Use case · Shopping feed managers

Google product categories for feed managers

If you own a Shopping feed, the category column is a column you are answerable for. The workflow here fills it for the whole catalog in one pass, sends only the rows the classifier is confident about straight to Merchant Center, and puts the uncertain ones in a separate file with the runner-up beside each — so review time goes where it changes the outcome.

What this comes down to

  • Every row gets a Google category ID, the full path and a confidence score.
  • Only rows scoring 0.90 or higher go into the upload-ready supplemental feed.
  • Rows below 0.85 go to a review file with the runner-up category beside them.
  • Identical products are charged once, so variants do not multiply the cost.

SellerSort Team · Updated

The workflow in four steps

  1. ExportExport the catalog as CSV, TSV or XLSX — up to 50,000 rows and 20 MB.
  2. Map and previewConfirm the column mapping and read the free 10-row preview.
  3. RunRun the full file; the credit count is shown before you start.
  4. Ship and reviewUpload supplemental-feed.csv; work through needs-review.csv.

The job the column does

google_product_category is optional — Google assigns a category to every product — but the value you send is what Google accepts for category-specific attribute rules and Google Ads targeting, and the same column feeds Meta catalogs and Microsoft Shopping. A wrong or invalid value is rejected and may keep a product from appearing. So the column is worth getting right, and the catalog is too big to do it by hand.

The workflow

Export the catalog you already have — from the store, from Merchant Center, from your feed tool. The Bulk Mapper guesses which columns are title, description, brand, product type and ID, and you confirm them. It classifies the first ten rows free on the same engine and thresholds as the full run, so you can see whether the mapping is right before spending anything.

The run returns three files. full.csv is every input row with the category, ID, path and confidence appended. supplemental-feed.csv holds only the rows at confidence 0.90 or above, with the two columns Merchant Center needs. needs-review.csv holds the rows below 0.85, each with the runner-up — that is the list worth a human’s attention.

Rows between 0.85 and 0.90 are deliberately in neither the feed nor the review file: probably right, but not confident enough to upload unseen. Filter full.csv on the confidence column to find them.

Shipping it

Upload supplemental-feed.csv as a supplemental data source, linked to your primary source. It matches on id, so it only works if your export carried the IDs Merchant Center uses — check that before the run, not after. The full procedure, including the add-on that reveals the Supplemental sources tab, is in override Google’s automatic category.

When Google publishes a new taxonomy version, re-run only the rows whose category changed rather than the whole file.

What it costs

One credit classifies one distinct row. Rows with the same title, description, brand and type are charged once, rows without a title are skipped and not charged, and a row that cannot be classified is refunded. The exact count appears after the free preview, before you spend. Credit packs start at $9 for 2,000 rows and nothing renews or expires — see pricing.

Making it a routine, not a project

Categories drift as a catalog grows: new products arrive without one, and an import rule writes the wrong value into a batch. Re-running the catalog in verify mode every so often turns that into a short list of rows that disagree with the classifier, instead of a clean-up project once a year.

Questions

Will this fix disapproved products?

A missing category does not cause disapprovals, so filling it is not a disapproval fix. What it prevents is invalid category values, which are rejected, and it gives Google your category for the cases where Google accepts an override.

Can I keep my existing categories and only check them?

Yes. The Bulk Mapper has a mode that verifies the categories already in your file rather than assigning new ones, so you can find the rows that disagree.