Product data for AI shopping assistants
By Philipp Kant 3 min read
Product data for AI shopping assistants is the structured information about each product (identifiers, attributes, price, availability) that assistants like ChatGPT, Google’s AI Mode, and Perplexity read when they answer a shopping question. An assistant recommends what it can verify against the buyer’s constraints, so incomplete data means invisible products.
This turns a PIM question into a revenue question.
Assistants recommend what they can verify
A buyer no longer types “hiking boots”. They ask: “waterproof hiking boots for wide feet under 150 euros, available this week.” Every part of that sentence is a filter:
- waterproof needs a material or feature attribute,
- wide feet needs a fit or width attribute,
- under 150 euros needs a current price,
- this week needs availability.
A product whose data does not state an attribute cannot pass that filter, however good the product is. The assistant does not guess in your favor. It picks the competitor whose data answers the question.
The attributes that decide a recommendation
| Attribute | Why it decides | Where it has to live |
|---|---|---|
| GTIN / EAN, brand, MPN | Identifies the product across sources | Feed, schema markup |
| Price and currency | Every budget question filters on it | Feed, schema Offer, page |
| Availability | Assistants avoid recommending what cannot ship | Feed, schema Offer |
| Material, size, fit, color | The constraints buyers actually type | Feed attributes, page text |
| Specific use cases | “for trail running”, “for small kitchens” | Page text, bullets |
| Reviews and ratings | Tie-breaker between similar products | Schema AggregateRating, page |
| Shipping and returns | Often part of the question | Feed, page |
The common gap is not the price. It is the attributes in the middle of that table: material, dimensions, fit, compatibility. They live in somebody’s head, a supplier PDF, or a free-text description.
Feed, schema markup, and page have to agree
Assistants read product data from more than one place: merchant feeds,
schema.org Product markup on the page, and the page text itself. When
the three disagree (a different price in the feed, a size only in the
description, an outdated availability), the product looks unreliable.
One source of truth fixes that: the PIM holds the attribute once, and feed, markup, and page are generated from it. That is the core of what MDM/PIM actually solves, and it matters more now that machines compare products, not only people.
How to get a catalog ready
- Measure completeness. Run the catalog export through our Product Data Check to see which attributes are missing, inconsistent, or stuck in free text.
- Fill the gaps at the source. In the PIM, not in the shop template. If the data model has no field for fit or material, add one. Our PIM migration checklist covers the data layer.
- Generate channel copy from the attributes. The Product Copy Generator shows what attribute-driven copy looks like and which gaps hold it back.
- Publish structured markup. Product, Offer, and rating markup on every product page; the Schema Markup Check shows what a page carries today.
- Let the assistants in. A robots.txt that blocks AI search crawlers hides the whole catalog. The AI Crawler Check shows who gets through, and 26% of top sites block AI crawlers shows how common the mistake is.
FAQ
How do AI shopping assistants choose which products to recommend? They match the constraints in the question (use, size, material, price, availability) against structured product data and page content, and recommend the products whose data verifiably meets them.
What product data do I need for AI shopping answers? Identifiers (GTIN, brand, MPN), current price and availability, and the attributes buyers filter on: material, size, fit, color, compatibility, use case. Plus reviews and shipping terms where they exist.
Is product schema markup enough? No. Markup, feeds, and page text are read together. They need the same values, which is easiest when all three are generated from one PIM.
Do I need a PIM for this? For a few dozen products, a disciplined spreadsheet works. Past a few hundred products or more than two channels, a PIM is what keeps the attributes complete and consistent.
If your catalog’s attributes live in PDFs and free text, that is the data layer we build.
