cannot be equated. Describing these differences cleanly is a sensible start for business-to-agent. Pribbly can help prepare offers and enquiry paths so they are easy to understand; this does not result in a guaranteed AI recommendation or top position.