In short: Google Merchant Center, the new Agentic Commerce feed and your own shop search share one common truth: your product data. Structure it cleanly once - titles, attributes, variants, GTINs, availability, images - and you serve all three channels at the same time. Neglect it, and you lose three times over. Here is my checklist from real Shopware projects.
One data problem, three symptoms
In projects I meet the same pattern in three disguises:
- "Merchant Center keeps rejecting our products." Missing GTINs, prices differing between feed and landing page, unclear availability.
- "Our shop search finds nothing." Properties live as prose in the description instead of structured attributes on the product.
- "AI search never recommends us." ChatGPT and friends can only recommend what they can read by machine - since Shopware 6.7.10 there is even a dedicated sales channel with a JSONL feed for exactly this (my article on it).
All three symptoms share the same cause. Which is the good news: you only have to solve the problem once.
The checklist: what clean product data in Shopware looks like
1. Titles with a system
A good product title follows a pattern: brand + product type + decisive attributes (size, color, quantity). Not creative, but consistent - feeds and AI systems parse patterns, not poetry. In Shopware it often pays to build title logic via feed templates instead of maintaining every title by hand.
2. Properties as properties, not prose
Material, dimensions, compatibility, ingredients: everything a customer might filter by or an AI might compare belongs in properties and custom fields - not only in the description. The description explains and sells; the attributes structure. Both have their job.
3. Variants modeled cleanly
Sizes and colors as standalone products instead of variants is one of the most expensive modeling mistakes: it fragments rankings, confuses feeds (see item_group_id in Merchant Center) and makes availability logic unreliable. The rule: if it is the same product in a different version, it is a variant.
4. GTIN, MPN, manufacturer
Unique identifiers are partly mandatory for Merchant Center and the most reliable way for AI systems to match a product unambiguously and link it with external sources (tests, reviews, price comparisons). Missing GTINs are no longer a minor offense in 2026.
5. Prices and availability: consistent all the way into the feed
The classic among Merchant Center rejections: the feed says 49.90 €, the product page says 47.41 € after customer group logic. In Shopware such deviations come from tier prices, customer groups, promotions and cache states. The feed must run through the same price logic as the storefront - a technical topic, not an editorial one.
6. Images that work outside your shop
A main image with a neutral background and no text overlays (Merchant Center requirement), plus real detail and application shots. AI interfaces show your product image in a foreign context - it has to stand on its own.
7. Structured data on the product page
Product schema with price, availability, ratings - consistent with the feed. With the JSON-LD support from 6.7.9 (feature flag) this becomes much cleaner than the old microdata approach. What matters is consistency: if schema, feed and visible page state different prices, your store loses trust with Google and with AI crawlers.
Why this pays off twice now
Until recently, product data maintenance was a Google Shopping topic. Now that AI assistants give buying recommendations and Shopware offers a direct feed interface to platforms like ChatGPT via the Agentic Commerce channel, the same data set is the entry ticket to a second, growing channel. The effort stays the same - the leverage has doubled.
My recommended order: identifiers and variants first (structural base), then prices/availability in the feed (eliminate rejection reasons), then attributes and titles (raise visibility), finally images and descriptions (conversion). That way you do not build a data maintenance desert - you work along the line of measurable benefit.