Perfect Product Data for AI: How Deep Structured Data Puts Your Brand in the Answer

AI assistants like ChatGPT, Perplexity, Gemini, Claude and Copilot do not read your page the way a human sees it. They read the text and a machine-readable layer beneath it: structured data in schema.org format, usually as JSON-LD.

The basic schemas for that (FAQPage, a Product with name and price, Organization) are mandatory but not an advantage. Every developer will soon do them, they are the floor. The real leverage sits above: data that answers the actual question behind a prompt.

How people really ask

Nobody searches for "espresso machine 25 cm wide". They search a situation:

"I need a portafilter for a small kitchen that also makes cappuccino. The water tank has to go in from the side, there is a cabinet above the machine."

The assistant breaks that into follow-ups: dual boiler or single? How wide and tall? Can the tank be filled from the side? Milk frothing? Budget? Every one of those follow-ups is a field in your product data, or a gap that drops you out of the answer.

And queries are getting more specific, not necessarily longer. Whether as a full sentence or keyword-like, the decisive constraints, the small kitchen, the side-loading tank, the budget, are usually in the first prompt already. Each one is a data field you either answer or you don’t.

The minimum versus the depth

This is the Product minimum most shops ship. Formally correct, but it answers none of the questions above:

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Roma Dual Espressomaschine",
  "brand": { "@type": "Brand", "name": "Naturbohne" },
  "offers": {
    "@type": "Offer",
    "priceCurrency": "EUR",
    "price": "899.00",
    "availability": "https://schema.org/InStock"
  }
}

And this is the same product when the data actually answers the questions. The carrier is additionalProperty for free-form specs plus width/height/depth as QuantitativeValue:

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Naturbohne Roma Dual, Siebträger-Espressomaschine",
  "brand": { "@type": "Brand", "name": "Naturbohne" },
  "category": "Siebträger-Espressomaschine",
  "description": "Dual-Boiler-Siebträger für Espresso und Cappuccino, kompakt für kleine Küchen.",
  "image": "https://example.com/roma-dual.jpg",
  "width":  { "@type": "QuantitativeValue", "value": 25, "unitCode": "CMT" },
  "height": { "@type": "QuantitativeValue", "value": 38, "unitCode": "CMT" },
  "depth":  { "@type": "QuantitativeValue", "value": 42, "unitCode": "CMT" },
  "weight": { "@type": "QuantitativeValue", "value": 14, "unitCode": "KGM" },
  "additionalProperty": [
    { "@type": "PropertyValue", "name": "Boiler-System", "value": "Dual-Boiler, getrennt für Brühen und Dampf" },
    { "@type": "PropertyValue", "name": "Wassertank", "value": "2,5 l, seitlich entnehmbar, Befüllung von oben nicht möglich" },
    { "@type": "PropertyValue", "name": "Milchgetränke", "value": "Ja, professionelle Dampflanze für Cappuccino und Latte" },
    { "@type": "PropertyValue", "name": "Aufheizzeit", "value": "ca. 6 Minuten" },
    { "@type": "PropertyValue", "name": "Entkalkungsintervall", "value": "alle 2 Monate bei mittlerer Wasserhärte" },
    { "@type": "PropertyValue", "name": "Ideal für", "value": "kleine Küchen und Einsteiger mit Espresso-Anspruch" }
  ],
  "offers": {
    "@type": "Offer",
    "priceCurrency": "EUR",
    "price": "899.00",
    "availability": "https://schema.org/InStock",
    "url": "https://example.com/roma-dual"
  }
}

Every line in the additionalProperty block answers a concrete prompt: the water-tank line answers "can I fill it from the side", the milk line "does it make cappuccino", the descaling interval the maintenance question. That is the difference between findable and recommended.

Why depth wins the upper funnel

The point is not just completeness, it is who owns the answer. When a user asks about a niche they do not yet tie to a brand, the AI builds the shortlist from whatever structured facts it can find. Supply the deep data and your brand shows up in the upper funnel, in front of people who did not know you before.

And beware the wrong metric: being cited as a source is not the same as being recommended. Monitoring tools count how often your brand shows up as a source on broad category prompts. But the user decides nothing there, they are still on the journey. The recommendation only lands when the AI narrows to concrete constraints, and there whoever answers the constraints in their data wins, not whoever collected the most source mentions.

Do not supply it and the retailer will. The big electronics retailer maintains exactly these specs, plus price, and then stands in front of you in the answer. It owns the comparison, and usually whoever got there first wins.

The retailer tells price. The brand tells heritage and story. But both have to be machine-readable, or the AI only hears the retailer.