One direction: the platform publishes a file, Google fetches it
Google Merchant Center is the account where Google keeps a seller’s product data and from which it feeds free product listings and product ads. Endora Commerce generates a file and serves it at a URL that Merchant Center polls as a scheduled fetch. Nothing flows the other way: product approval statuses and performance reports are checked in Merchant Center.
A feed always covers one sales channel and one language. The channel determines the catalog, the prices and the links to product pages; the language decides which translations of names, descriptions and category paths are written to the file.
What is in the file
The Google Merchant Center template is XML in RSS 2.0 format with the g: namespace, generated at variant level: every SKU variant is a separate item, which is what Google expects. The field list covers Google’s full required set.
The system template cannot be edited, but it can be duplicated and the copy changed. In the editor, each output field is bound to a source picked from a list: a product field, an attribute, a custom field, an image, a price, a category path. This helps with identifiers: Google conditionally requires a brand and either a GTIN or an MPN, and in technical catalogs the manufacturer part number often lives in an attribute or a custom field. A preview shows the value of each field for a sample product before anything is published.
Prices and visibility: Google’s requirements versus the B2B model
Google’s product data specification requires that:
- the price in the feed matches the price on the landing page and at checkout,
- outside the US and Canada, the price includes VAT,
- any customer can buy the product at the submitted price, without signing up for a membership program.
Three mechanisms on the platform side correspond to this. First, a feed can be net or gross, with the tax country set on the feed; an item for which a gross feed finds no tax rule is reported in the run report. Second, the price comes from a named price list or from the channel’s anonymous price resolution, meaning the price a visitor who is not logged in sees. Third, only products that are active, not archived, members of the channel and visible to an anonymous visitor reach the file. A product reserved for named organisations stays out of the feed. Within those limits, you narrow the feed with the same rule builder used for promotions, and the panel shows a live count of the items the rule matches.
If prices in your shop are visible only after login, or every account has its own terms, the sensible candidate for Merchant Center is a channel with public prices, not the entire wholesale catalog.
Going live
- Create a feed in the admin panel: the Google Merchant Center template, a sales channel, a language and a name. The remaining settings have working defaults.
- Run the first generation. The feed URL appears only after the first successful run, so that nobody can copy a link that answers with a 404.
- In Merchant Center, add a product data source from a file and paste the URL as a scheduled fetch.
- Set the generation schedule on the feed (hourly, every 4 hours, daily, or a custom cron expression with a time zone) so that the file is ready before the fetch time set in Google.
On Google’s side, the file must be smaller than 4 GB, and the Googlebot and AdsBot-Google crawlers must not be blocked.
The URL contains a random token, and anyone who knows it can read a file that contains your prices, so treat it as a credential. Rotating it invalidates the old URL immediately: Merchant Center stops receiving updates until you paste the new one.
Google categories
The platform ships with Google’s product taxonomy (5,595 categories, in Polish and English), so generating a feed never depends on reaching Google’s servers. You map a shop category to a Google node and all its subcategories inherit the mapping unless they have one of their own. A missing mapping is not an error: the field is omitted, the item is still written to the file, and the run records a warning.
What you see after each run
Every generation records the number of items considered, emitted and skipped, plus warnings. Problems are grouped by reason, with the SKU of each item and a CSV export of the full list. Typical reasons are:
- no price in the feed’s currency and price list,
- no publicly reachable image,
- an empty field that Google requires, with no fallback value,
- no translation in the feed’s language (a fallback language is used),
- no Google category mapping.
A good file is never replaced by a worse one. If more than half of the considered items are skipped (the default threshold, adjustable in settings), or the rule matches nothing, the previously published file keeps being served at the URL. A failed generation notifies an administrator once, on the transition into failure.
Large catalogs
Serialization streams rather than loading the whole catalog into memory: in a test on 100,000 products it produces a 38 MB XML file with peak memory growth under 20 MB.
The same engine also generates the Meta product catalog: the assortment rule and the price list work the same way, only the output format and the taxonomy differ.