A reliable Google Merchant feed on a catalogue of several thousand SKUs
A B2B distributor was seeing a large share of its catalogue rejected by Google Merchant Center, with no clear visibility into the causes.
The context
The catalogue held several thousand SKUs, fed by multiple suppliers, with descriptions and attributes of uneven quality. A generic feed module already existed, but a significant share of products was rejected by Google Merchant Center: missing identifiers, poorly mapped categories, required fields missing depending on the product type. The rejection rate severely limited the catalogue's visibility on Google Shopping, without it being easy to know exactly which products were causing problems and why.
The goal was to make the feed reliable enough that a much larger share of the catalogue would actually be accepted and distributed.
What I did
-
Auditing existing rejections
Analysing Google Merchant Center feedback to rank rejection causes by frequency and ease of fixing.
-
Mapping Google categories
Matching each internal catalogue category to Google's product taxonomy, handling cases where an internal category had no obvious standard match.
-
Completing required fields
Automatically generating the required identifiers and attributes when missing from product sheets, based on data already available in the catalogue.
-
Pre-submission checks
Adding a feed validation step before submission, to catch an anomaly before it triggered a mass rejection on Google's side.
-
Tracking remaining rejections
Setting up a tracking dashboard listing still-rejected products and their reason, for prioritised fixes rather than random correction.
Frequently asked questions
Why does a B2B catalogue cause more difficulties than a typical B2C catalogue?
Does the feed need to be regenerated every time the catalogue changes?
How do you know how many products are actually being distributed?
Does this work apply to feeds other than Google Merchant?
Describe your need in one minute
A few targeted questions so I can reply with an estimate rather than another questionnaire.